diff --git a/common/ledger/util/badgerdbhelper/badgerdb_helper.go b/common/ledger/util/badgerdbhelper/badgerdb_helper.go new file mode 100644 index 00000000000..ef16783d93d --- /dev/null +++ b/common/ledger/util/badgerdbhelper/badgerdb_helper.go @@ -0,0 +1,268 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package badgerdbhelper + +import ( + "fmt" + "os" + "path/filepath" + "sync" + + badger "github.com/dgraph-io/badger/v4" + "github.com/hyperledger/fabric-lib-go/common/flogging" + "github.com/hyperledger/fabric/internal/fileutil" + "github.com/pkg/errors" +) + +var logger = flogging.MustGetLogger("badgerdbhelper") + +type dbState int32 + +const ( + // ManifestFilename is the filename for the manifest file. + ManifestFilename = "MANIFEST" + + closed dbState = iota + opened +) + +// DB - a wrapper on an actual store +type DB struct { + conf *Conf + db *badger.DB + dbState dbState + mutex sync.RWMutex +} + +// CreateDB constructs a `DB` +func CreateDB(conf *Conf) *DB { + return &DB{ + conf: conf, + dbState: closed, + } +} + +// Open opens the underlying db +func (dbInst *DB) Open() { + dbInst.mutex.Lock() + defer dbInst.mutex.Unlock() + if dbInst.dbState == opened { + return + } + dbPath := dbInst.conf.DBPath + var err error + var dirEmpty bool + if dirEmpty, err = fileutil.CreateDirIfMissing(dbPath); err != nil { + panic(fmt.Sprintf("Error creating dir if missing: %s", err)) + } + if !dirEmpty { + if _, err := os.Stat(filepath.Join(dbPath, ManifestFilename)); err != nil { + panic(fmt.Sprintf("Error opening badgerdb: %s", err)) + } + } + if dbInst.db, err = badger.Open(badger.DefaultOptions(dbPath)); err != nil { + panic(fmt.Sprintf("Error opening badgerdb: %s", err)) + } + dbInst.dbState = opened +} + +// IsEmpty returns whether or not a database is empty +func (dbInst *DB) IsEmpty() (bool, error) { + dbInst.mutex.RLock() + defer dbInst.mutex.RUnlock() + var hasItems bool + opts := badger.DefaultIteratorOptions + opts.PrefetchValues = false + err := dbInst.db.View(func(txn *badger.Txn) error { + it := txn.NewIterator(opts) + defer it.Close() + it.Rewind() + hasItems = it.Valid() + return nil + }) + return !hasItems, + errors.Wrapf(err, "error while trying to see if the badgerdb at path [%s] is empty", dbInst.conf.DBPath) +} + +// Close closes the underlying db +func (dbInst *DB) Close() { + dbInst.mutex.Lock() + defer dbInst.mutex.Unlock() + if dbInst.dbState == closed { + return + } + if err := dbInst.db.Close(); err != nil { + logger.Errorf("Error closing badgerdb: %s", err) + } + dbInst.dbState = closed +} + +// Get returns the value for the given key +func (dbInst *DB) Get(key []byte) ([]byte, error) { + dbInst.mutex.RLock() + defer dbInst.mutex.RUnlock() + var value []byte + err := dbInst.db.View(func(txn *badger.Txn) error { + item, err := txn.Get(key) + if err != nil { + return err + } + value, err = item.ValueCopy(nil) + if err != nil { + return err + } + return nil + }) + if errors.Is(err, badger.ErrKeyNotFound) { + value = nil + err = nil + } + if err != nil { + logger.Errorf("Error retrieving badgerdb key [%#v]: %s", key, err) + return nil, errors.Wrapf(err, "error retrieving badgerdb key [%#v]", key) + } + return value, nil +} + +// Put saves the key/value. Last arg is sync on/off flag. It is unused as +// all badgerdb writes can survive process crashes or k8s environments +// with sync set to false. Sync is false by default. +func (dbInst *DB) Put(key []byte, value []byte, _ bool) error { + dbInst.mutex.RLock() + defer dbInst.mutex.RUnlock() + err := dbInst.db.Update(func(txn *badger.Txn) error { + err := txn.Set(key, value) + return err + }) + if err != nil { + logger.Errorf("Error writing badgerdb key [%#v]", key) + return errors.Wrapf(err, "error writing badgerdb key [%#v]", key) + } + return nil +} + +// Delete deletes the given key. Last arg is sync on/off flag. It is unused as +// all badgerdb writes can survive process crashes or k8s environments +// with sync set to false. Sync is false by default. +func (dbInst *DB) Delete(key []byte, _ bool) error { + dbInst.mutex.RLock() + defer dbInst.mutex.RUnlock() + err := dbInst.db.Update(func(txn *badger.Txn) error { + err := txn.Delete(key) + return err + }) + if err != nil { + logger.Errorf("Error deleting badgerdb key [%#v]", key) + return errors.Wrapf(err, "error deleting badgerdb key [%#v]", key) + } + return nil +} + +type RangeIterator struct { + iterator *badger.Iterator + startKey []byte + endKey []byte +} + +// Key wraps Badger's functions to make function similar Leveldb Key +func (itr *RangeIterator) Key() []byte { + return itr.iterator.Item().KeyCopy(nil) +} + +// GetIterator returns an iterator over key-value store. The iterator should be released after the use. +// The resultset contains all the keys that are present in the db between the startKey (inclusive) and the endKey (exclusive). +// A nil startKey represents the first available key and a nil endKey represent a logical key after the last available key +func (dbInst *DB) GetIterator(startKey []byte, endKey []byte) RangeIterator { + dbInst.mutex.RLock() + defer dbInst.mutex.RUnlock() + txn := dbInst.db.NewTransaction(true) + defer txn.Discard() + itr := txn.NewIterator(badger.DefaultIteratorOptions) + defer itr.Close() + return RangeIterator{ + iterator: itr, + startKey: startKey, + endKey: endKey, + } +} + +// WriteBatch writes a batch. Last arg is sync on/off flag. It is unused as +// all badgerdb writes can survive process crashes or k8s environments +// with sync set to false. Sync is false by default. +func (dbInst *DB) WriteBatch(batch *badger.WriteBatch, _ bool) error { + dbInst.mutex.RLock() + defer dbInst.mutex.RUnlock() + if err := batch.Flush(); err != nil { + return errors.Wrap(err, "error writing batch to badgerdb") + } + return nil +} + +// FileLock encapsulate the DB that holds the file lock. +// As the FileLock to be used by a single process/goroutine, +// there is no need for the semaphore to synchronize the +// FileLock usage. +type FileLock struct { + db *badger.DB + filePath string +} + +// NewFileLock returns a new file based lock manager. +func NewFileLock(filePath string) *FileLock { + return &FileLock{ + filePath: filePath, + } +} + +// Lock acquire a file lock. We achieve this by opening +// a db for the given filePath. Internally, badgerdb acquires a +// file lock while opening a db. If the db is opened again by the same or +// another process not in the read-only mode, error would be returned. +// When the db is closed or the owner process dies, the lock would be +// released and hence the other process can open the db. +func (f *FileLock) Lock() error { + var err error + var dirEmpty bool + if dirEmpty, err = fileutil.CreateDirIfMissing(f.filePath); err != nil { + panic(fmt.Sprintf("Error creating dir if missing: %s", err)) + } + if !dirEmpty { + if _, err := os.Stat(filepath.Join(f.filePath, ManifestFilename)); err != nil { + panic(fmt.Sprintf("Error opening badgerdb: %s", err)) + } + } + db, err := badger.Open(badger.DefaultOptions(f.filePath)) + if fmt.Sprint(err) == fmt.Sprintf("Cannot acquire directory lock on \"%s\". Another process is using this Badger database. error: resource temporarily unavailable", f.filePath) { + return errors.Errorf("lock is already acquired on file %s", f.filePath) + } + if err != nil { + panic(fmt.Sprintf("Error acquiring lock on file %s: %s", f.filePath, err)) + } + + // only mutate the lock db reference AFTER validating that the lock was held. + f.db = db + + return nil +} + +// Determine if the lock is currently held open. +func (f *FileLock) IsLocked() bool { + return f.db != nil +} + +// Unlock releases a previously acquired lock. We achieve this by closing +// the previously opened db. FileUnlock can be called multiple times. +func (f *FileLock) Unlock() { + if f.db == nil { + return + } + if err := f.db.Close(); err != nil { + logger.Warningf("unable to release the lock on file %s: %s", f.filePath, err) + return + } + f.db = nil +} diff --git a/common/ledger/util/badgerdbhelper/badgerdb_helper_test.go b/common/ledger/util/badgerdbhelper/badgerdb_helper_test.go new file mode 100644 index 00000000000..3b11e38e88e --- /dev/null +++ b/common/ledger/util/badgerdbhelper/badgerdb_helper_test.go @@ -0,0 +1,304 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package badgerdbhelper + +import ( + "fmt" + "io/ioutil" + "math/rand" + "os" + "path/filepath" + "testing" + "time" + + "github.com/stretchr/testify/require" +) + +func TestBadgerDBHelperWriteWithoutOpen(t *testing.T) { + env := newTestDBEnv(t, testDBPath) + defer env.cleanup() + db := env.db + defer func() { + if recover() == nil { + t.Fatalf("A panic is expected when writing to db before opening") + } + }() + db.Put([]byte("key"), []byte("value"), false) +} + +func TestBadgerDBHelperReadWithoutOpen(t *testing.T) { + env := newTestDBEnv(t, testDBPath) + defer env.cleanup() + db := env.db + defer func() { + if recover() == nil { + t.Fatalf("A panic is expected when writing to db before opening") + } + }() + db.Get([]byte("key")) +} + +func TestBadgerDBHelper(t *testing.T) { + env := newTestDBEnv(t, testDBPath) + // defer env.cleanup() + db := env.db + + db.Open() + // second time open should not have any side effect + db.Open() + IsEmpty, err := db.IsEmpty() + require.NoError(t, err) + require.True(t, IsEmpty) + db.Put([]byte("key1"), []byte("value1"), false) + db.Put([]byte("key2"), []byte("value2"), true) + db.Put([]byte("key3"), []byte("value3"), true) + + val, _ := db.Get([]byte("key2")) + require.Equal(t, "value2", string(val)) + + db.Delete([]byte("key1"), false) + db.Delete([]byte("key2"), true) + + val1, err1 := db.Get([]byte("key1")) + require.NoError(t, err1, "") + require.Equal(t, "", string(val1)) + + val2, err2 := db.Get([]byte("key2")) + require.NoError(t, err2, "") + require.Equal(t, "", string(val2)) + + db.Close() + // second time Close should not have any side effect + db.Close() + + _, err = db.IsEmpty() + require.Error(t, err) + + val3, err3 := db.Get([]byte("key3")) + require.Error(t, err3) + require.Equal(t, "", string(val3)) + + db.Open() + IsEmpty, err = db.IsEmpty() + require.NoError(t, err) + require.False(t, IsEmpty) + + batch := db.db.NewWriteBatch() + batch.Set([]byte("key1"), []byte("value1")) + batch.Set([]byte("key2"), []byte("value2")) + batch.Delete([]byte("key3")) + db.WriteBatch(batch, true) + + val1, err1 = db.Get([]byte("key1")) + require.NoError(t, err1, "") + require.Equal(t, "value1", string(val1)) + + val2, err2 = db.Get([]byte("key2")) + require.NoError(t, err2, "") + require.Equal(t, "value2", string(val2)) + + val3, err3 = db.Get([]byte("key3")) + require.NoError(t, err3, "") + require.Equal(t, "", string(val3)) + + keys := []string{} + itr := db.GetIterator(nil, nil) + for itr.iterator.Rewind(); itr.iterator.Valid(); itr.iterator.Next() { + keys = append(keys, string(itr.Key())) + } + require.Equal(t, []string{"key1", "key2"}, keys) +} + +func TestFileLock(t *testing.T) { + // create 1st fileLock manager + fileLockPath := testDBPath + "/fileLock" + fileLock1 := NewFileLock(fileLockPath) + require.Nil(t, fileLock1.db) + require.Equal(t, fileLock1.filePath, fileLockPath) + + // acquire the file lock using the fileLock manager 1 + err := fileLock1.Lock() + require.NoError(t, err) + require.NotNil(t, fileLock1.db) + + // create 2nd fileLock manager + fileLock2 := NewFileLock(fileLockPath) + require.Nil(t, fileLock2.db) + require.Equal(t, fileLock2.filePath, fileLockPath) + + // try to acquire the file lock again using the fileLock2 + // would result in an error + err = fileLock2.Lock() + expectedErr := fmt.Sprintf("lock is already acquired on file %s", fileLockPath) + require.EqualError(t, err, expectedErr) + require.Nil(t, fileLock2.db) + + // release the file lock acquired using fileLock1 + fileLock1.Unlock() + require.Nil(t, fileLock1.db) + + // As the fileLock1 has released the lock, + // the fileLock2 can acquire the lock. + err = fileLock2.Lock() + require.NoError(t, err) + require.NotNil(t, fileLock2.db) + + // release the file lock acquired using fileLock 2 + fileLock2.Unlock() + require.Nil(t, fileLock1.db) + + // unlock can be called multiple times and it is safe + fileLock2.Unlock() + require.Nil(t, fileLock1.db) + + // cleanup + require.NoError(t, os.RemoveAll(fileLockPath)) +} + +func TestFileLockLockUnlockLock(t *testing.T) { + // create an open lock + lockPath := testDBPath + "/fileLock" + lock := NewFileLock(lockPath) + require.Nil(t, lock.db) + require.Equal(t, lock.filePath, lockPath) + require.False(t, lock.IsLocked()) + + defer lock.Unlock() + defer os.RemoveAll(lockPath) + + // lock + require.NoError(t, lock.Lock()) + require.True(t, lock.IsLocked()) + + // lock + require.ErrorContains(t, lock.Lock(), "lock is already acquired") + + // unlock + lock.Unlock() + require.False(t, lock.IsLocked()) + + // lock - this should not error + require.NoError(t, lock.Lock()) + require.True(t, lock.IsLocked()) +} + +func TestCreateDBInEmptyDir(t *testing.T) { + require.NoError(t, os.RemoveAll(testDBPath), "") + require.NoError(t, os.MkdirAll(testDBPath, 0o775), "") + db := CreateDB(&Conf{DBPath: testDBPath}) + defer db.Close() + defer func() { + if r := recover(); r != nil { + t.Fatalf("Panic is not expected when opening db in an existing empty dir. %s", r) + } + }() + db.Open() +} + +func TestCreateDBInNonEmptyDir(t *testing.T) { + require.NoError(t, os.RemoveAll(testDBPath), "") + require.NoError(t, os.MkdirAll(testDBPath, 0o775), "") + file, err := os.Create(filepath.Join(testDBPath, "dummyfile.txt")) + require.NoError(t, err, "") + file.Close() + db := CreateDB(&Conf{DBPath: testDBPath}) + defer db.Close() + defer func() { + if r := recover(); r == nil { + t.Fatalf("A panic is expected when opening db in an existing non-empty dir. %s", r) + } + }() + db.Open() +} + +func BenchmarkBadgerDBHelper(b *testing.B) { + b.Run("get-badgerdb-little-data", BenchmarkGetBadgerDBWithLittleData) + // b.Run("get-badgerdb-big-data", BenchmarkGetBadgerDBWithBigData) + b.Run("put-badgerdb", BenchmarkPutBadgerDB) + b.Run("put-badgerdb-type-2", BenchmarkPutBadgerDB2) +} + +func BenchmarkGetBadgerDBWithLittleData(b *testing.B) { + db := createAndOpenDB() + db.Put([]byte("key1"), []byte("value1"), true) + db.Put([]byte("key2"), []byte("value2"), true) + db.Put([]byte("key3"), []byte(""), true) + db.Put([]byte("key4"), []byte("value4"), true) + db.Put([]byte("key5"), []byte("null"), true) + createdKeysAmount := 5 + randSource := rand.NewSource(time.Now().UnixNano()) + r := rand.New(randSource) + keys := make([][]byte, 500) + for i := range keys { + keys[i] = []byte(fmt.Sprintf("key%d", (r.Int() % createdKeysAmount))) + } + b.ResetTimer() + for i := 0; i < b.N; i++ { + _, _ = db.Get(keys[i%500]) + } +} + +func BenchmarkGetBadgerDBWithBigData(b *testing.B) { + db := createAndOpenDB() + keysTotalAmount := 1000 + keysToGetApproxAmount := 500 + for i := 0; i < keysTotalAmount; i++ { + _ = db.Put([]byte(createTestKey(i)), []byte(createTestValue("testdb", i)), true) + } + randSource := rand.NewSource(time.Now().UnixNano()) + r := rand.New(randSource) + keysToGet := make([][]byte, keysToGetApproxAmount) + for i := range keysToGet { + keysToGet[i] = []byte(createTestKey(r.Int() % keysTotalAmount)) + } + b.ResetTimer() + for i := 0; i < b.N; i++ { + _, _ = db.Get(keysToGet[i%keysToGetApproxAmount]) + } +} + +func BenchmarkPutBadgerDB(b *testing.B) { + db := createAndOpenDB() + keysAmount := 100000 + keys := make([][]byte, 0, keysAmount) + values := make([][]byte, 0, keysAmount) + for i := 0; i < keysAmount; i++ { + key := []byte(createTestKey(i)) + value := []byte(createTestValue("testdb", i)) + keys = append(keys, key) + values = append(values, value) + } + b.ResetTimer() + for i := 0; i < b.N; i++ { + _ = db.Put(keys[i%keysAmount], values[i%keysAmount], true) + } +} + +func BenchmarkPutBadgerDB2(b *testing.B) { + db := createAndOpenDB() + var key []byte + var value []byte + b.ResetTimer() + for i := 0; i < b.N; i++ { + b.StopTimer() + key = []byte(createTestKey(i)) + value = []byte(createTestValue("testdb", i)) + b.StartTimer() + _ = db.Put(key, value, true) + } +} + +func createAndOpenDB() *DB { + dbPath, _ := ioutil.TempDir("", "badgerdb") + defer os.RemoveAll(dbPath) + db := CreateDB(&Conf{ + DBPath: dbPath, + ExpectedFormat: "2.0", + }) + db.Open() + return db +} diff --git a/common/ledger/util/badgerdbhelper/badgerdb_provider.go b/common/ledger/util/badgerdbhelper/badgerdb_provider.go new file mode 100644 index 00000000000..d0ee53e1528 --- /dev/null +++ b/common/ledger/util/badgerdbhelper/badgerdb_provider.go @@ -0,0 +1,408 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package badgerdbhelper + +import ( + "bytes" + "errors" + "fmt" + "sync" + + badger "github.com/dgraph-io/badger/v4" + "github.com/hyperledger/fabric/common/ledger/dataformat" +) + +const ( + // internalDBName is used to keep track of data related to internals such as data format + // _ is used as name because this is not allowed as a channelname + internalDBName = "_" +) + +var ( + dbNameKeySep = []byte{0x00} + lastKeyIndicator = byte(0x01) + formatVersionKey = []byte{'f'} // a single key in db whose value indicates the version of the data format +) + +// closeFunc closes the db handle +type closeFunc func() + +// Conf configuration for `Provider` +// +// `ExpectedFormat` is the expected value of the format key in the internal database. +// At the time of opening the db, A check is performed that +// either the db is empty (i.e., opening for the first time) or the value +// of the formatVersionKey is equal to `ExpectedFormat`. Otherwise, an error is returned. +// A nil value for ExpectedFormat indicates that the format is never set and hence there is no such record. +type Conf struct { + DBPath string + ExpectedFormat string +} + +// Provider enables to use a single badgerdb as multiple logical badgerdbs +type Provider struct { + db *DB + + mux sync.Mutex + dbHandles map[string]*DBHandle +} + +// NewProvider constructs a Provider +func NewProvider(conf *Conf) (*Provider, error) { + db, err := openDBAndCheckFormat(conf) + if err != nil { + return nil, err + } + return &Provider{ + db: db, + dbHandles: make(map[string]*DBHandle), + }, nil +} + +func openDBAndCheckFormat(conf *Conf) (d *DB, e error) { + db := CreateDB(conf) + db.Open() + + defer func() { + if e != nil { + db.Close() + } + }() + + internalDB := &DBHandle{ + db: db, + dbName: internalDBName, + } + + dbEmpty, err := db.IsEmpty() + if err != nil { + return nil, err + } + + if dbEmpty && conf.ExpectedFormat != "" { + logger.Infof("DB is empty Setting db format as %s", conf.ExpectedFormat) + if err := internalDB.Put(formatVersionKey, []byte(conf.ExpectedFormat), true); err != nil { + return nil, err + } + return db, nil + } + + formatVersion, err := internalDB.Get(formatVersionKey) + if err != nil { + return nil, err + } + logger.Debugf("Checking for db format at path [%s]", conf.DBPath) + + if !bytes.Equal(formatVersion, []byte(conf.ExpectedFormat)) { + logger.Errorf("The db at path [%s] contains data in unexpected format. expected data format = [%s] (%#v), data format = [%s] (%#v).", + conf.DBPath, conf.ExpectedFormat, []byte(conf.ExpectedFormat), formatVersion, formatVersion) + return nil, &dataformat.ErrFormatMismatch{ + ExpectedFormat: conf.ExpectedFormat, + Format: string(formatVersion), + DBInfo: fmt.Sprintf("badgerdb at [%s]", conf.DBPath), + } + } + logger.Debug("format is latest, nothing to do") + return db, nil +} + +// GetDataFormat returns the format of the data +func (p *Provider) GetDataFormat() (string, error) { + f, err := p.GetDBHandle(internalDBName).Get(formatVersionKey) + return string(f), err +} + +// GetDBHandle returns a handle to a named db +func (p *Provider) GetDBHandle(dbName string) *DBHandle { + p.mux.Lock() + defer p.mux.Unlock() + dbHandle := p.dbHandles[dbName] + if dbHandle == nil { + closeFunc := func() { + p.mux.Lock() + defer p.mux.Unlock() + delete(p.dbHandles, dbName) + } + dbHandle = &DBHandle{dbName, p.db, closeFunc} + p.dbHandles[dbName] = dbHandle + } + return dbHandle +} + +// Close closes the underlying badgerdb +func (p *Provider) Close() { + p.db.Close() +} + +// Drop drops all the data for the given dbName +func (p *Provider) Drop(dbName string) error { + dbHandle := p.GetDBHandle(dbName) + defer dbHandle.Close() + return dbHandle.deleteAll() +} + +// DBHandle is an handle to a named db +type DBHandle struct { + dbName string + db *DB + closeFunc closeFunc +} + +func (h *DBHandle) GetMaxBatchSize() int64 { + return h.db.db.MaxBatchSize() +} + +func (h *DBHandle) GetMaxBatchCount() int64 { + return h.db.db.MaxBatchCount() +} + +// Get returns the value for the given key +func (h *DBHandle) Get(key []byte) ([]byte, error) { + return h.db.Get(constructLevelKey(h.dbName, key)) +} + +// Put saves the key/value +func (h *DBHandle) Put(key []byte, value []byte, sync bool) error { + return h.db.Put(constructLevelKey(h.dbName, key), value, sync) +} + +// Delete deletes the given key +func (h *DBHandle) Delete(key []byte, sync bool) error { + return h.db.Delete(constructLevelKey(h.dbName, key), sync) +} + +// DeleteAll deletes all the keys that belong to the channel (dbName). +func (h *DBHandle) deleteAll() error { + iter, err := h.GetIterator(nil, nil) + if err != nil { + return err + } + defer iter.iterator.Close() + + // use badgerdb iterator directly to be more efficient + dbIter := iter.iterator + + numKeys := 0 + var batchSize int64 + batchSize = 0 + batch := h.db.db.NewWriteBatch() + defer batch.Cancel() + for dbIter.Seek([]byte(h.dbName)); dbIter.ValidForPrefix([]byte(h.dbName)); dbIter.Next() { + key := dbIter.Item().KeyCopy(nil) + numKeys++ + batchSize = batchSize + int64(len(key)) + batch.Delete(key) + if batchSize >= h.db.db.MaxBatchSize() || numKeys == int(h.db.db.MaxBatchCount()) { + if err := h.db.WriteBatch(batch, true); err != nil { + return err + } + logger.Infof("Have removed %d entries for channel %s in badgerdb %s", numKeys, h.dbName, h.db.conf.DBPath) + batchSize = 0 + numKeys = 0 + batch = h.db.db.NewWriteBatch() + defer batch.Cancel() + } + } + if err = batch.Error(); err == nil { + return h.db.WriteBatch(batch, true) + } + return nil +} + +// IsEmpty returns true if no data exists for the DBHandle +func (h *DBHandle) IsEmpty() (bool, error) { + itr, err := h.GetIterator(nil, nil) + if err != nil { + return false, err + } + defer itr.iterator.Close() + + return !itr.Next(), nil +} + +// NewUpdateBatch returns a new UpdateBatch that can be used to update the db +func (h *DBHandle) NewUpdateBatch() *UpdateBatch { + return &UpdateBatch{ + dbName: h.dbName, + WriteBatch: h.db.db.NewWriteBatch(), + } +} + +// WriteBatch writes a batch in an atomic way +func (h *DBHandle) WriteBatch(batch *UpdateBatch, sync bool) error { + if h.db.db.IsClosed() { + return errors.New("error writing batch to badgerdb") + } + if batch == nil || batch.Error() != nil { + return nil + } + if err := h.db.WriteBatch(batch.WriteBatch, sync); err != nil { + return err + } + return nil +} + +// GetIterator gets an handle to iterator. The iterator should be released after the use. +// The resultset contains all the keys that are present in the db between the startKey (inclusive) and the endKey (exclusive). +// A nil startKey represents the first available key and a nil endKey represent a logical key after the last available key +func (h *DBHandle) GetIterator(startKey []byte, endKey []byte) (*Iterator, error) { + sKey := constructLevelKey(h.dbName, startKey) + eKey := constructLevelKey(h.dbName, endKey) + if endKey == nil { + // replace the last byte 'dbNameKeySep' by 'lastKeyIndicator' + eKey[len(eKey)-1] = lastKeyIndicator + } + logger.Debugf("Getting iterator for range [%#v] - [%#v]", sKey, eKey) + if h.db.dbState == closed { + err := errors.New("internal badgerdb error while obtaining db iterator: badgerdb: closed") + return nil, err + } + itr := h.db.GetIterator(sKey, eKey) + return &Iterator{h.dbName, itr, true, false, false}, nil +} + +// Close closes the DBHandle after its db data have been deleted +func (h *DBHandle) Close() { + if h.closeFunc != nil { + h.closeFunc() + } +} + +// UpdateBatch encloses the details of multiple `updates` +type UpdateBatch struct { + *badger.WriteBatch + dbName string +} + +// Put adds a KV +func (b *UpdateBatch) Put(key []byte, value []byte) { + if value == nil { + panic("Nil value not allowed") + } + if err := b.Set(constructLevelKey(b.dbName, key), value); err != nil { + logger.Errorf("Error while setting key [%#v] and value [%#v]: %#v", constructLevelKey(b.dbName, key), value, err) + } +} + +// Delete deletes a Key and associated value +func (b *UpdateBatch) Delete(key []byte) { + b.WriteBatch.Delete(constructLevelKey(b.dbName, key)) +} + +// Iterator extends actual badgerdb iterator +type Iterator struct { + dbName string + RangeIterator + justOpened bool + outOfRange bool + IgnoreNext bool +} + +// Key wraps actual badgerdb iterator method +func (itr *Iterator) Key() []byte { + key := itr.iterator.Item().KeyCopy(nil) + return retrieveAppKey(key) +} + +// Next wraps Badger's functions to make function similar Leveldb Next +func (itr *Iterator) Next() bool { + if itr.outOfRange { + return false + } + // Check does iterator need start from startKey + if itr.justOpened { + itr.iterator.Seek(itr.startKey) + itr.justOpened = false + if !itr.iterator.ValidForPrefix([]byte(itr.dbName)) { + return false + } + if itr.endKey != nil && bytes.Compare(itr.endKey, itr.iterator.Item().Key()) <= 0 { + itr.outOfRange = true + return false + } + return true + } + if !itr.IgnoreNext { + itr.iterator.Next() + } + itr.IgnoreNext = false + if !itr.iterator.ValidForPrefix([]byte(itr.dbName)) { + itr.outOfRange = true + return false + } + if itr.endKey != nil && bytes.Compare(itr.endKey, itr.iterator.Item().Key()) <= 0 { + itr.outOfRange = true + return false + } + if bytes.Equal(itr.endKey, itr.iterator.Item().Key()) { + itr.outOfRange = true + return false + } + return true +} + +// Release wraps Badger's function to make function similar Leveldb +func (itr *Iterator) Release() { + itr.iterator.Close() +} + +// First wraps Badger's function to make function similar Leveldb +func (itr *Iterator) First() bool { + itr.outOfRange = false + itr.iterator.Rewind() + return itr.iterator.Valid() +} + +func (itr *Iterator) Valid() bool { + if itr.outOfRange { + return false + } + if !itr.iterator.ValidForPrefix([]byte(itr.dbName)) || + bytes.Equal(itr.endKey, itr.iterator.Item().Key()) { + itr.outOfRange = true + return false + } + return true +} + +func (itr *Iterator) Value() []byte { + v, err := itr.iterator.Item().ValueCopy(nil) + if err != nil { + logger.Errorf("Error while getting ValueCopy: %#v", err) + return nil + } + return v +} + +// Seek moves the iterator to the first key/value pair +// whose key is greater than or equal to the given key. +// It returns whether such pair exist. +func (itr *Iterator) Seek(key []byte) bool { + itr.justOpened = false + itr.outOfRange = false + levelKey := constructLevelKey(itr.dbName, key) + if bytes.Compare(levelKey, itr.startKey) == -1 { + itr.iterator.Seek(itr.startKey) + return itr.iterator.ValidForPrefix([]byte(itr.dbName)) + } else if bytes.Compare(levelKey, itr.endKey) == 1 { + itr.iterator.Seek(itr.endKey) + itr.outOfRange = true + return itr.iterator.ValidForPrefix([]byte(itr.dbName)) + } + itr.iterator.Seek(levelKey) + return itr.iterator.ValidForPrefix([]byte(itr.dbName)) +} + +// constructLevelKey is similar to the same function from leveldb_provider +func constructLevelKey(dbName string, key []byte) []byte { + return append(append([]byte(dbName), dbNameKeySep...), key...) +} + +// retrieveAppKey is similar to the same function from leveldb_provider +func retrieveAppKey(levelKey []byte) []byte { + return bytes.SplitN(levelKey, dbNameKeySep, 2)[1] +} diff --git a/common/ledger/util/badgerdbhelper/badgerdb_provider_test.go b/common/ledger/util/badgerdbhelper/badgerdb_provider_test.go new file mode 100644 index 00000000000..44b190be48c --- /dev/null +++ b/common/ledger/util/badgerdbhelper/badgerdb_provider_test.go @@ -0,0 +1,570 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package badgerdbhelper + +import ( + "fmt" + "os" + "testing" + + "github.com/hyperledger/fabric-lib-go/common/flogging" + "github.com/hyperledger/fabric/common/ledger/dataformat" + + "github.com/stretchr/testify/require" +) + +func TestMain(m *testing.M) { + flogging.ActivateSpec("badgerdbhelper=debug") + os.Exit(m.Run()) +} + +func TestDBBasicWriteAndReads(t *testing.T) { + testDBBasicWriteAndReads(t, "db1", "db2", "") +} + +func TestIterator(t *testing.T) { + env := newTestProviderEnv(t, testDBPath) + defer env.cleanup() + p := env.provider + + db1 := p.GetDBHandle("db1") + db2 := p.GetDBHandle("db2") + db3 := p.GetDBHandle("db3") + for i := 0; i < 20; i++ { + db1.Put([]byte(createTestKey(i)), []byte(createTestValue("db1", i)), false) + db2.Put([]byte(createTestKey(i)), []byte(createTestValue("db2", i)), false) + db3.Put([]byte(createTestKey(i)), []byte(createTestValue("db3", i)), false) + } + + rangeTestCases := []struct { + startKey []byte + endKey []byte + expectedKeys []string + expectedValues []string + }{ + { + startKey: []byte(createTestKey(2)), + endKey: []byte(createTestKey(4)), + expectedKeys: createTestKeys(2, 3), + expectedValues: createTestValues("db2", 2, 3), + }, + { + startKey: []byte(createTestKey(2)), + endKey: nil, + expectedKeys: createTestKeys(2, 19), + expectedValues: createTestValues("db2", 2, 19), + }, + { + startKey: nil, + endKey: nil, + expectedKeys: createTestKeys(0, 19), + expectedValues: createTestValues("db2", 0, 19), + }, + } + + for i, testCase := range rangeTestCases { + t.Run( + fmt.Sprintf("range testCase %d", i), + func(t *testing.T) { + itr, err := db2.GetIterator(testCase.startKey, testCase.endKey) + require.NoError(t, err) + defer itr.Release() + checkItrResults(t, itr, testCase.expectedKeys, testCase.expectedValues) + }, + ) + } + + rangeWithSeekTestCases := []struct { + startKey []byte + endKey []byte + seekToKey []byte + itrAtKeyAfterSeek []byte + expectedKeys []string + expectedValues []string + }{ + { + startKey: nil, + endKey: nil, + seekToKey: []byte(createTestKey(10)), + itrAtKeyAfterSeek: []byte(createTestKey(10)), + expectedKeys: createTestKeys(11, 19), + expectedValues: createTestValues("db1", 11, 19), + }, + { + startKey: []byte(createTestKey(11)), + endKey: nil, + seekToKey: []byte(createTestKey(5)), + itrAtKeyAfterSeek: []byte(createTestKey(11)), + expectedKeys: createTestKeys(12, 19), + expectedValues: createTestValues("db1", 12, 19), + }, + { + startKey: nil, + endKey: nil, + seekToKey: []byte(createTestKey(19)), + itrAtKeyAfterSeek: []byte(createTestKey(19)), + expectedKeys: nil, + expectedValues: nil, + }, + } + + for i, testCase := range rangeWithSeekTestCases { + t.Run( + fmt.Sprintf("range with seek testCase %d", i), + func(t *testing.T) { + itr, err := db1.GetIterator(testCase.startKey, testCase.endKey) + require.NoError(t, err) + defer itr.Release() + require.True(t, itr.Seek(testCase.seekToKey)) + k := itr.Key() + require.Equal(t, testCase.itrAtKeyAfterSeek, k) + checkItrResults(t, itr, testCase.expectedKeys, testCase.expectedValues) + }, + ) + } + + t.Run("test-first-prev", func(t *testing.T) { + itr, err := db1.GetIterator(nil, nil) + require.NoError(t, err) + defer itr.Release() + require.True(t, itr.Seek([]byte(createTestKey(10)))) + require.Equal(t, []byte(createTestKey(10)), itr.Key()) + checkItrResults(t, itr, createTestKeys(11, 19), createTestValues("db1", 11, 19)) + + require.True(t, itr.First()) + require.True(t, itr.Seek([]byte(createTestKey(10)))) + require.Equal(t, []byte(createTestKey(10)), itr.Key()) + // require.True(t, itr.Prev()) // There is no function Prev in Badger. So I use Seek to return to previous key + require.True(t, itr.Seek([]byte(createTestKey(9)))) + checkItrResults(t, itr, createTestKeys(10, 19), createTestValues("db1", 10, 19)) // + + require.True(t, itr.First()) + require.False(t, itr.Seek([]byte(createTestKey(20)))) + require.True(t, itr.First()) + checkItrResults(t, itr, createTestKeys(1, 19), createTestValues("db1", 1, 19)) + + require.True(t, itr.First()) + // require.False(t, itr.Prev()) // There is no function Prev in Badger. + checkItrResults(t, itr, createTestKeys(1, 19), createTestValues("db1", 1, 19)) // Changed because Prev is not used + + // require.True(t, itr.First()) + // require.True(t, itr.Last()) // There is no function Last in Badger. + // checkItrResults(t, itr, nil, nil) + }) + + t.Run("test-error-path", func(t *testing.T) { + env.provider.Close() + itr, err := db1.GetIterator(nil, nil) + require.EqualError(t, err, "internal badgerdb error while obtaining db iterator: badgerdb: closed") + require.Nil(t, itr) + }) +} + +func TestBatchedUpdates(t *testing.T) { + env := newTestProviderEnv(t, testDBPath) + defer env.cleanup() + p := env.provider + + db1 := p.GetDBHandle("db1") + db2 := p.GetDBHandle("db2") + + dbs := []*DBHandle{db1, db2} + for _, db := range dbs { + batch := db.NewUpdateBatch() + batch.Put([]byte("key1"), []byte("value1")) + batch.Put([]byte("key2"), []byte("value2")) + batch.Put([]byte("key3"), []byte("value3")) + db.WriteBatch(batch, true) + } + + for _, db := range dbs { + batch := db.NewUpdateBatch() + batch.Delete([]byte("key2")) + db.WriteBatch(batch, true) + } + + for _, db := range dbs { + val1, _ := db.Get([]byte("key1")) + require.Equal(t, "value1", string(val1)) + + val2, err2 := db.Get([]byte("key2")) + require.NoError(t, err2, "") + require.Nil(t, val2) + + val3, _ := db.Get([]byte("key3")) + require.Equal(t, "value3", string(val3)) + } +} + +func TestDrop(t *testing.T) { + env := newTestProviderEnv(t, testDBPath) + defer env.cleanup() + p := env.provider + + db1 := p.GetDBHandle("db1") + db2 := p.GetDBHandle("db2") + db3 := p.GetDBHandle("db3") + + require.Contains(t, p.dbHandles, "db1") + require.Contains(t, p.dbHandles, "db2") + require.Contains(t, p.dbHandles, "db3") + + for i := 0; i < 20; i++ { + db1.Put([]byte(createTestKey(i)), []byte(createTestValue("db1", i)), false) + db2.Put([]byte(createTestKey(i)), []byte(createTestValue("db2", i)), false) + } + // db3 is used to test remove when multiple batches are needed (each long key has 125 bytes) + for i := 0; i < 2; i++ { + db3.Put([]byte(createTestLongKey(i)), []byte(createTestValue("db3", i)), false) + } + + expectedSetup := []struct { + db *DBHandle + expectedKeys []string + expectedValues []string + }{ + { + db: db1, + expectedKeys: createTestKeys(0, 19), + expectedValues: createTestValues("db1", 0, 19), + }, + { + db: db2, + expectedKeys: createTestKeys(0, 19), + expectedValues: createTestValues("db2", 0, 19), + }, + { + db: db3, + expectedKeys: createTestLongKeys(0, 1), + expectedValues: createTestValues("db3", 0, 1), + }, + } + + for _, dbSetup := range expectedSetup { + itr, err := dbSetup.db.GetIterator(nil, nil) + require.NoError(t, err) + checkItrResults(t, itr, dbSetup.expectedKeys, dbSetup.expectedValues) + itr.Release() + } + + require.NoError(t, p.Drop("db1")) + require.NoError(t, p.Drop("db3")) + + require.NotContains(t, p.dbHandles, "db1") + require.NotContains(t, p.dbHandles, "db3") + require.Contains(t, p.dbHandles, "db2") + + expectedResults := []struct { + db *DBHandle + expectedKeys []string + expectedValues []string + }{ + { + db: db1, + expectedKeys: nil, + expectedValues: nil, + }, + { + db: db2, + expectedKeys: createTestKeys(0, 19), + expectedValues: createTestValues("db2", 0, 19), + }, + { + db: db3, + expectedKeys: nil, + expectedValues: nil, + }, + } + + for _, result := range expectedResults { + itr, err := result.db.GetIterator(nil, nil) + require.NoError(t, err) + checkItrResults(t, itr, result.expectedKeys, result.expectedValues) + itr.Release() + } + + // negative test + p.Close() + require.EqualError(t, db2.deleteAll(), "internal badgerdb error while obtaining db iterator: badgerdb: closed") +} + +func TestFormatCheck(t *testing.T) { + testCases := []struct { + dataFormat string + dataExists bool + expectedFormat string + expectedErr *dataformat.ErrFormatMismatch + }{ + { + dataFormat: "", + dataExists: true, + expectedFormat: "", + expectedErr: nil, + }, + { + dataFormat: "", + dataExists: false, + expectedFormat: "", + expectedErr: nil, + }, + { + dataFormat: "", + dataExists: false, + expectedFormat: "2.0", + expectedErr: nil, + }, + { + dataFormat: "", + dataExists: true, + expectedFormat: "2.0", + expectedErr: &dataformat.ErrFormatMismatch{Format: "", ExpectedFormat: "2.0"}, + }, + { + dataFormat: "2.0", + dataExists: true, + expectedFormat: "2.0", + expectedErr: nil, + }, + { + dataFormat: "2.0", + dataExists: true, + expectedFormat: "3.0", + expectedErr: &dataformat.ErrFormatMismatch{Format: "2.0", ExpectedFormat: "3.0"}, + }, + } + + for i, testCase := range testCases { + t.Run( + fmt.Sprintf("testCase %d", i), + func(t *testing.T) { + testFormatCheck(t, testCase.dataFormat, testCase.expectedFormat, testCase.dataExists, testCase.expectedErr) + }) + } +} + +func TestClose(t *testing.T) { + env := newTestProviderEnv(t, testDBPath) + defer env.cleanup() + p := env.provider + + db1 := p.GetDBHandle("db1") + db2 := p.GetDBHandle("db2") + + expectedDBHandles := map[string]*DBHandle{ + "db1": db1, + "db2": db2, + } + require.Equal(t, expectedDBHandles, p.dbHandles) + + db1.Close() + expectedDBHandles = map[string]*DBHandle{ + "db2": db2, + } + require.Equal(t, expectedDBHandles, p.dbHandles) + + db2.Close() + require.Equal(t, map[string]*DBHandle{}, p.dbHandles) +} + +func TestIsEmpty(t *testing.T) { + var env *testDBProviderEnv + var db1, db2 *DBHandle + + setup := func() { + env = newTestProviderEnv(t, testDBPath) + p := env.provider + db1 = p.GetDBHandle("db1") + db2 = p.GetDBHandle("db2") + } + + cleanup := func() { + env.cleanup() + } + + t.Run("both the dbs are empty", func(t *testing.T) { + setup() + defer cleanup() + + empty, err := db1.IsEmpty() + require.NoError(t, err) + require.True(t, empty) + + empty, err = db2.IsEmpty() + require.NoError(t, err) + require.True(t, empty) + }) + + t.Run("only one db is empty", func(t *testing.T) { + setup() + defer cleanup() + + db1.Put([]byte("key"), []byte("value"), false) + empty, err := db1.IsEmpty() + require.NoError(t, err) + require.False(t, empty) + + empty, err = db2.IsEmpty() + require.NoError(t, err) + require.True(t, empty) + }) + + t.Run("both the dbs contain data", func(t *testing.T) { + setup() + defer cleanup() + + db1.Put([]byte("key"), []byte("value"), false) + db2.Put([]byte("key"), []byte("value"), false) + + empty, err := db1.IsEmpty() + require.NoError(t, err) + require.False(t, empty) + + empty, err = db2.IsEmpty() + require.NoError(t, err) + require.False(t, empty) + }) + + t.Run("iter error", func(t *testing.T) { + setup() + defer cleanup() + + env.provider.Close() + empty, err := db1.IsEmpty() + require.EqualError(t, err, "internal badgerdb error while obtaining db iterator: badgerdb: closed") + require.False(t, empty) + + empty, err = db2.IsEmpty() + require.EqualError(t, err, "internal badgerdb error while obtaining db iterator: badgerdb: closed") + require.False(t, empty) + }) +} + +func testFormatCheck(t *testing.T, dataFormat, expectedFormat string, dataExists bool, expectedErr *dataformat.ErrFormatMismatch) { + require.NoError(t, os.RemoveAll(testDBPath)) + defer func() { + require.NoError(t, os.RemoveAll(testDBPath)) + }() + + // setup test pre-conditions (create a db with dbformat) + p, err := NewProvider(&Conf{DBPath: testDBPath, ExpectedFormat: dataFormat}) + require.NoError(t, err) + f, err := p.GetDataFormat() + require.NoError(t, err) + require.Equal(t, dataFormat, f) + if dataExists { + require.NoError(t, p.GetDBHandle("testdb").Put([]byte("key"), []byte("value"), true)) + } + + // close and reopen with new conf + p.Close() + p, err = NewProvider(&Conf{DBPath: testDBPath, ExpectedFormat: expectedFormat}) + if expectedErr != nil { + expectedErr.DBInfo = fmt.Sprintf("badgerdb at [%s]", testDBPath) + require.Equal(t, err, expectedErr) + return + } + require.NoError(t, err) + f, err = p.GetDataFormat() + require.NoError(t, err) + require.Equal(t, expectedFormat, f) +} + +func testDBBasicWriteAndReads(t *testing.T, dbNames ...string) { + env := newTestProviderEnv(t, testDBPath) + defer env.cleanup() + p := env.provider + + for _, dbName := range dbNames { + db := p.GetDBHandle(dbName) + db.Put([]byte("key1"), []byte("value1_"+dbName), false) + db.Put([]byte("key2"), []byte("value2_"+dbName), false) + db.Put([]byte("key3"), []byte("value3_"+dbName), false) + } + + for _, dbName := range dbNames { + db := p.GetDBHandle(dbName) + val, err := db.Get([]byte("key1")) + require.NoError(t, err, "") + require.Equal(t, []byte("value1_"+dbName), val) + + val, err = db.Get([]byte("key2")) + require.NoError(t, err, "") + require.Equal(t, []byte("value2_"+dbName), val) + + val, err = db.Get([]byte("key3")) + require.NoError(t, err, "") + require.Equal(t, []byte("value3_"+dbName), val) + } + + for _, dbName := range dbNames { + db := p.GetDBHandle(dbName) + require.NoError(t, db.Delete([]byte("key1"), false), "") + val, err := db.Get([]byte("key1")) + require.NoError(t, err, "") + require.Nil(t, val) + + require.NoError(t, db.Delete([]byte("key2"), false), "") + val, err = db.Get([]byte("key2")) + require.NoError(t, err, "") + require.Nil(t, val) + + require.NoError(t, db.Delete([]byte("key3"), false), "") + val, err = db.Get([]byte("key3")) + require.NoError(t, err, "") + require.Nil(t, val) + } +} + +func checkItrResults(t *testing.T, itr *Iterator, expectedKeys []string, expectedValues []string) { + var actualKeys []string + var actualValues []string + for itr.Next(); itr.Valid(); itr.Next() { + actualKeys = append(actualKeys, string(itr.Key())) + actualValues = append(actualValues, string(itr.Value())) + } + require.Equal(t, expectedKeys, actualKeys) + require.Equal(t, expectedValues, actualValues) + require.Equal(t, false, itr.Next()) +} + +func createTestKey(i int) string { + return fmt.Sprintf("key_%06d", i) +} + +const padding100 = "_0123456789_0123456789_0123456789_0123456789_0123456789_0123456789_0123456789_0123456789_0123456789_" + +func createTestLongKey(i int) string { + return fmt.Sprintf("key_%s_%10d", padding100, i) +} + +func createTestValue(dbname string, i int) string { + return fmt.Sprintf("value_%s_%06d", dbname, i) +} + +func createTestKeys(start int, end int) []string { + var keys []string + for i := start; i <= end; i++ { + keys = append(keys, createTestKey(i)) + } + return keys +} + +func createTestLongKeys(start int, end int) []string { + var keys []string + for i := start; i <= end; i++ { + keys = append(keys, createTestLongKey(i)) + } + return keys +} + +func createTestValues(dbname string, start int, end int) []string { + var values []string + for i := start; i <= end; i++ { + values = append(values, createTestValue(dbname, i)) + } + return values +} diff --git a/common/ledger/util/badgerdbhelper/pkg_test.go b/common/ledger/util/badgerdbhelper/pkg_test.go new file mode 100644 index 00000000000..ce9c6863268 --- /dev/null +++ b/common/ledger/util/badgerdbhelper/pkg_test.go @@ -0,0 +1,60 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package badgerdbhelper + +import ( + "os" + "testing" + + "github.com/stretchr/testify/require" +) + +const testDBPath = "/tmp/fabric/ledgertests/util/badgerdbhelper" + +type testDBEnv struct { + t *testing.T + path string + db *DB +} + +type testDBProviderEnv struct { + t *testing.T + path string + provider *Provider +} + +func newTestDBEnv(t *testing.T, path string) *testDBEnv { + testDBEnv := &testDBEnv{t: t, path: path} + testDBEnv.cleanup() + testDBEnv.db = CreateDB(&Conf{DBPath: path}) + return testDBEnv +} + +func newTestProviderEnv(t *testing.T, path string) *testDBProviderEnv { + testProviderEnv := &testDBProviderEnv{t: t, path: path} + testProviderEnv.cleanup() + var err error + testProviderEnv.provider, err = NewProvider(&Conf{DBPath: path}) + if err != nil { + panic(err) + } + return testProviderEnv +} + +func (dbEnv *testDBEnv) cleanup() { + if dbEnv.db != nil { + dbEnv.db.Close() + } + require.NoError(dbEnv.t, os.RemoveAll(dbEnv.path)) +} + +func (providerEnv *testDBProviderEnv) cleanup() { + if providerEnv.provider != nil { + providerEnv.provider.Close() + } + require.NoError(providerEnv.t, os.RemoveAll(providerEnv.path)) +} diff --git a/common/ledger/util/leveldbhelper/leveldb_helper_test.go b/common/ledger/util/leveldbhelper/leveldb_helper_test.go index 0231bc5b23b..90cb2287a4f 100644 --- a/common/ledger/util/leveldbhelper/leveldb_helper_test.go +++ b/common/ledger/util/leveldbhelper/leveldb_helper_test.go @@ -8,9 +8,12 @@ package leveldbhelper import ( "fmt" + "io/ioutil" + "math/rand" "os" "path/filepath" "testing" + "time" "github.com/stretchr/testify/require" "github.com/syndtr/goleveldb/leveldb" @@ -213,3 +216,91 @@ func TestCreateDBInNonEmptyDir(t *testing.T) { }() db.Open() } + +func BenchmarkLevelDBHelper(b *testing.B) { + b.Run("get-leveldb-little-data", BenchmarkGetLevelDBWithLittleData) + // b.Run("get-leveldb-big-data", BenchmarkGetLevelDBWithBigData) + b.Run("put-leveldb", BenchmarkPutLevelDB) + b.Run("put-leveldb-type-2", BenchmarkPutLevelDB2) +} + +func BenchmarkGetLevelDBWithLittleData(b *testing.B) { + db := createAndOpenDB() + db.Put([]byte("key1"), []byte("value1"), true) + db.Put([]byte("key2"), []byte("value2"), true) + db.Put([]byte("key3"), []byte(""), true) + db.Put([]byte("key4"), []byte("value4"), true) + db.Put([]byte("key5"), []byte("null"), true) + createdKeysAmount := 5 + randSource := rand.NewSource(time.Now().UnixNano()) + r := rand.New(randSource) + keys := make([][]byte, 500) + for i := range keys { + keys[i] = []byte(fmt.Sprintf("key%d", (r.Int() % createdKeysAmount))) + } + b.ResetTimer() + for i := 0; i < b.N; i++ { + _, _ = db.Get(keys[i%500]) + } +} + +func BenchmarkGetLevelDBWithBigData(b *testing.B) { + db := createAndOpenDB() + keysTotalAmount := 1000 + keysToGetApproxAmount := 500 + for i := 0; i < keysTotalAmount; i++ { + _ = db.Put([]byte(createTestKey(i)), []byte(createTestValue("testdb", i)), true) + } + randSource := rand.NewSource(time.Now().UnixNano()) + r := rand.New(randSource) + keysToGet := make([][]byte, keysToGetApproxAmount) + for i := range keysToGet { + keysToGet[i] = []byte(createTestKey(r.Int() % keysTotalAmount)) + } + b.ResetTimer() + for i := 0; i < b.N; i++ { + _, _ = db.Get(keysToGet[i%keysToGetApproxAmount]) + } +} + +func BenchmarkPutLevelDB(b *testing.B) { + db := createAndOpenDB() + keysAmount := 100000 + keys := make([][]byte, 0, keysAmount) + values := make([][]byte, 0, keysAmount) + for i := 0; i < keysAmount; i++ { + key := []byte(createTestKey(i)) + value := []byte(createTestValue("testdb", i)) + keys = append(keys, key) + values = append(values, value) + } + b.ResetTimer() + for i := 0; i < b.N; i++ { + _ = db.Put(keys[i], values[i], true) + } +} + +func BenchmarkPutLevelDB2(b *testing.B) { + db := createAndOpenDB() + var key []byte + var value []byte + b.ResetTimer() + for i := 0; i < b.N; i++ { + b.StopTimer() + key = []byte(createTestKey(i)) + value = []byte(createTestValue("testdb", i)) + b.StartTimer() + _ = db.Put(key, value, true) + } +} + +func createAndOpenDB() *DB { + dbPath, _ := ioutil.TempDir("", "stateleveldb") + defer os.RemoveAll(dbPath) + db := CreateDB(&Conf{ + DBPath: dbPath, + ExpectedFormat: "2.0", + }) + db.Open() + return db +} diff --git a/core/ledger/kvledger/kv_ledger_provider.go b/core/ledger/kvledger/kv_ledger_provider.go index 445ee38c19f..a87f5f88d07 100644 --- a/core/ledger/kvledger/kv_ledger_provider.go +++ b/core/ledger/kvledger/kv_ledger_provider.go @@ -228,7 +228,7 @@ func (p *Provider) initStateDBProvider() error { } stateDBConfig := &privacyenabledstate.StateDBConfig{ StateDBConfig: p.initializer.Config.StateDBConfig, - LevelDBPath: StateDBPath(p.initializer.Config.RootFSPath), + StateDBPath: StateDBPath(p.initializer.Config.RootFSPath), } sysNamespaces := p.initializer.DeployedChaincodeInfoProvider.Namespaces() p.dbProvider, err = privacyenabledstate.NewDBProvider( diff --git a/core/ledger/kvledger/ledger_filepath.go b/core/ledger/kvledger/ledger_filepath.go index 724d56e5fe3..a7519685405 100644 --- a/core/ledger/kvledger/ledger_filepath.go +++ b/core/ledger/kvledger/ledger_filepath.go @@ -32,6 +32,7 @@ func PvtDataStorePath(rootFSPath string) string { // StateDBPath returns the absolute path of state level DB func StateDBPath(rootFSPath string) string { + // this catalog name doesn't mean that StateDB is goleveldb return filepath.Join(rootFSPath, "stateLeveldb") } diff --git a/core/ledger/kvledger/txmgmt/privacyenabledstate/db.go b/core/ledger/kvledger/txmgmt/privacyenabledstate/db.go index 270c618c55b..76987afd9df 100644 --- a/core/ledger/kvledger/txmgmt/privacyenabledstate/db.go +++ b/core/ledger/kvledger/txmgmt/privacyenabledstate/db.go @@ -19,6 +19,7 @@ import ( "github.com/hyperledger/fabric/core/ledger/internal/version" "github.com/hyperledger/fabric/core/ledger/kvledger/bookkeeping" "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb" + "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb/statebadgerdb" "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb/statecouchdb" "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb/stateleveldb" "github.com/hyperledger/fabric/core/ledger/util" @@ -37,10 +38,10 @@ const ( type StateDBConfig struct { // ledger.StateDBConfig is used to configure the stateDB for the ledger. *ledger.StateDBConfig - // LevelDBPath is the filesystem path when statedb type is "goleveldb". + // StateDBPath is the filesystem path when statedb type is "goleveldb" or "Badger". // It is internally computed by the ledger component, // so it is not in ledger.StateDBConfig and not exposed to other components. - LevelDBPath string + StateDBPath string } // DBProvider encapsulates other providers such as VersionedDBProvider and @@ -66,8 +67,12 @@ func NewDBProvider( if vdbProvider, err = statecouchdb.NewVersionedDBProvider(stateDBConf.CouchDB, metricsProvider, sysNamespaces); err != nil { return nil, err } + } else if stateDBConf != nil && stateDBConf.StateDatabase == ledger.Badger { + if vdbProvider, err = statebadgerdb.NewVersionedDBProvider(stateDBConf.StateDBPath); err != nil { + return nil, err + } } else { - if vdbProvider, err = stateleveldb.NewVersionedDBProvider(stateDBConf.LevelDBPath); err != nil { + if vdbProvider, err = stateleveldb.NewVersionedDBProvider(stateDBConf.StateDBPath); err != nil { return nil, err } } diff --git a/core/ledger/kvledger/txmgmt/privacyenabledstate/test_exports.go b/core/ledger/kvledger/txmgmt/privacyenabledstate/test_exports.go index 1be84f5ce32..9a46f513fce 100644 --- a/core/ledger/kvledger/txmgmt/privacyenabledstate/test_exports.go +++ b/core/ledger/kvledger/txmgmt/privacyenabledstate/test_exports.go @@ -147,7 +147,7 @@ func (env *CouchDBTestEnv) Init(t testing.TB) { RedoLogPath: redoPath, }, }, - LevelDBPath: "", + StateDBPath: "", } env.bookkeeperTestEnv = bookkeeping.NewTestEnv(t) diff --git a/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb.go b/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb.go new file mode 100644 index 00000000000..589d9f408dc --- /dev/null +++ b/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb.go @@ -0,0 +1,407 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package statebadgerdb + +import ( + "github.com/hyperledger/fabric-lib-go/common/flogging" + "github.com/hyperledger/fabric/common/ledger/dataformat" + "github.com/hyperledger/fabric/common/ledger/util/badgerdbhelper" + "github.com/hyperledger/fabric/core/ledger/internal/version" + "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb" + kvdb "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb/stateleveldb" + "github.com/pkg/errors" +) + +var logger = flogging.MustGetLogger("statebadgerdb") + +var ( + dataKeyPrefix = []byte{'d'} + dataKeyStopper = []byte{'e'} + nsKeySep = []byte{0x00} + lastKeyIndicator = byte(0x01) + savePointKey = []byte{'s'} + maxDataImportBatchSize = 4 * 1024 * 1024 +) + +// VersionedDBProvider implements interface VersionedDBProvider +type VersionedDBProvider struct { + dbProvider *badgerdbhelper.Provider +} + +// NewVersionedDBProvider instantiates VersionedDBProvider +func NewVersionedDBProvider(dbPath string) (*VersionedDBProvider, error) { + logger.Debugf("constructing VersionedDBProvider dbPath=%s", dbPath) + dbProvider, err := badgerdbhelper.NewProvider( + &badgerdbhelper.Conf{ + DBPath: dbPath, + ExpectedFormat: dataformat.CurrentFormat, + }) + if err != nil { + return nil, err + } + return &VersionedDBProvider{dbProvider}, nil +} + +// GetDBHandle gets the handle to a named database +func (provider *VersionedDBProvider) GetDBHandle(dbName string, namespaceProvider statedb.NamespaceProvider) (statedb.VersionedDB, error) { + return newVersionedDB(provider.dbProvider.GetDBHandle(dbName), dbName), nil +} + +// ImportFromSnapshot loads the public state and pvtdata hashes from the snapshot files previously generated +func (provider *VersionedDBProvider) ImportFromSnapshot( + dbName string, + savepoint *version.Height, + itr statedb.FullScanIterator, +) error { + vdb := newVersionedDB(provider.dbProvider.GetDBHandle(dbName), dbName) + return vdb.importState(itr, savepoint) +} + +// BytesKeySupported returns true if a db created supports bytes as a key +func (provider *VersionedDBProvider) BytesKeySupported() bool { + return true +} + +// Close closes the underlying db +func (provider *VersionedDBProvider) Close() { + provider.dbProvider.Close() +} + +// Drop drops channel-specific data from the state badgerdb. +// It is not an error if a database does not exist. +func (provider *VersionedDBProvider) Drop(dbName string) error { + return provider.dbProvider.Drop(dbName) +} + +// VersionedDB implements VersionedDB interface +type versionedDB struct { + db *badgerdbhelper.DBHandle + dbName string +} + +// newVersionedDB constructs an instance of VersionedDB +func newVersionedDB(db *badgerdbhelper.DBHandle, dbName string) *versionedDB { + return &versionedDB{db, dbName} +} + +// Open implements method in VersionedDB interface +func (vdb *versionedDB) Open() error { + // do nothing because shared db is used + return nil +} + +// Close implements method in VersionedDB interface +func (vdb *versionedDB) Close() { + // do nothing because shared db is used +} + +// ValidateKeyValue implements method in VersionedDB interface +func (vdb *versionedDB) ValidateKeyValue(key string, value []byte) error { + return nil +} + +// BytesKeySupported implements method in VersionedDB interface +func (vdb *versionedDB) BytesKeySupported() bool { + return true +} + +// GetState implements method in VersionedDB interface +func (vdb *versionedDB) GetState(namespace string, key string) (*statedb.VersionedValue, error) { + logger.Debugf("GetState(). ns=%s, key=%s", namespace, key) + dbVal, err := vdb.db.Get(kvdb.EncodeDataKey(namespace, key)) + if err != nil { + return nil, err + } + if dbVal == nil { + return nil, nil + } + return kvdb.DecodeValue(dbVal) +} + +// GetVersion implements method in VersionedDB interface +func (vdb *versionedDB) GetVersion(namespace string, key string) (*version.Height, error) { + versionedValue, err := vdb.GetState(namespace, key) + if err != nil { + return nil, err + } + if versionedValue == nil { + return nil, nil + } + return versionedValue.Version, nil +} + +// GetStateMultipleKeys implements method in VersionedDB interface +func (vdb *versionedDB) GetStateMultipleKeys(namespace string, keys []string) ([]*statedb.VersionedValue, error) { + vals := make([]*statedb.VersionedValue, len(keys)) + for i, key := range keys { + val, err := vdb.GetState(namespace, key) + if err != nil { + return nil, err + } + vals[i] = val + } + return vals, nil +} + +// GetStateRangeScanIterator implements method in VersionedDB interface +// startKey is inclusive +// endKey is exclusive +func (vdb *versionedDB) GetStateRangeScanIterator(namespace string, startKey string, endKey string) (statedb.ResultsIterator, error) { + // pageSize = 0 denotes unlimited page size + return vdb.GetStateRangeScanIteratorWithPagination(namespace, startKey, endKey, 0) +} + +// GetStateRangeScanIteratorWithPagination implements method in VersionedDB interface +func (vdb *versionedDB) GetStateRangeScanIteratorWithPagination(namespace string, startKey string, endKey string, pageSize int32) (statedb.QueryResultsIterator, error) { + dataStartKey := kvdb.EncodeDataKey(namespace, startKey) + dataEndKey := kvdb.EncodeDataKey(namespace, endKey) + if endKey == "" { + dataEndKey[len(dataEndKey)-1] = lastKeyIndicator + } else { + logger.Debugf("endKey is not empty") + dataEndKey = kvdb.EncodeDataKey(namespace, endKey) + } + dbItr, err := vdb.db.GetIterator(dataStartKey, dataEndKey) + if err != nil { + return nil, err + } + return newKVScanner(namespace, dbItr, pageSize), nil +} + +// ExecuteQuery implements method in VersionedDB interface +func (vdb *versionedDB) ExecuteQuery(namespace, query string) (statedb.ResultsIterator, error) { + return nil, errors.New("ExecuteQuery not supported for badgerdb") +} + +// ExecuteQueryWithPagination implements method in VersionedDB interface +func (vdb *versionedDB) ExecuteQueryWithPagination(namespace, query, bookmark string, pageSize int32) (statedb.QueryResultsIterator, error) { + return nil, errors.New("ExecuteQueryWithMetadata not supported for badgerdb") +} + +// ApplyUpdates implements method in VersionedDB interface +func (vdb *versionedDB) ApplyUpdates(batch *statedb.UpdateBatch, height *version.Height) error { + dbBatch := vdb.db.NewUpdateBatch() + defer dbBatch.WriteBatch.Cancel() + namespaces := batch.GetUpdatedNamespaces() + for _, ns := range namespaces { + updates := batch.GetUpdates(ns) + for k, vv := range updates { + dataKey := kvdb.EncodeDataKey(ns, k) + logger.Debugf("Channel [%s]: Applying key(string)=[%s] key(bytes)=[%#v]", vdb.dbName, string(dataKey), dataKey) + + if vv.Value == nil { + dbBatch.Delete(dataKey) + } else { + encodedVal, err := kvdb.EncodeValue(vv) + if err != nil { + return err + } + dbBatch.Put(dataKey, encodedVal) + } + } + } + // Record a savepoint at a given height + // If a given height is nil, it denotes that we are committing pvt data of old blocks. + // In this case, we should not store a savepoint for recovery. The lastUpdatedOldBlockList + // in the pvtstore acts as a savepoint for pvt data. + if height != nil { + dbBatch.Put(savePointKey, height.ToBytes()) + } + // Setting snyc to true as a precaution, false may be an ok optimization after further testing. + return vdb.db.WriteBatch(dbBatch, true) +} + +// GetLatestSavePoint implements method in VersionedDB interface +func (vdb *versionedDB) GetLatestSavePoint() (*version.Height, error) { + versionBytes, err := vdb.db.Get(savePointKey) + if err != nil { + return nil, err + } + if versionBytes == nil { + return nil, nil + } + version, _, err := version.NewHeightFromBytes(versionBytes) + if err != nil { + return nil, err + } + return version, nil +} + +// GetFullScanIterator implements method in VersionedDB interface. This function returns a +// FullScanIterator that can be used to iterate over entire data in the statedb for a channel. +// `skipNamespace` parameter can be used to control if the consumer wants the FullScanIterator +// to skip one or more namespaces from the returned results. The intended use of this iterator +// is to generate the snapshot files for the statebadgerdb +func (vdb *versionedDB) GetFullScanIterator(skipNamespace func(string) bool) (statedb.FullScanIterator, error) { + return newFullDBScanner(vdb.db, skipNamespace) +} + +// importState implements method in VersionedDB interface. The function is expected to be used +// for importing the state from a previously snapshotted state. The parameter itr provides access to +// the snapshotted state. +func (vdb *versionedDB) importState(itr statedb.FullScanIterator, savepoint *version.Height) error { + if itr == nil { + return vdb.db.Put(savePointKey, savepoint.ToBytes(), true) + } + dbBatch := vdb.db.NewUpdateBatch() + defer dbBatch.Cancel() + if err := vdb.db.WriteBatch(dbBatch, true); err != nil { + return err + } + dbBatch = vdb.db.NewUpdateBatch() + // numKeys := 0 + batchSize := 0 + // maxBatchSize := int(vdb.db.GetMaxBatchSize()) + // maxBatchCount := int(vdb.db.GetMaxBatchCount()) + for { + versionedKV, err := itr.Next() + if err != nil { + return err + } + if versionedKV == nil { + break + } + dbKey := kvdb.EncodeDataKey(versionedKV.Namespace, versionedKV.Key) + dbValue, err := kvdb.EncodeValue(versionedKV.VersionedValue) + if err != nil { + return err + } + batchSize += len(dbKey) + len(dbValue) + // numKeys++ + dbBatch.Put(dbKey, dbValue) + if batchSize >= maxDataImportBatchSize { + if err := vdb.db.WriteBatch(dbBatch, true); err != nil { + return err + } + batchSize = 0 + // numKeys = 0 + dbBatch.Cancel() + dbBatch = vdb.db.NewUpdateBatch() + } + } + dbBatch.Put(savePointKey, savepoint.ToBytes()) + return vdb.db.WriteBatch(dbBatch, true) +} + +// IsEmpty return true if the statedb does not have any content +func (vdb *versionedDB) IsEmpty() (bool, error) { + return vdb.db.IsEmpty() +} + +type kvScanner struct { + namespace string + dbItr *badgerdbhelper.Iterator + requestedLimit int32 + totalRecordsReturned int32 +} + +func newKVScanner(namespace string, dbItr *badgerdbhelper.Iterator, requestedLimit int32) *kvScanner { + return &kvScanner{namespace, dbItr, requestedLimit, 0} +} + +func (scanner *kvScanner) Next() (*statedb.VersionedKV, error) { + if scanner.requestedLimit > 0 && scanner.totalRecordsReturned >= scanner.requestedLimit { + return nil, nil + } + if !scanner.dbItr.Next() { + return nil, nil + } + + dbKey := scanner.dbItr.Key() + dbVal := scanner.dbItr.Value() + dbValCopy := make([]byte, len(dbVal)) + copy(dbValCopy, dbVal) + _, key := kvdb.DecodeDataKey(dbKey) + vv, err := kvdb.DecodeValue(dbValCopy) + if err != nil { + return nil, err + } + + scanner.totalRecordsReturned++ + return &statedb.VersionedKV{ + CompositeKey: &statedb.CompositeKey{ + Namespace: scanner.namespace, + Key: key, + }, + VersionedValue: vv, + }, nil +} + +func (scanner *kvScanner) Close() { + scanner.dbItr.Release() +} + +func (scanner *kvScanner) GetBookmarkAndClose() string { + retval := "" + if scanner.dbItr.Next() { + dbKey := scanner.dbItr.Key() + _, key := kvdb.DecodeDataKey(dbKey) + retval = key + } + scanner.Close() + return retval +} + +type fullDBScanner struct { + db *badgerdbhelper.DBHandle + dbItr *badgerdbhelper.Iterator + toSkip func(namespace string) bool + isClosed bool +} + +func newFullDBScanner(db *badgerdbhelper.DBHandle, skipNamespace func(namespace string) bool) (*fullDBScanner, error) { + dbItr, err := db.GetIterator(dataKeyPrefix, dataKeyStopper) + if err != nil { + return nil, err + } + return &fullDBScanner{ + db: db, + dbItr: dbItr, + toSkip: skipNamespace, + isClosed: false, + }, + nil +} + +// Next returns the key-values in the lexical order of +func (s *fullDBScanner) Next() (*statedb.VersionedKV, error) { + if s.isClosed { + return nil, errors.Errorf("internal badgerdb error while retrieving data from db iterator") + } + for s.dbItr.Next() { + ns, key := kvdb.DecodeDataKey(s.dbItr.Key()) + compositeKey := &statedb.CompositeKey{ + Namespace: ns, + Key: key, + } + + versionedVal, err := kvdb.DecodeValue(s.dbItr.Value()) + if err != nil { + return nil, err + } + + switch { + case !s.toSkip(ns): + return &statedb.VersionedKV{ + CompositeKey: compositeKey, + VersionedValue: versionedVal, + }, nil + default: + s.dbItr.Seek(kvdb.DataKeyStarterForNextNamespace(ns)) + s.dbItr.IgnoreNext = true + } + } + return nil, nil +} + +func (s *fullDBScanner) Close() { + if s == nil { + return + } + s.isClosed = true + s.dbItr.Release() +} diff --git a/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb_test.go b/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb_test.go new file mode 100644 index 00000000000..40bbe829ac7 --- /dev/null +++ b/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb_test.go @@ -0,0 +1,283 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package statebadgerdb + +import ( + "errors" + "testing" + + "github.com/hyperledger/fabric/core/ledger/internal/version" + "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb" + "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb/commontests" + kvdb "github.com/hyperledger/fabric/core/ledger/kvledger/txmgmt/statedb/stateleveldb" + "github.com/stretchr/testify/require" +) + +func TestBasicRW(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestBasicRW(t, env.DBProvider) +} + +func TestMultiDBBasicRW(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestMultiDBBasicRW(t, env.DBProvider) +} + +func TestDeletes(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestDeletes(t, env.DBProvider) +} + +func TestIterator(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestIterator(t, env.DBProvider) + t.Run("test-iter-error-path", func(t *testing.T) { + db, err := env.DBProvider.GetDBHandle("testiterator", nil) + require.NoError(t, err) + env.DBProvider.Close() + itr, err := db.GetStateRangeScanIterator("ns1", "", "") + require.EqualError(t, err, "internal badgerdb error while obtaining db iterator: badgerdb: closed") + require.Nil(t, itr) + }) +} + +func TestDataKeyEncoding(t *testing.T) { + testDataKeyEncoding(t, "ledger1", "ns", "key") + testDataKeyEncoding(t, "ledger2", "ns", "") +} + +func testDataKeyEncoding(t *testing.T, _ string, ns string, key string) { + dataKey := kvdb.EncodeDataKey(ns, key) + t.Logf("dataKey=%#v", dataKey) + ns1, key1 := kvdb.DecodeDataKey(dataKey) + require.Equal(t, ns, ns1) + require.Equal(t, key, key1) +} + +// TestQueryOnBadgerDB tests queries on badgerdb. +func TestQueryOnBadgerDB(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + db, err := env.DBProvider.GetDBHandle("testquery", nil) + require.NoError(t, err) + require.NoError(t, db.Open()) + defer db.Close() + batch := statedb.NewUpdateBatch() + jsonValue1 := `{"asset_name": "marble1","color": "blue","size": 1,"owner": "tom"}` + batch.Put("ns1", "key1", []byte(jsonValue1), version.NewHeight(1, 1)) + + savePoint := version.NewHeight(2, 22) + require.NoError(t, db.ApplyUpdates(batch, savePoint)) + + // query for owner=jerry, use namespace "ns1" + // As queries are not supported in badgerdb out of the box, + // call to ExecuteQuery() should return a error message. + // But actually queries are possible with usage the tool + // https://github.com/timshannon/badgerhold + itr, err := db.ExecuteQuery("ns1", `{"selector":{"owner":"jerry"}}`) + require.Error(t, err, "ExecuteQuery not supported for badgerdb") + require.Nil(t, itr) +} + +func TestGetStateMultipleKeys(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestGetStateMultipleKeys(t, env.DBProvider) +} + +func TestGetVersion(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestGetVersion(t, env.DBProvider) +} + +func TestUtilityFunctions(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + + db, err := env.DBProvider.GetDBHandle("testutilityfunctions", nil) + require.NoError(t, err) + + require.True(t, env.DBProvider.BytesKeySupported()) + require.True(t, db.BytesKeySupported()) + + // ValidateKeyValue should return nil for a valid key and value + require.NoError(t, db.ValidateKeyValue("testKey", []byte("testValue")), "badgerdb should accept all key-values") +} + +func TestValueAndMetadataWrites(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestValueAndMetadataWrites(t, env.DBProvider) +} + +func TestPaginatedRangeQuery(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestPaginatedRangeQuery(t, env.DBProvider) +} + +func TestRangeQuerySpecialCharacters(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestRangeQuerySpecialCharacters(t, env.DBProvider) +} + +func TestApplyUpdatesWithNilHeight(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestApplyUpdatesWithNilHeight(t, env.DBProvider) +} + +func TestDataExportImport(t *testing.T) { + // smaller batch size for testing to cover the boundary case of writing the final batch + maxDataImportBatchSize = 10 + env := NewTestVDBEnv(t) + defer env.Cleanup() + commontests.TestDataExportImport( + t, + env.DBProvider, + ) +} + +func TestFullScanIteratorErrorPropagation(t *testing.T) { + var env *TestVDBEnv + var cleanup func() + var vdbProvider *VersionedDBProvider + var vdb *versionedDB + + initEnv := func() { + env = NewTestVDBEnv(t) + vdbProvider = env.DBProvider + db, err := vdbProvider.GetDBHandle("TestFullScanIteratorErrorPropagation", nil) + require.NoError(t, err) + vdb = db.(*versionedDB) + cleanup = func() { + env.Cleanup() + } + } + + reInitEnv := func() { + env.Cleanup() + initEnv() + } + + initEnv() + defer cleanup() + + // error from function GetFullScanIterator + vdbProvider.Close() + _, err := vdb.GetFullScanIterator( + func(string) bool { + return false + }, + ) + require.Contains(t, err.Error(), "internal badgerdb error while obtaining db iterator:") + + // error from function Next + reInitEnv() + itr, err := vdb.GetFullScanIterator( + func(string) bool { + return false + }, + ) + require.NoError(t, err) + itr.Close() + _, err = itr.Next() + require.Contains(t, err.Error(), "internal badgerdb error while retrieving data from db iterator") +} + +func TestImportStateErrorPropagation(t *testing.T) { + var env *TestVDBEnv + var cleanup func() + var vdbProvider *VersionedDBProvider + + initEnv := func() { + env = NewTestVDBEnv(t) + vdbProvider = env.DBProvider + cleanup = func() { + env.Cleanup() + } + } + + t.Run("error-reading-from-source", func(t *testing.T) { + initEnv() + defer cleanup() + + err := vdbProvider.ImportFromSnapshot( + "test-db", + version.NewHeight(2, 2), + &dummyFullScanIter{ + err: errors.New("error while reading from source"), + }, + ) + + require.EqualError(t, err, "error while reading from source") + }) + + t.Run("error-writing-to-db", func(t *testing.T) { + initEnv() + defer cleanup() + + vdbProvider.Close() + err := vdbProvider.ImportFromSnapshot("test-db", version.NewHeight(2, 2), + &dummyFullScanIter{ + kv: &statedb.VersionedKV{ + CompositeKey: &statedb.CompositeKey{ + Namespace: "ns", + Key: "key", + }, + VersionedValue: &statedb.VersionedValue{ + Value: []byte("value"), + Version: version.NewHeight(1, 1), + }, + }, + }, + ) + require.Contains(t, err.Error(), "error writing batch to badgerdb") + }) +} + +func TestDrop(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + + checkDBsAfterDropFunc := func(channelName string) { + empty, err := env.DBProvider.dbProvider.GetDBHandle(channelName).IsEmpty() + require.NoError(t, err) + require.True(t, empty) + } + + commontests.TestDrop(t, env.DBProvider, checkDBsAfterDropFunc) +} + +func TestDropErrorPath(t *testing.T) { + env := NewTestVDBEnv(t) + defer env.Cleanup() + + _, err := env.DBProvider.GetDBHandle("testdroperror", nil) + require.NoError(t, err) + + env.DBProvider.Close() + require.EqualError(t, env.DBProvider.Drop("testdroperror"), "internal badgerdb error while obtaining db iterator: badgerdb: closed") +} + +type dummyFullScanIter struct { + err error + kv *statedb.VersionedKV +} + +func (d *dummyFullScanIter) Next() (*statedb.VersionedKV, error) { + return d.kv, d.err +} + +func (d *dummyFullScanIter) Close() { +} diff --git a/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb_test_export.go b/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb_test_export.go new file mode 100644 index 00000000000..77ac1cdbf88 --- /dev/null +++ b/core/ledger/kvledger/txmgmt/statedb/statebadgerdb/statebadgerdb_test_export.go @@ -0,0 +1,41 @@ +/* +Copyright IBM Corp. All Rights Reserved. + +SPDX-License-Identifier: Apache-2.0 +*/ + +package statebadgerdb + +import ( + "io/ioutil" + "os" + "testing" + + "github.com/stretchr/testify/require" +) + +// TestVDBEnv provides a badger db backed versioned db for testing +type TestVDBEnv struct { + t testing.TB + DBProvider *VersionedDBProvider + dbPath string +} + +// NewTestVDBEnv instantiates and new badger db backed TestVDB +func NewTestVDBEnv(t testing.TB) *TestVDBEnv { + t.Logf("Creating new TestVDBEnv") + dbPath, err := ioutil.TempDir("", "statebadgerdb") + if err != nil { + t.Fatalf("Failed to create badgerdb directory: %s", err) + } + dbProvider, err := NewVersionedDBProvider(dbPath) + require.NoError(t, err) + return &TestVDBEnv{t, dbProvider, dbPath} +} + +// Cleanup closes the db and removes the db folder +func (env *TestVDBEnv) Cleanup() { + env.t.Logf("Cleaningup TestVDBEnv") + env.DBProvider.Close() + os.RemoveAll(env.dbPath) +} diff --git a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb.go b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb.go index 6865630d1e2..3520aae267a 100644 --- a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb.go +++ b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb.go @@ -114,14 +114,14 @@ func (vdb *versionedDB) BytesKeySupported() bool { // GetState implements method in VersionedDB interface func (vdb *versionedDB) GetState(namespace string, key string) (*statedb.VersionedValue, error) { logger.Debugf("GetState(). ns=%s, key=%s", namespace, key) - dbVal, err := vdb.db.Get(encodeDataKey(namespace, key)) + dbVal, err := vdb.db.Get(EncodeDataKey(namespace, key)) if err != nil { return nil, err } if dbVal == nil { return nil, nil } - return decodeValue(dbVal) + return DecodeValue(dbVal) } // GetVersion implements method in VersionedDB interface @@ -159,8 +159,8 @@ func (vdb *versionedDB) GetStateRangeScanIterator(namespace string, startKey str // GetStateRangeScanIteratorWithPagination implements method in VersionedDB interface func (vdb *versionedDB) GetStateRangeScanIteratorWithPagination(namespace string, startKey string, endKey string, pageSize int32) (statedb.QueryResultsIterator, error) { - dataStartKey := encodeDataKey(namespace, startKey) - dataEndKey := encodeDataKey(namespace, endKey) + dataStartKey := EncodeDataKey(namespace, startKey) + dataEndKey := EncodeDataKey(namespace, endKey) if endKey == "" { dataEndKey[len(dataEndKey)-1] = lastKeyIndicator } @@ -188,13 +188,13 @@ func (vdb *versionedDB) ApplyUpdates(batch *statedb.UpdateBatch, height *version for _, ns := range namespaces { updates := batch.GetUpdates(ns) for k, vv := range updates { - dataKey := encodeDataKey(ns, k) + dataKey := EncodeDataKey(ns, k) logger.Debugf("Channel [%s]: Applying key(string)=[%s] key(bytes)=[%#v]", vdb.dbName, string(dataKey), dataKey) if vv.Value == nil { dbBatch.Delete(dataKey) } else { - encodedVal, err := encodeValue(vv) + encodedVal, err := EncodeValue(vv) if err != nil { return err } @@ -255,8 +255,8 @@ func (vdb *versionedDB) importState(itr statedb.FullScanIterator, savepoint *ver if versionedKV == nil { break } - dbKey := encodeDataKey(versionedKV.Namespace, versionedKV.Key) - dbValue, err := encodeValue(versionedKV.VersionedValue) + dbKey := EncodeDataKey(versionedKV.Namespace, versionedKV.Key) + dbValue, err := EncodeValue(versionedKV.VersionedValue) if err != nil { return err } @@ -279,18 +279,18 @@ func (vdb *versionedDB) IsEmpty() (bool, error) { return vdb.db.IsEmpty() } -func encodeDataKey(ns, key string) []byte { +func EncodeDataKey(ns, key string) []byte { k := append(dataKeyPrefix, []byte(ns)...) k = append(k, nsKeySep...) return append(k, []byte(key)...) } -func decodeDataKey(encodedDataKey []byte) (string, string) { +func DecodeDataKey(encodedDataKey []byte) (string, string) { split := bytes.SplitN(encodedDataKey, nsKeySep, 2) return string(split[0][1:]), string(split[1]) } -func dataKeyStarterForNextNamespace(ns string) []byte { +func DataKeyStarterForNextNamespace(ns string) []byte { k := append(dataKeyPrefix, []byte(ns)...) return append(k, lastKeyIndicator) } @@ -318,8 +318,8 @@ func (scanner *kvScanner) Next() (*statedb.VersionedKV, error) { dbVal := scanner.dbItr.Value() dbValCopy := make([]byte, len(dbVal)) copy(dbValCopy, dbVal) - _, key := decodeDataKey(dbKey) - vv, err := decodeValue(dbValCopy) + _, key := DecodeDataKey(dbKey) + vv, err := DecodeValue(dbValCopy) if err != nil { return nil, err } @@ -342,7 +342,7 @@ func (scanner *kvScanner) GetBookmarkAndClose() string { retval := "" if scanner.dbItr.Next() { dbKey := scanner.dbItr.Key() - _, key := decodeDataKey(dbKey) + _, key := DecodeDataKey(dbKey) retval = key } scanner.Close() @@ -371,13 +371,12 @@ func newFullDBScanner(db *leveldbhelper.DBHandle, skipNamespace func(namespace s // Next returns the key-values in the lexical order of func (s *fullDBScanner) Next() (*statedb.VersionedKV, error) { for s.dbItr.Next() { - ns, key := decodeDataKey(s.dbItr.Key()) + ns, key := DecodeDataKey(s.dbItr.Key()) compositeKey := &statedb.CompositeKey{ Namespace: ns, Key: key, } - - versionedVal, err := decodeValue(s.dbItr.Value()) + versionedVal, err := DecodeValue(s.dbItr.Value()) if err != nil { return nil, err } @@ -389,7 +388,7 @@ func (s *fullDBScanner) Next() (*statedb.VersionedKV, error) { VersionedValue: versionedVal, }, nil default: - s.dbItr.Seek(dataKeyStarterForNextNamespace(ns)) + s.dbItr.Seek(DataKeyStarterForNextNamespace(ns)) s.dbItr.Prev() } } diff --git a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb_test.go b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb_test.go index efe3a94f062..25840780793 100644 --- a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb_test.go +++ b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/stateleveldb_test.go @@ -54,9 +54,9 @@ func TestDataKeyEncoding(t *testing.T) { } func testDataKeyEncoding(t *testing.T, dbName string, ns string, key string) { - dataKey := encodeDataKey(ns, key) + dataKey := EncodeDataKey(ns, key) t.Logf("dataKey=%#v", dataKey) - ns1, key1 := decodeDataKey(dataKey) + ns1, key1 := DecodeDataKey(dataKey) require.Equal(t, ns, ns1) require.Equal(t, key, key1) } diff --git a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding.go b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding.go index 79cfeaba576..36e412ad586 100644 --- a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding.go +++ b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding.go @@ -13,7 +13,7 @@ import ( ) // encodeValue encodes the value, version, and metadata -func encodeValue(v *statedb.VersionedValue) ([]byte, error) { +func EncodeValue(v *statedb.VersionedValue) ([]byte, error) { return proto.Marshal( &DBValue{ Version: v.Version.ToBytes(), @@ -24,7 +24,7 @@ func encodeValue(v *statedb.VersionedValue) ([]byte, error) { } // decodeValue decodes the statedb value bytes -func decodeValue(encodedValue []byte) (*statedb.VersionedValue, error) { +func DecodeValue(encodedValue []byte) (*statedb.VersionedValue, error) { dbValue := &DBValue{} err := proto.Unmarshal(encodedValue, dbValue) if err != nil { diff --git a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding_test.go b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding_test.go index 5fef82084ad..7dd6142a69c 100644 --- a/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding_test.go +++ b/core/ledger/kvledger/txmgmt/statedb/stateleveldb/value_encoding_test.go @@ -45,9 +45,9 @@ func TestEncodeDecodeVersionedValues(t *testing.T) { } func testEncodeDecodeVersionedValues(t *testing.T, v *statedb.VersionedValue) { - encodedVal, err := encodeValue(v) + encodedVal, err := EncodeValue(v) require.NoError(t, err) - decodedVal, err := decodeValue(encodedVal) + decodedVal, err := DecodeValue(encodedVal) require.NoError(t, err) require.Equal(t, v, decodedVal) } diff --git a/core/ledger/kvledger/unjoin_channel.go b/core/ledger/kvledger/unjoin_channel.go index 3cae52b4df6..f81d3153410 100644 --- a/core/ledger/kvledger/unjoin_channel.go +++ b/core/ledger/kvledger/unjoin_channel.go @@ -88,7 +88,7 @@ func removeLedgerData(config *ledger.Config, ledgerID string) error { &noopHealthCheckRegistry{}, &privacyenabledstate.StateDBConfig{ StateDBConfig: config.StateDBConfig, - LevelDBPath: StateDBPath(config.RootFSPath), + StateDBPath: StateDBPath(config.RootFSPath), }, []string{}, ) diff --git a/core/ledger/ledger_interface.go b/core/ledger/ledger_interface.go index ce92d105d4f..6ab169099d5 100644 --- a/core/ledger/ledger_interface.go +++ b/core/ledger/ledger_interface.go @@ -25,6 +25,7 @@ import ( const ( GoLevelDB = "goleveldb" CouchDB = "CouchDB" + Badger = "Badger" ) // Initializer encapsulates dependencies for PeerLedgerProvider diff --git a/go.mod b/go.mod index c06ff76b34f..2bcfee9de0a 100644 --- a/go.mod +++ b/go.mod @@ -44,6 +44,8 @@ require ( gopkg.in/yaml.v2 v2.4.0 ) +require github.com/dgraph-io/badger/v4 v4.2.0 + require ( github.com/AdaLogics/go-fuzz-headers v0.0.0-20230811130428-ced1acdcaa24 // indirect github.com/Azure/go-ansiterm v0.0.0-20210617225240-d185dfc1b5a1 // indirect @@ -61,6 +63,7 @@ require ( github.com/consensys/bavard v0.1.13 // indirect github.com/consensys/gnark-crypto v0.12.1 // indirect github.com/containerd/containerd v1.6.26 // indirect + github.com/dgraph-io/ristretto v0.1.1 // indirect github.com/docker/docker v24.0.9+incompatible // indirect github.com/docker/go-connections v0.4.0 // indirect github.com/docker/go-units v0.5.0 // indirect @@ -71,7 +74,10 @@ require ( github.com/go-logr/logr v1.4.1 // indirect github.com/go-task/slim-sprig v0.0.0-20230315185526-52ccab3ef572 // indirect github.com/gogo/protobuf v1.3.2 // indirect + github.com/golang/glog v1.1.2 // indirect + github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da // indirect github.com/golang/snappy v0.0.4 // indirect + github.com/google/flatbuffers v1.12.1 // indirect github.com/google/go-cmp v0.6.0 // indirect github.com/google/pprof v0.0.0-20240130152714-0ed6a68c8d9e // indirect github.com/hashicorp/hcl v1.0.0 // indirect @@ -105,6 +111,7 @@ require ( github.com/subosito/gotenv v1.6.0 // indirect github.com/sykesm/zap-logfmt v0.0.4 // indirect go.etcd.io/etcd/pkg/v3 v3.5.9 // indirect + go.opencensus.io v0.24.0 // indirect go.uber.org/multierr v1.11.0 // indirect golang.org/x/crypto v0.21.0 // indirect golang.org/x/mod v0.14.0 // indirect diff --git a/go.sum b/go.sum index 881d4577413..02550782f46 100644 --- a/go.sum +++ b/go.sum @@ -144,7 +144,13 @@ github.com/creack/pty v1.1.11/go.mod h1:oKZEueFk5CKHvIhNR5MUki03XCEU+Q6VDXinZuGJ github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38= github.com/davecgh/go-spew v1.1.1 h1:vj9j/u1bqnvCEfJOwUhtlOARqs3+rkHYY13jYWTU97c= github.com/davecgh/go-spew v1.1.1/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38= +github.com/dgraph-io/badger/v4 v4.2.0 h1:kJrlajbXXL9DFTNuhhu9yCx7JJa4qpYWxtE8BzuWsEs= +github.com/dgraph-io/badger/v4 v4.2.0/go.mod h1:qfCqhPoWDFJRx1gp5QwwyGo8xk1lbHUxvK9nK0OGAak= +github.com/dgraph-io/ristretto v0.1.1 h1:6CWw5tJNgpegArSHpNHJKldNeq03FQCwYvfMVWajOK8= +github.com/dgraph-io/ristretto v0.1.1/go.mod h1:S1GPSBCYCIhmVNfcth17y2zZtQT6wzkzgwUve0VDWWA= github.com/dgrijalva/jwt-go v3.2.0+incompatible/go.mod h1:E3ru+11k8xSBh+hMPgOLZmtrrCbhqsmaPHjLKYnJCaQ= +github.com/dgryski/go-farm v0.0.0-20190423205320-6a90982ecee2 h1:tdlZCpZ/P9DhczCTSixgIKmwPv6+wP5DGjqLYw5SUiA= +github.com/dgryski/go-farm v0.0.0-20190423205320-6a90982ecee2/go.mod h1:SqUrOPUnsFjfmXRMNPybcSiG0BgUW2AuFH8PAnS2iTw= github.com/dgryski/go-sip13 v0.0.0-20181026042036-e10d5fee7954/go.mod h1:vAd38F8PWV+bWy6jNmig1y/TA+kYO4g3RSRF0IAv0no= github.com/docker/docker v24.0.9+incompatible h1:HPGzNmwfLZWdxHqK9/II92pyi1EpYKsAqcl4G0Of9v0= github.com/docker/docker v24.0.9+incompatible/go.mod h1:eEKB0N0r5NX/I1kEveEz05bcu8tLC/8azJZsviup8Sk= @@ -209,11 +215,15 @@ github.com/gogo/protobuf v1.2.1/go.mod h1:hp+jE20tsWTFYpLwKvXlhS1hjn+gTNwPg2I6zV github.com/gogo/protobuf v1.3.2 h1:Ov1cvc58UF3b5XjBnZv7+opcTcQFZebYjWzi34vdm4Q= github.com/gogo/protobuf v1.3.2/go.mod h1:P1XiOD3dCwIKUDQYPy72D8LYyHL2YPYrpS2s69NZV8Q= github.com/golang/glog v0.0.0-20160126235308-23def4e6c14b/go.mod h1:SBH7ygxi8pfUlaOkMMuAQtPIUF8ecWP5IEl/CR7VP2Q= +github.com/golang/glog v1.1.2 h1:DVjP2PbBOzHyzA+dn3WhHIq4NdVu3Q+pvivFICf/7fo= +github.com/golang/glog v1.1.2/go.mod h1:zR+okUeTbrL6EL3xHUDxZuEtGv04p5shwip1+mL/rLQ= github.com/golang/groupcache v0.0.0-20160516000752-02826c3e7903/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc= github.com/golang/groupcache v0.0.0-20190129154638-5b532d6fd5ef/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc= github.com/golang/groupcache v0.0.0-20190702054246-869f871628b6/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc= github.com/golang/groupcache v0.0.0-20191227052852-215e87163ea7/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc= github.com/golang/groupcache v0.0.0-20200121045136-8c9f03a8e57e/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc= +github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da h1:oI5xCqsCo564l8iNU+DwB5epxmsaqB+rhGL0m5jtYqE= +github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da/go.mod h1:cIg4eruTrX1D+g88fzRXU5OdNfaM+9IcxsU14FzY7Hc= github.com/golang/mock v1.1.1/go.mod h1:oTYuIxOrZwtPieC+H1uAHpcLFnEyAGVDL/k47Jfbm0A= github.com/golang/mock v1.2.0/go.mod h1:oTYuIxOrZwtPieC+H1uAHpcLFnEyAGVDL/k47Jfbm0A= github.com/golang/mock v1.3.1/go.mod h1:sBzyDLLjw3U8JLTeZvSv8jJB+tU5PVekmnlKIyFUx0Y= @@ -245,6 +255,8 @@ github.com/golang/snappy v0.0.4 h1:yAGX7huGHXlcLOEtBnF4w7FQwA26wojNCwOYAEhLjQM= github.com/golang/snappy v0.0.4/go.mod h1:/XxbfmMg8lxefKM7IXC3fBNl/7bRcc72aCRzEWrmP2Q= github.com/google/btree v0.0.0-20180813153112-4030bb1f1f0c/go.mod h1:lNA+9X1NB3Zf8V7Ke586lFgjr2dZNuvo3lPJSGZ5JPQ= github.com/google/btree v1.0.0/go.mod h1:lNA+9X1NB3Zf8V7Ke586lFgjr2dZNuvo3lPJSGZ5JPQ= +github.com/google/flatbuffers v1.12.1 h1:MVlul7pQNoDzWRLTw5imwYsl+usrS1TXG2H4jg6ImGw= +github.com/google/flatbuffers v1.12.1/go.mod h1:1AeVuKshWv4vARoZatz6mlQ0JxURH0Kv5+zNeJKJCa8= github.com/google/go-cmp v0.2.0/go.mod h1:oXzfMopK8JAjlY9xF4vHSVASa0yLyX7SntLO5aqRK0M= github.com/google/go-cmp v0.3.0/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU= github.com/google/go-cmp v0.3.1/go.mod h1:8QqcDgzrUqlUb/G2PQTWiueGozuR1884gddMywk6iLU= @@ -252,6 +264,7 @@ github.com/google/go-cmp v0.4.0/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/ github.com/google/go-cmp v0.4.1/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= github.com/google/go-cmp v0.5.0/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= github.com/google/go-cmp v0.5.1/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= +github.com/google/go-cmp v0.5.3/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= github.com/google/go-cmp v0.5.4/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= github.com/google/go-cmp v0.5.5/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= github.com/google/go-cmp v0.5.6/go.mod h1:v8dTdLbMG2kIc/vJvl+f65V22dbkXbowE6jgT/gNBxE= @@ -568,6 +581,7 @@ github.com/stretchr/testify v1.6.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/ github.com/stretchr/testify v1.7.0/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg= github.com/stretchr/testify v1.7.1/go.mod h1:6Fq8oRcR53rry900zMqJjRRixrwX3KX962/h/Wwjteg= github.com/stretchr/testify v1.8.0/go.mod h1:yNjHg4UonilssWZ8iaSj1OCr/vHnekPRkoO+kdMU+MU= +github.com/stretchr/testify v1.8.1/go.mod h1:w2LPCIKwWwSfY2zedu0+kehJoqGctiVI29o6fzry7u4= github.com/stretchr/testify v1.8.4 h1:CcVxjf3Q8PM0mHUKJCdn+eZZtm5yQwehR5yeSVQQcUk= github.com/stretchr/testify v1.8.4/go.mod h1:sz/lmYIOXD/1dqDmKjjqLyZ2RngseejIcXlSw2iwfAo= github.com/subosito/gotenv v1.2.0/go.mod h1:N0PQaV/YGNqwC0u51sEeR/aUtSLEXKX9iv69rRypqCw= @@ -606,6 +620,8 @@ go.opencensus.io v0.22.0/go.mod h1:+kGneAE2xo2IficOXnaByMWTGM9T73dGwxeWcUqIpI8= go.opencensus.io v0.22.2/go.mod h1:yxeiOL68Rb0Xd1ddK5vPZ/oVn4vY4Ynel7k9FzqtOIw= go.opencensus.io v0.22.3/go.mod h1:yxeiOL68Rb0Xd1ddK5vPZ/oVn4vY4Ynel7k9FzqtOIw= go.opencensus.io v0.22.4/go.mod h1:yxeiOL68Rb0Xd1ddK5vPZ/oVn4vY4Ynel7k9FzqtOIw= +go.opencensus.io v0.24.0 h1:y73uSU6J157QMP2kn2r30vwW1A2W2WFwSCGnAVxeaD0= +go.opencensus.io v0.24.0/go.mod h1:vNK8G9p7aAivkbmorf4v+7Hgx+Zs0yY+0fOtgBfjQKo= go.opentelemetry.io/proto/otlp v0.7.0/go.mod h1:PqfVotwruBrMGOCsRd/89rSnXhoiJIqeYNgFYFoEGnI= go.uber.org/atomic v1.3.2/go.mod h1:gD2HeocX3+yG+ygLZcrzQJaqmWj9AIm7n08wl/qW/PE= go.uber.org/atomic v1.4.0/go.mod h1:gD2HeocX3+yG+ygLZcrzQJaqmWj9AIm7n08wl/qW/PE= @@ -701,6 +717,7 @@ golang.org/x/net v0.0.0-20200707034311-ab3426394381/go.mod h1:/O7V0waA8r7cgGh81R golang.org/x/net v0.0.0-20200813134508-3edf25e44fcc/go.mod h1:/O7V0waA8r7cgGh81Ro3o1hOxt32SMVPicZroKQ2sZA= golang.org/x/net v0.0.0-20200822124328-c89045814202/go.mod h1:/O7V0waA8r7cgGh81Ro3o1hOxt32SMVPicZroKQ2sZA= golang.org/x/net v0.0.0-20201021035429-f5854403a974/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU= +golang.org/x/net v0.0.0-20201110031124-69a78807bb2b/go.mod h1:sp8m0HH+o8qH0wwXwYZr8TS3Oi6o0r6Gce1SSxlDquU= golang.org/x/net v0.0.0-20210525063256-abc453219eb5/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y= golang.org/x/net v0.23.0 h1:7EYJ93RZ9vYSZAIb2x3lnuvqO5zneoD6IvWjuhfxjTs= golang.org/x/net v0.23.0/go.mod h1:JKghWKKOSdJwpW2GEx0Ja7fmaKnMsbu+MWVZTokSYmg= @@ -777,6 +794,7 @@ golang.org/x/sys v0.0.0-20210603081109-ebe580a85c40/go.mod h1:oPkhp1MJrh7nUepCBc golang.org/x/sys v0.0.0-20210616094352-59db8d763f22/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg= golang.org/x/sys v0.0.0-20220204135822-1c1b9b1eba6a/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg= golang.org/x/sys v0.0.0-20220715151400-c0bba94af5f8/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg= +golang.org/x/sys v0.0.0-20221010170243-090e33056c14/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg= golang.org/x/sys v0.18.0 h1:DBdB3niSjOA/O0blCZBqDefyWNYveAYMNF1Wum0DYQ4= golang.org/x/sys v0.18.0/go.mod h1:/VUhepiaJMQUp4+oa/7Zr1D23ma6VTLIYjOOTFZPUcA= golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo= @@ -927,6 +945,7 @@ google.golang.org/grpc v1.29.1/go.mod h1:itym6AZVZYACWQqET3MqgPpjcuV5QH3BxFS3Iji google.golang.org/grpc v1.30.0/go.mod h1:N36X2cJ7JwdamYAgDz+s+rVMFjt3numwzf/HckM8pak= google.golang.org/grpc v1.31.0/go.mod h1:N36X2cJ7JwdamYAgDz+s+rVMFjt3numwzf/HckM8pak= google.golang.org/grpc v1.33.1/go.mod h1:fr5YgcSWrqhRRxogOsw7RzIpsmvOZ6IcH4kBYTpR3n0= +google.golang.org/grpc v1.33.2/go.mod h1:JMHMWHQWaTccqQQlmk3MJZS+GWXOdAesneDmEnv2fbc= google.golang.org/grpc v1.36.0/go.mod h1:qjiiYl8FncCW8feJPdyg3v6XW24KsRHe+dy9BAGRRjU= google.golang.org/grpc v1.47.0/go.mod h1:vN9eftEi1UMyUsIF80+uQXhHjbXYbm0uXoFCACuMGWk= google.golang.org/grpc v1.61.0 h1:TOvOcuXn30kRao+gfcvsebNEa5iZIiLkisYEkf7R7o0= diff --git a/sampleconfig/core.yaml b/sampleconfig/core.yaml index 654e7c207ce..6c8f5fe5c19 100644 --- a/sampleconfig/core.yaml +++ b/sampleconfig/core.yaml @@ -670,10 +670,10 @@ ledger: blockchain: state: - # stateDatabase - options are "goleveldb", "CouchDB" + # stateDatabase - options are "goleveldb", "CouchDB", "Badger" # goleveldb - default state database stored in goleveldb. # CouchDB - store state database in CouchDB - stateDatabase: goleveldb + stateDatabase: Badger # Limit on the number of records to return per query totalQueryLimit: 100000 couchDBConfig: diff --git a/vendor/github.com/dgraph-io/badger/v4/.deepsource.toml b/vendor/github.com/dgraph-io/badger/v4/.deepsource.toml new file mode 100644 index 00000000000..266045f0ecb --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/.deepsource.toml @@ -0,0 +1,18 @@ +version = 1 + +test_patterns = [ + 'integration/testgc/**', + '**/*_test.go' +] + +exclude_patterns = [ + +] + +[[analyzers]] +name = 'go' +enabled = true + + + [analyzers.meta] + import_path = 'github.com/dgraph-io/badger' diff --git a/vendor/github.com/dgraph-io/badger/v4/.gitignore b/vendor/github.com/dgraph-io/badger/v4/.gitignore new file mode 100644 index 00000000000..f78c74c4c90 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/.gitignore @@ -0,0 +1,5 @@ +p/ +badger-test*/ +.idea/ + +vendor \ No newline at end of file diff --git a/vendor/github.com/dgraph-io/badger/v4/.go-version b/vendor/github.com/dgraph-io/badger/v4/.go-version new file mode 100644 index 00000000000..bc4493477ae --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/.go-version @@ -0,0 +1 @@ +1.19 diff --git a/vendor/github.com/dgraph-io/badger/v4/.golangci.yml b/vendor/github.com/dgraph-io/badger/v4/.golangci.yml new file mode 100644 index 00000000000..a338bab64af --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/.golangci.yml @@ -0,0 +1,32 @@ +run: + skip-dirs: + skip-files: + +linters-settings: + lll: + line-length: 120 + staticcheck: + checks: + - all + - '-SA1019' # it is okay to use math/rand at times. + gosec: + excludes: # these are not relevant for us right now + - G114 + - G204 + - G306 + - G404 + +linters: + disable-all: true + enable: + - errcheck + - gofmt + - goimports + - gosec + - gosimple + - govet + - ineffassign + - lll + - staticcheck + - unconvert + - unused diff --git a/vendor/github.com/dgraph-io/badger/v4/CHANGELOG.md b/vendor/github.com/dgraph-io/badger/v4/CHANGELOG.md new file mode 100644 index 00000000000..a2ebd9b0ba7 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/CHANGELOG.md @@ -0,0 +1,796 @@ +# Changelog +All notable changes to this project will be documented in this file. + +The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/). + +## [4.2.0] - 2023-08-03 + +### Breaking + +- feat(metrics): fix and update metrics in badger (#1948) +- fix(metrics): remove badger version in the metrics name (#1982) + +### Fixed + +- fix(logging): fix direct access to logger (#1980) +- fix(sec): bump google.golang.org/grpc from 1.20.1 to 1.53.0 (#1977) +- fix(sec): update gopkg.in/yaml.v2 package (#1969) +- fix(test): fix flakiness of TestPersistLFDiscardStats (#1963) +- fix(stream): setup oracle correctly in stream writer (#1968) (#1904) +- fix(stream): improve incremental stream writer (#1901) +- fix(test): improve the params in BenchmarkDbGrowth (#1967) +- fix(sync): sync active memtable and value log on Db.Sync (#1847) (#1953) +- fix(test): handle draining of closed channel, speed up test. (#1957) +- fix(test): fix table checksum test. Test on uncompressed. (#1952) +- fix(level): change split key range right key to use ts=0 (#1932) +- fix(test): the new test case PagebufferReader5 introduced an error. (#1936) +- fix(test): add missing unlock in TestPersistLFDiscardStats (#1924) +- fix(PageBufferReader): should conform to io.Reader interface (#1935) +- fix(publisher): publish updates after persistence in WAL (#1917) + +### CI + +- chore(ci): split off coverage workflow (#1944) +- chore(ci): adding trivy scanning workflow (#1925) + +## [4.1.0] - 2023-03-30 + +This release adds support for incremental stream writer. We also do some cleanup in the docs +and resolve some CI issues for community PR's. We resolve high and medium CVE's and fix +[#1833](https://github.com/dgraph-io/badger/issues/1833). + +### Features +- feat(stream): add support for incremental stream writer (#1722) (#1874) + +### Fixes +- chore: upgrade xxhash from v1.1.0 to v2.1.2 (#1910) (fixes [#1833](https://github.com/dgraph-io/badger/issues/1833)) + +### Security +- chore(deps): bump golang.org/x/net from 0.0.0-20201021035429-f5854403a974 to 0.7.0 (#1885) + +### CVE's + +- [CVE-2021-31525](https://github.com/dgraph-io/badger/security/dependabot/7) +- [CVE-2022-41723](https://github.com/dgraph-io/badger/security/dependabot/4) +- [CVE-2022-27664](https://github.com/dgraph-io/badger/security/dependabot/5) +- [CVE-2021-33194](https://github.com/dgraph-io/badger/security/dependabot/9) +- [CVE-2022-41723](https://github.com/dgraph-io/badger/security/dependabot/13) +- [CVE-2021-33194](https://github.com/dgraph-io/badger/security/dependabot/16) +- [CVE-2021-38561](https://github.com/dgraph-io/badger/security/dependabot/8) + +### Chores + +- fix(docs): update README (#1915) +- cleanup sstable file after tests (#1912) +- chore(ci): add dgraph regression tests (#1908) +- docs: fix the default value in docs (#1909) +- chore: update URL for unsupported manifest version error (#1905) +- docs(README): add raft-badger to projects using badger (#1902) +- sync the docs with README with projects using badger (#1903) +- fix: update code comments for WithNumCompactors (#1900) +- docs: add loggie to projects using badger (#1882) +- chore(memtable): refactor code for memtable flush (#1866) +- resolve coveralls issue for community PR's (#1892, #1894, #1896) + +## [4.0.1] - 2023-02-28 + +We issue a follow up release in order to resolve a bug in subscriber. We also generate updated protobufs for Badger v4. + +### Fixed + - fix(pb): fix generated protos #1888 + - fix(publisher): initialize the atomic variable #1889 + +### Chores + - chore(cd): tag based deployments #1887 + - chore(ci): fail fast when testing #1890 + +## [4.0.0] - 2023-02-27 + +> **Warning** +> The tag v4.0.0 has been retracted due to a bug in publisher. +> Use v4.0.1 (see #1889) + +This release fixes a bug in the maxHeaderSize parameter that could lead +to panics. We introduce an external magic number to keep track of external +dependencies. We bump up the minimum required Go version to 1.19. No changes +were made to the format of data on disk. This is a major release because +we are making a switch to SemVer in order to make it easier for the community +to understand when breaking API and data format changes are made. + +### Fixed +- fix: update maxHeaderSize #1877 +- feat(externalMagic): Introduce external magic number (#1745) #1852 +- fix(bench): bring in benchmark fixes from main #1863 + +### Chores +- upgrade go to 1.19 #1868 +- enable linters (gosimple, govet, lll, unused, staticcheck, errcheck, ineffassign, gofmt) #1871 #1870 #1876 +- remove dependency on io/ioutil #1879 +- various doc and comment fixes #1857 +- moving from CalVer to SemVer + +## [3.2103.5] - 2022-12-15 + +We release Badger CLI tool binaries for amd64 and now arm64. This release does not involve any core code changes to Badger. We add a CD job for building Badger for arm64. + +## [3.2103.4] - 2022-11-04 + +### Fixed + - fix(manifest): fix manifest corruption due to race condition in concurrent compactions (#1756) + +### Chores + - We bring the release branch to parity with main by updating the CI/CD jobs, Readme, Codeowners, PR and issue templates, etc. + +## [3.2103.3] - 2022-10-14 + +### Remarks + + - This is a minor patch release that fixes arm64 related issues. The issues in the `z` package in Ristretto were resolved in Ristretto v0.1.1. + +### Fixed + + - fix(arm64): bump ristretto v0.1.0 --> v0.1.1 (#1806) + +## [3.2103.2] - 2021-10-07 + +### Fixed + + - fix(compact): close vlog after the compaction at L0 has completed (#1752) + - fix(builder): put the upper limit on reallocation (#1748) + - deps: Bump github.com/google/flatbuffers to v1.12.1 (#1746) + - fix(levels): Avoid a deadlock when acquiring read locks in levels (#1744) + - fix(pubsub): avoid deadlock in publisher and subscriber (#1749) (#1751) + +## [3.2103.1] - 2021-07-08 + +### Fixed + - fix(compaction): copy over the file ID when building tables #1713 + - fix: Fix conflict detection for managed DB (#1716) + - fix(pendingWrites): don't skip the pending entries with version=0 (#1721) + +### Features + - feat(zstd): replace datadog's zstd with Klauspost's zstd (#1709) + +## [3.2103.0] - 2021-06-02 + +### Breaking + - Subscribe: Add option to subscribe with holes in prefixes. (#1658) + +### Fixed + - fix(compaction): Remove compaction backoff mechanism (#1686) + - Add a name to mutexes to make them unexported (#1678) + - fix(merge-operator): don't read the deleted keys (#1675) + - fix(discard): close the discard stats file on db close (#1672) + - fix(iterator): fix iterator when data does not exist in read only mode (#1670) + - fix(badger): Do not reuse variable across badger commands (#1624) + - fix(dropPrefix): check properly if the key is present in a table (#1623) + +### Performance + - Opt(Stream): Optimize how we deduce key ranges for iteration (#1687) + - Increase value threshold from 1 KB to 1 MB (#1664) + - opt(DropPrefix): check if there exist some data to drop before dropping prefixes (#1621) + +### Features + - feat(options): allow special handling and checking when creating options from superflag (#1688) + - overwrite default Options from SuperFlag string (#1663) + - Support SinceTs in iterators (#1653) + - feat(info): Add a flag to parse and print DISCARD file (#1662) + - feat(vlog): making vlog threshold dynamic 6ce3b7c (#1635) + - feat(options): add NumGoroutines option for default Stream.numGo (#1656) + - feat(Trie): Working prefix match with holes (#1654) + - feat: add functionality to ban a prefix (#1638) + - feat(compaction): Support Lmax to Lmax compaction (#1615) + +### New APIs +- Badger.DB + - BanNamespace + - BannedNamespaces + - Ranges +- Badger.Options + - FromSuperFlag + - WithNumGoRoutines + - WithNamespaceOffset + - WithVLogPercentile +- Badger.Trie + - AddMatch + - DeleteMatch +- Badger.Table + - StaleDataSize +- Badger.Table.Builder + - AddStaleKey +- Badger.InitDiscardStats + +### Removed APIs +- Badger.DB + - KeySplits +- Badger.Options + - SkipVlog + +### Changed APIs +- Badger.DB + - Subscribe +- Badger.Options + - WithValueThreshold + +## [3.2011.1] - 2021-01-22 + +### Fixed + - Fix(compaction): Set base level correctly after stream (#1651) + - Fix: update ristretto and use filepath (#1652) + - Fix(badger): Do not reuse variable across badger commands (#1650) + - Fix(build): fix 32-bit build (#1646) + - Fix(table): always sync SST to disk (#1645) + +## [3.2011.0] - 2021-01-15 + +This release is not backward compatible with Badger v2.x.x + +### Breaking: + - opt(compactions): Improve compaction performance (#1574) + - Change how Badger handles WAL (#1555) + - feat(index): Use flatbuffers instead of protobuf (#1546) + +### Fixed: + - Fix(GC): Set bits correctly for moved keys (#1619) + - Fix(tableBuilding): reduce scope of valuePointer (#1617) + - Fix(compaction): fix table size estimation on compaction (#1613) + - Fix(OOM): Reuse pb.KVs in Stream (#1609) + - Fix race condition in L0StallMs variable (#1605) + - Fix(stream): Stop produceKVs on error (#1604) + - Fix(skiplist): Remove z.Buffer from skiplist (#1600) + - Fix(readonly): fix the file opening mode (#1592) + - Fix: Disable CompactL0OnClose by default (#1586) + - Fix(compaction): Don't drop data when split overlaps with top tables (#1587) + - Fix(subcompaction): Close builder before throttle.Done (#1582) + - Fix(table): Add onDisk size (#1569) + - Fix(Stream): Only send done markers if told to do so + - Fix(value log GC): Fix a bug which caused value log files to not be GCed. + - Fix segmentation fault when cache sizes are small. (#1552) + - Fix(builder): Too many small tables when compression is enabled (#1549) + - Fix integer overflow error when building for 386 (#1541) + - Fix(writeBatch): Avoid deadlock in commit callback (#1529) + - Fix(db): Handle nil logger (#1534) + - Fix(maxVersion): Use choosekey instead of KeyToList (#1532) + - Fix(Backup/Restore): Keep all versions (#1462) + - Fix(build): Fix nocgo builds. (#1493) + - Fix(cleanup): Avoid truncating in value.Open on error (#1465) + - Fix(compaction): Don't use cache for table compaction (#1467) + - Fix(compaction): Use separate compactors for L0, L1 (#1466) + - Fix(options): Do not implicitly enable cache (#1458) + - Fix(cleanup): Do not close cache before compaction (#1464) + - Fix(replay): Update head for LSM entires also (#1456) + - fix(levels): Cleanup builder resources on building an empty table (#1414) + +### Performance + - perf(GC): Remove move keys (#1539) + - Keep the cheaper parts of the index within table struct. (#1608) + - Opt(stream): Use z.Buffer to stream data (#1606) + - opt(builder): Use z.Allocator for building tables (#1576) + - opt(memory): Use z.Calloc for allocating KVList (#1563) + - opt: Small memory usage optimizations (#1562) + - KeySplits checks tables and memtables when number of splits is small. (#1544) + - perf: Reduce memory usage by better struct packing (#1528) + - perf(tableIterator): Don't do next on NewIterator (#1512) + - Improvements: Manual Memory allocation via Calloc (#1459) + - Various bug fixes: Break up list and run DropAll func (#1439) + - Add a limit to the size of the batches sent over a stream. (#1412) + - Commit does not panic after Finish, instead returns an error (#1396) + - levels: Compaction incorrectly drops some delete markers (#1422) + - Remove vlog file if bootstrap, syncDir or mmap fails (#1434) + +### Features: + - Use opencensus for tracing (#1566) + - Export functions from Key Registry (#1561) + - Allow sizes of block and index caches to be updated. (#1551) + - Add metric for number of tables being compacted (#1554) + - feat(info): Show index and bloom filter size (#1543) + - feat(db): Add db.MaxVersion API (#1526) + - Expose DB options in Badger. (#1521) + - Feature: Add a Calloc based Buffer (#1471) + - Add command to stream contents of DB into another DB. (#1463) + - Expose NumAlloc metrics via expvar (#1470) + - Support fully disabling the bloom filter (#1319) + - Add --enc-key flag in badger info tool (#1441) + +### New APIs +- Badger.DB + - CacheMaxCost (#1551) + - Levels (#1574) + - LevelsToString (#1574) + - Opts (#1521) +- Badger.Options + - WithBaseLevelSize (#1574) + - WithBaseTableSize (#1574) + - WithMemTableSize (#1574) +- Badger.KeyRegistry + - DataKey (#1561) + - LatestDataKey (#1561) + +### Removed APIs +- Badger.Options + - WithKeepL0InMemory (#1555) + - WithLevelOneSize (#1574) + - WithLoadBloomsOnOpen (#1555) + - WithLogRotatesToFlush (#1574) + - WithMaxTableSize (#1574) + - WithTableLoadingMode (#1555) + - WithTruncate (#1555) + - WithValueLogLoadingMode (#1555) + +## [2.2007.4] - 2021-08-25 + +### Fixed + - Fix build on Plan 9 (#1451) (#1508) (#1738) + +### Features + - feat(zstd): backport replacement of DataDog's zstd with Klauspost's zstd (#1736) + +## [2.2007.3] - 2021-07-21 + +### Fixed + - fix(maxVersion): Use choosekey instead of KeyToList (#1532) #1533 + - fix(flatten): Add --num_versions flag (#1518) #1520 + - fix(build): Fix integer overflow on 32-bit architectures #1558 + - fix(pb): avoid protobuf warning due to common filename (#1519) + +### Features + - Add command to stream contents of DB into another DB. (#1486) + +### New APIs + - DB.StreamDB + - DB.MaxVersion + +## [2.2007.2] - 2020-08-31 + +### Fixed + - Compaction: Use separate compactors for L0, L1 (#1466) + - Rework Block and Index cache (#1473) + - Add IsClosed method (#1478) + - Cleanup: Avoid truncating in vlog.Open on error (#1465) + - Cleanup: Do not close cache before compactions (#1464) + +### New APIs +- Badger.DB + - BlockCacheMetrics (#1473) + - IndexCacheMetrics (#1473) +- Badger.Option + - WithBlockCacheSize (#1473) + - WithIndexCacheSize (#1473) + +### Removed APIs [Breaking Changes] +- Badger.DB + - DataCacheMetrics (#1473) + - BfCacheMetrics (#1473) +- Badger.Option + - WithMaxCacheSize (#1473) + - WithMaxBfCacheSize (#1473) + - WithKeepBlockIndicesInCache (#1473) + - WithKeepBlocksInCache (#1473) + +## [2.2007.1] - 2020-08-19 + +### Fixed + - Remove vlog file if bootstrap, syncDir or mmap fails (#1434) + - levels: Compaction incorrectly drops some delete markers (#1422) + - Replay: Update head for LSM entires also (#1456) + +## [2.2007.0] - 2020-08-10 + +### Fixed + - Add a limit to the size of the batches sent over a stream. (#1412) + - Fix Sequence generates duplicate values (#1281) + - Fix race condition in DoesNotHave (#1287) + - Fail fast if cgo is disabled and compression is ZSTD (#1284) + - Proto: make badger/v2 compatible with v1 (#1293) + - Proto: Rename dgraph.badger.v2.pb to badgerpb2 (#1314) + - Handle duplicates in ManagedWriteBatch (#1315) + - Ensure `bitValuePointer` flag is cleared for LSM entry values written to LSM (#1313) + - DropPrefix: Return error on blocked writes (#1329) + - Confirm `badgerMove` entry required before rewrite (#1302) + - Drop move keys when its key prefix is dropped (#1331) + - Iterator: Always add key to txn.reads (#1328) + - Restore: Account for value size as well (#1358) + - Compaction: Expired keys and delete markers are never purged (#1354) + - GC: Consider size of value while rewriting (#1357) + - Force KeepL0InMemory to be true when InMemory is true (#1375) + - Rework DB.DropPrefix (#1381) + - Update head while replaying value log (#1372) + - Avoid panic on multiple closer.Signal calls (#1401) + - Return error if the vlog writes exceeds more than 4GB (#1400) + +### Performance + - Clean up transaction oracle as we go (#1275) + - Use cache for storing block offsets (#1336) + +### Features + - Support disabling conflict detection (#1344) + - Add leveled logging (#1249) + - Support entry version in Write batch (#1310) + - Add Write method to batch write (#1321) + - Support multiple iterators in read-write transactions (#1286) + +### New APIs +- Badger.DB + - NewManagedWriteBatch (#1310) + - DropPrefix (#1381) +- Badger.Option + - WithDetectConflicts (#1344) + - WithKeepBlockIndicesInCache (#1336) + - WithKeepBlocksInCache (#1336) +- Badger.WriteBatch + - DeleteAt (#1310) + - SetEntryAt (#1310) + - Write (#1321) + +### Changes to Default Options + - DefaultOptions: Set KeepL0InMemory to false (#1345) + - Increase default valueThreshold from 32B to 1KB (#1346) + +### Deprecated +- Badger.Option + - WithEventLogging (#1203) + +### Reverts +This sections lists the changes which were reverted because of non-reproducible crashes. +- Compress/Encrypt Blocks in the background (#1227) + + +## [2.0.3] - 2020-03-24 + +### Fixed + +- Add support for watching nil prefix in subscribe API (#1246) + +### Performance + +- Compress/Encrypt Blocks in the background (#1227) +- Disable cache by default (#1257) + +### Features + +- Add BypassDirLock option (#1243) +- Add separate cache for bloomfilters (#1260) + +### New APIs +- badger.DB + - BfCacheMetrics (#1260) + - DataCacheMetrics (#1260) +- badger.Options + - WithBypassLockGuard (#1243) + - WithLoadBloomsOnOpen (#1260) + - WithMaxBfCacheSize (#1260) + +## [2.0.3] - 2020-03-24 + +### Fixed + +- Add support for watching nil prefix in subscribe API (#1246) + +### Performance + +- Compress/Encrypt Blocks in the background (#1227) +- Disable cache by default (#1257) + +### Features + +- Add BypassDirLock option (#1243) +- Add separate cache for bloomfilters (#1260) + +### New APIs +- badger.DB + - BfCacheMetrics (#1260) + - DataCacheMetrics (#1260) +- badger.Options + - WithBypassLockGuard (#1243) + - WithLoadBloomsOnOpen (#1260) + - WithMaxBfCacheSize (#1260) + +## [2.0.2] - 2020-03-02 + +### Fixed + +- Cast sz to uint32 to fix compilation on 32 bit. (#1175) +- Fix checkOverlap in compaction. (#1166) +- Avoid sync in inmemory mode. (#1190) +- Support disabling the cache completely. (#1185) +- Add support for caching bloomfilters. (#1204) +- Fix int overflow for 32bit. (#1216) +- Remove the 'this entry should've caught' log from value.go. (#1170) +- Rework concurrency semantics of valueLog.maxFid. (#1187) + +### Performance + +- Use fastRand instead of locked-rand in skiplist. (#1173) +- Improve write stalling on level 0 and 1. (#1186) +- Disable compression and set ZSTD Compression Level to 1. (#1191) + +## [2.0.1] - 2020-01-02 + +### New APIs + +- badger.Options + - WithInMemory (f5b6321) + - WithZSTDCompressionLevel (3eb4e72) + +- Badger.TableInfo + - EstimatedSz (f46f8ea) + +### Features + +- Introduce in-memory mode in badger. (#1113) + +### Fixed + +- Limit manifest's change set size. (#1119) +- Cast idx to uint32 to fix compilation on i386. (#1118) +- Fix request increment ref bug. (#1121) +- Fix windows dataloss issue. (#1134) +- Fix VerifyValueChecksum checks. (#1138) +- Fix encryption in stream writer. (#1146) +- Fix segmentation fault in vlog.Read. (header.Decode) (#1150) +- Fix merge iterator duplicates issue. (#1157) + +### Performance + +- Set level 15 as default compression level in Zstd. (#1111) +- Optimize createTable in stream_writer.go. (#1132) + +## [2.0.0] - 2019-11-12 + +### New APIs + +- badger.DB + - NewWriteBatchAt (7f43769) + - CacheMetrics (b9056f1) + +- badger.Options + - WithMaxCacheSize (b9056f1) + - WithEventLogging (75c6a44) + - WithBlockSize (1439463) + - WithBloomFalsePositive (1439463) + - WithKeepL0InMemory (ee70ff2) + - WithVerifyValueChecksum (ee70ff2) + - WithCompression (5f3b061) + - WithEncryptionKey (a425b0e) + - WithEncryptionKeyRotationDuration (a425b0e) + - WithChecksumVerificationMode (7b4083d) + +### Features + +- Data cache to speed up lookups and iterations. (#1066) +- Data compression. (#1013) +- Data encryption-at-rest. (#1042) + +### Fixed + +- Fix deadlock when flushing discard stats. (#976) +- Set move key's expiresAt for keys with TTL. (#1006) +- Fix unsafe usage in Decode. (#1097) +- Fix race condition on db.orc.nextTxnTs. (#1101) +- Fix level 0 GC dataloss bug. (#1090) +- Fix deadlock in discard stats. (#1070) +- Support checksum verification for values read from vlog. (#1052) +- Store entire L0 in memory. (#963) +- Fix table.Smallest/Biggest and iterator Prefix bug. (#997) +- Use standard proto functions for Marshal/Unmarshal and Size. (#994) +- Fix boundaries on GC batch size. (#987) +- VlogSize to store correct directory name to expvar.Map. (#956) +- Fix transaction too big issue in restore. (#957) +- Fix race condition in updateDiscardStats. (#973) +- Cast results of len to uint32 to fix compilation in i386 arch. (#961) +- Making the stream writer APIs goroutine-safe. (#959) +- Fix prefix bug in key iterator and allow all versions. (#950) +- Drop discard stats if we can't unmarshal it. (#936) +- Fix race condition in flushDiscardStats function. (#921) +- Ensure rewrite in vlog is within transactional limits. (#911) +- Fix discard stats moved by GC bug. (#929) +- Fix busy-wait loop in Watermark. (#920) + +### Performance + +- Introduce fast merge iterator. (#1080) +- Binary search based table picker. (#983) +- Flush vlog buffer if it grows beyond threshold. (#1067) +- Introduce StreamDone in Stream Writer. (#1061) +- Performance Improvements to block iterator. (#977) +- Prevent unnecessary safecopy in iterator parseKV. (#971) +- Use pointers instead of binary encoding. (#965) +- Reuse block iterator inside table iterator. (#972) +- [breaking/format] Remove vlen from entry header. (#945) +- Replace FarmHash with AESHash for Oracle conflicts. (#952) +- [breaking/format] Optimize Bloom filters. (#940) +- [breaking/format] Use varint for header encoding (without header length). (#935) +- Change file picking strategy in compaction. (#894) +- [breaking/format] Block level changes. (#880) +- [breaking/format] Add key-offset index to the end of SST table. (#881) + + +## [1.6.0] - 2019-07-01 + +This is a release including almost 200 commits, so expect many changes - some of them +not backward compatible. + +Regarding backward compatibility in Badger versions, you might be interested on reading +[VERSIONING.md](VERSIONING.md). + +_Note_: The hashes in parentheses correspond to the commits that impacted the given feature. + +### New APIs + +- badger.DB + - DropPrefix (291295e) + - Flatten (7e41bba) + - KeySplits (4751ef1) + - MaxBatchCount (b65e2a3) + - MaxBatchSize (b65e2a3) + - PrintKeyValueHistogram (fd59907) + - Subscribe (26128a7) + - Sync (851e462) + +- badger.DefaultOptions() and badger.LSMOnlyOptions() (91ce687) + - badger.Options.WithX methods + +- badger.Entry (e9447c9) + - NewEntry + - WithMeta + - WithDiscard + - WithTTL + +- badger.Item + - KeySize (fd59907) + - ValueSize (5242a99) + +- badger.IteratorOptions + - PickTable (7d46029, 49a49e3) + - Prefix (7d46029) + +- badger.Logger (fbb2778) + +- badger.Options + - CompactL0OnClose (7e41bba) + - Logger (3f66663) + - LogRotatesToFlush (2237832) + +- badger.Stream (14cbd89, 3258067) +- badger.StreamWriter (7116e16) +- badger.TableInfo.KeyCount (fd59907) +- badger.TableManifest (2017987) +- badger.Tx.NewKeyIterator (49a49e3) +- badger.WriteBatch (6daccf9, 7e78e80) + +### Modified APIs + +#### Breaking changes: + +- badger.DefaultOptions and badger.LSMOnlyOptions are now functions rather than variables (91ce687) +- badger.Item.Value now receives a function that returns an error (439fd46) +- badger.Txn.Commit doesn't receive any params now (6daccf9) +- badger.DB.Tables now receives a boolean (76b5341) + +#### Not breaking changes: + +- badger.LSMOptions changed values (799c33f) +- badger.DB.NewIterator now allows multiple iterators per RO txn (41d9656) +- badger.Options.TableLoadingMode's new default is options.MemoryMap (6b97bac) + +### Removed APIs + +- badger.ManagedDB (d22c0e8) +- badger.Options.DoNotCompact (7e41bba) +- badger.Txn.SetWithX (e9447c9) + +### Tools: + +- badger bank disect (13db058) +- badger bank test (13db058) --mmap (03870e3) +- badger fill (7e41bba) +- badger flatten (7e41bba) +- badger info --histogram (fd59907) --history --lookup --show-keys --show-meta --with-prefix (09e9b63) --show-internal (fb2eed9) +- badger benchmark read (239041e) +- badger benchmark write (6d3b67d) + +## [1.5.5] - 2019-06-20 + +* Introduce support for Go Modules + +## [1.5.3] - 2018-07-11 +Bug Fixes: +* Fix a panic caused due to item.vptr not copying over vs.Value, when looking + for a move key. + +## [1.5.2] - 2018-06-19 +Bug Fixes: +* Fix the way move key gets generated. +* If a transaction has unclosed, or multiple iterators running simultaneously, + throw a panic. Every iterator must be properly closed. At any point in time, + only one iterator per transaction can be running. This is to avoid bugs in a + transaction data structure which is thread unsafe. + +* *Warning: This change might cause panics in user code. Fix is to properly + close your iterators, and only have one running at a time per transaction.* + +## [1.5.1] - 2018-06-04 +Bug Fixes: +* Fix for infinite yieldItemValue recursion. #503 +* Fix recursive addition of `badgerMove` prefix. https://github.com/dgraph-io/badger/commit/2e3a32f0ccac3066fb4206b28deb39c210c5266f +* Use file size based window size for sampling, instead of fixing it to 10MB. #501 + +Cleanup: +* Clarify comments and documentation. +* Move badger tool one directory level up. + +## [1.5.0] - 2018-05-08 +* Introduce `NumVersionsToKeep` option. This option is used to discard many + versions of the same key, which saves space. +* Add a new `SetWithDiscard` method, which would indicate that all the older + versions of the key are now invalid. Those versions would be discarded during + compactions. +* Value log GC moves are now bound to another keyspace to ensure latest versions + of data are always at the top in LSM tree. +* Introduce `ValueLogMaxEntries` to restrict the number of key-value pairs per + value log file. This helps bound the time it takes to garbage collect one + file. + +## [1.4.0] - 2018-05-04 +* Make mmap-ing of value log optional. +* Run GC multiple times, based on recorded discard statistics. +* Add MergeOperator. +* Force compact L0 on clsoe (#439). +* Add truncate option to warn about data loss (#452). +* Discard key versions during compaction (#464). +* Introduce new `LSMOnlyOptions`, to make Badger act like a typical LSM based DB. + +Bug fix: +* (Temporary) Check max version across all tables in Get (removed in next + release). +* Update commit and read ts while loading from backup. +* Ensure all transaction entries are part of the same value log file. +* On commit, run unlock callbacks before doing writes (#413). +* Wait for goroutines to finish before closing iterators (#421). + +## [1.3.0] - 2017-12-12 +* Add `DB.NextSequence()` method to generate monotonically increasing integer + sequences. +* Add `DB.Size()` method to return the size of LSM and value log files. +* Tweaked mmap code to make Windows 32-bit builds work. +* Tweaked build tags on some files to make iOS builds work. +* Fix `DB.PurgeOlderVersions()` to not violate some constraints. + +## [1.2.0] - 2017-11-30 +* Expose a `Txn.SetEntry()` method to allow setting the key-value pair + and all the metadata at the same time. + +## [1.1.1] - 2017-11-28 +* Fix bug where txn.Get was returing key deleted in same transaction. +* Fix race condition while decrementing reference in oracle. +* Update doneCommit in the callback for CommitAsync. +* Iterator see writes of current txn. + +## [1.1.0] - 2017-11-13 +* Create Badger directory if it does not exist when `badger.Open` is called. +* Added `Item.ValueCopy()` to avoid deadlocks in long-running iterations +* Fixed 64-bit alignment issues to make Badger run on Arm v7 + +## [1.0.1] - 2017-11-06 +* Fix an uint16 overflow when resizing key slice + +[Unreleased]: https://github.com/dgraph-io/badger/compare/v2.2007.2...HEAD +[2.2007.2]: https://github.com/dgraph-io/badger/compare/v2.2007.1...v2.2007.2 +[2.2007.1]: https://github.com/dgraph-io/badger/compare/v2.2007.0...v2.2007.1 +[2.2007.0]: https://github.com/dgraph-io/badger/compare/v2.0.3...v2.2007.0 +[2.0.3]: https://github.com/dgraph-io/badger/compare/v2.0.2...v2.0.3 +[2.0.2]: https://github.com/dgraph-io/badger/compare/v2.0.1...v2.0.2 +[2.0.1]: https://github.com/dgraph-io/badger/compare/v2.0.0...v2.0.1 +[2.0.0]: https://github.com/dgraph-io/badger/compare/v1.6.0...v2.0.0 +[1.6.0]: https://github.com/dgraph-io/badger/compare/v1.5.5...v1.6.0 +[1.5.5]: https://github.com/dgraph-io/badger/compare/v1.5.3...v1.5.5 +[1.5.3]: https://github.com/dgraph-io/badger/compare/v1.5.2...v1.5.3 +[1.5.2]: https://github.com/dgraph-io/badger/compare/v1.5.1...v1.5.2 +[1.5.1]: https://github.com/dgraph-io/badger/compare/v1.5.0...v1.5.1 +[1.5.0]: https://github.com/dgraph-io/badger/compare/v1.4.0...v1.5.0 +[1.4.0]: https://github.com/dgraph-io/badger/compare/v1.3.0...v1.4.0 +[1.3.0]: https://github.com/dgraph-io/badger/compare/v1.2.0...v1.3.0 +[1.2.0]: https://github.com/dgraph-io/badger/compare/v1.1.1...v1.2.0 +[1.1.1]: https://github.com/dgraph-io/badger/compare/v1.1.0...v1.1.1 +[1.1.0]: https://github.com/dgraph-io/badger/compare/v1.0.1...v1.1.0 +[1.0.1]: https://github.com/dgraph-io/badger/compare/v1.0.0...v1.0.1 diff --git a/vendor/github.com/dgraph-io/badger/v4/CODE_OF_CONDUCT.md b/vendor/github.com/dgraph-io/badger/v4/CODE_OF_CONDUCT.md new file mode 100644 index 00000000000..bf7bbc29dc4 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/CODE_OF_CONDUCT.md @@ -0,0 +1,5 @@ +# Code of Conduct + +Our Code of Conduct can be found here: + +https://dgraph.io/conduct diff --git a/vendor/github.com/dgraph-io/badger/v4/CONTRIBUTING.md b/vendor/github.com/dgraph-io/badger/v4/CONTRIBUTING.md new file mode 100644 index 00000000000..30512e9dba1 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/CONTRIBUTING.md @@ -0,0 +1,107 @@ +# Contribution Guide + +* [Before you get started](#before-you-get-started) + * [Code of Conduct](#code-of-conduct) +* [Your First Contribution](#your-first-contribution) + * [Find a good first topic](#find-a-good-first-topic) +* [Setting up your development environment](#setting-up-your-development-environment) + * [Fork the project](#fork-the-project) + * [Clone the project](#clone-the-project) + * [New branch for a new code](#new-branch-for-a-new-code) + * [Test](#test) + * [Commit and push](#commit-and-push) + * [Create a Pull Request](#create-a-pull-request) + * [Sign the CLA](#sign-the-cla) + * [Get a code review](#get-a-code-review) + +## Before you get started + +### Code of Conduct + +Please make sure to read and observe our [Code of Conduct](./CODE_OF_CONDUCT.md). + +## Your First Contribution + +### Find a good first topic + +You can start by finding an existing issue with the +[good first issue](https://github.com/dgraph-io/badger/labels/good%20first%20issue) or [help wanted](https://github.com/dgraph-io/badger/labels/help%20wanted) labels. These issues are well suited for new contributors. + + +## Setting up your development environment + +Badger uses [`Go Modules`](https://github.com/golang/go/wiki/Modules) +to manage dependencies. The version of Go should be **1.12** or above. + +### Fork the project + +- Visit https://github.com/dgraph-io/badger +- Click the `Fork` button (top right) to create a fork of the repository + +### Clone the project + +```sh +$ git clone https://github.com/$GITHUB_USER/badger +$ cd badger +$ git remote add upstream git@github.com:dgraph-io/badger.git + +# Never push to the upstream master +git remote set-url --push upstream no_push +``` + +### New branch for a new code + +Get your local master up to date: + +```sh +$ git fetch upstream +$ git checkout master +$ git rebase upstream/master +``` + +Create a new branch from the master: + +```sh +$ git checkout -b my_new_feature +``` + +And now you can finally add your changes to project. + +### Test + +Build and run all tests: + +```sh +$ ./test.sh +``` + +### Commit and push + +Commit your changes: + +```sh +$ git commit +``` + +When the changes are ready to review: + +```sh +$ git push origin my_new_feature +``` + +### Create a Pull Request + +Just open `https://github.com/$GITHUB_USER/badger/pull/new/my_new_feature` and +fill the PR description. + +### Sign the CLA + +Click the **Sign in with Github to agree** button to sign the CLA. [An example](https://cla-assistant.io/dgraph-io/badger?pullRequest=1377). + +### Get a code review + +If your pull request (PR) is opened, it will be assigned to one or more +reviewers. Those reviewers will do a code review. + +To address review comments, you should commit the changes to the same branch of +the PR on your fork. diff --git a/vendor/github.com/dgraph-io/badger/v4/LICENSE b/vendor/github.com/dgraph-io/badger/v4/LICENSE new file mode 100644 index 00000000000..d9a10c0d8e8 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/LICENSE @@ -0,0 +1,176 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS diff --git a/vendor/github.com/dgraph-io/badger/v4/Makefile b/vendor/github.com/dgraph-io/badger/v4/Makefile new file mode 100644 index 00000000000..fa4227816d9 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/Makefile @@ -0,0 +1,59 @@ +# +# Copyright 2022 Dgraph Labs, Inc. and Contributors +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. +# + +USER_ID = $(shell id -u) +HAS_JEMALLOC = $(shell test -f /usr/local/lib/libjemalloc.a && echo "jemalloc") +JEMALLOC_URL = "https://github.com/jemalloc/jemalloc/releases/download/5.2.1/jemalloc-5.2.1.tar.bz2" + + +.PHONY: all badger test jemalloc dependency + +badger: jemalloc + @echo "Compiling Badger binary..." + @$(MAKE) -C badger badger + @echo "Badger binary located in badger directory." + +test: jemalloc + @echo "Running Badger tests..." + @./test.sh + +jemalloc: + @if [ -z "$(HAS_JEMALLOC)" ] ; then \ + mkdir -p /tmp/jemalloc-temp && cd /tmp/jemalloc-temp ; \ + echo "Downloading jemalloc..." ; \ + curl -s -L ${JEMALLOC_URL} -o jemalloc.tar.bz2 ; \ + tar xjf ./jemalloc.tar.bz2 ; \ + cd jemalloc-5.2.1 ; \ + ./configure --with-jemalloc-prefix='je_' --with-malloc-conf='background_thread:true,metadata_thp:auto'; \ + make ; \ + if [ "$(USER_ID)" -eq "0" ]; then \ + make install ; \ + else \ + echo "==== Need sudo access to install jemalloc" ; \ + sudo make install ; \ + fi \ + fi + +dependency: + @echo "Installing dependencies..." + @sudo apt-get update + @sudo apt-get -y install \ + ca-certificates \ + curl \ + gnupg \ + lsb-release \ + build-essential \ + protobuf-compiler \ diff --git a/vendor/github.com/dgraph-io/badger/v4/README.md b/vendor/github.com/dgraph-io/badger/v4/README.md new file mode 100644 index 00000000000..0862aa4ca26 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/README.md @@ -0,0 +1,232 @@ +# BadgerDB + +[![Go Reference](https://pkg.go.dev/badge/github.com/dgraph-io/badger/v4.svg)](https://pkg.go.dev/github.com/dgraph-io/badger/v4) +[![Go Report Card](https://goreportcard.com/badge/github.com/dgraph-io/badger/v4)](https://goreportcard.com/report/github.com/dgraph-io/badger/v4) +[![Sourcegraph](https://sourcegraph.com/github.com/dgraph-io/badger/-/badge.svg)](https://sourcegraph.com/github.com/dgraph-io/badger?badge) +[![ci-badger-tests](https://github.com/dgraph-io/badger/actions/workflows/ci-badger-tests.yml/badge.svg)](https://github.com/dgraph-io/badger/actions/workflows/ci-badger-tests.yml) +[![ci-badger-bank-tests](https://github.com/dgraph-io/badger/actions/workflows/ci-badger-bank-tests.yml/badge.svg)](https://github.com/dgraph-io/badger/actions/workflows/ci-badger-bank-tests.yml) +[![ci-golang-lint](https://github.com/dgraph-io/badger/actions/workflows/ci-golang-lint.yml/badge.svg)](https://github.com/dgraph-io/badger/actions/workflows/ci-golang-lint.yml) + + +![Badger mascot](images/diggy-shadow.png) + +BadgerDB is an embeddable, persistent and fast key-value (KV) database written +in pure Go. It is the underlying database for [Dgraph](https://dgraph.io), a +fast, distributed graph database. It's meant to be a performant alternative to +non-Go-based key-value stores like RocksDB. + +## Project Status + +Badger is stable and is being used to serve data sets worth hundreds of +terabytes. Badger supports concurrent ACID transactions with serializable +snapshot isolation (SSI) guarantees. A Jepsen-style bank test runs nightly for +8h, with `--race` flag and ensures the maintenance of transactional guarantees. +Badger has also been tested to work with filesystem level anomalies, to ensure +persistence and consistency. Badger is being used by a number of projects which +includes Dgraph, Jaeger Tracing, UsenetExpress, and many more. + +The list of projects using Badger can be found [here](#projects-using-badger). + +Badger v1.0 was released in Nov 2017, and the latest version that is data-compatible +with v1.0 is v1.6.0. + +Badger v2.0 was released in Nov 2019 with a new storage format which won't +be compatible with all of the v1.x. Badger v2.0 supports compression, encryption and uses a cache to speed up lookup. + +Badger v3.0 was released in January 2021. This release improves compaction performance. + +Please consult the [Changelog] for more detailed information on releases. + +For more details on our version naming schema please read [Choosing a version](#choosing-a-version). + +[Changelog]:https://github.com/dgraph-io/badger/blob/master/CHANGELOG.md + +## Table of Contents +- [BadgerDB](#badgerdb) + - [Project Status](#project-status) + - [Table of Contents](#table-of-contents) + - [Getting Started](#getting-started) + - [Installing](#installing) + - [Installing Badger Command Line Tool](#installing-badger-command-line-tool) + - [Choosing a version](#choosing-a-version) + - [Badger Documentation](#badger-documentation) + - [Resources](#resources) + - [Blog Posts](#blog-posts) + - [Design](#design) + - [Comparisons](#comparisons) + - [Benchmarks](#benchmarks) + - [Projects Using Badger](#projects-using-badger) + - [Contributing](#contributing) + - [Contact](#contact) + +## Getting Started + +### Installing +To start using Badger, install Go 1.19 or above. Badger v3 needs go modules. From your project, run the following command + +```sh +$ go get github.com/dgraph-io/badger/v4 +``` +This will retrieve the library. + +#### Installing Badger Command Line Tool + +Badger provides a CLI tool which can perform certain operations like offline backup/restore. To install the Badger CLI, +retrieve the repository and checkout the desired version. Then run + +```sh +$ cd badger +$ go install . +``` +This will install the badger command line utility into your $GOBIN path. + +#### Choosing a version + +BadgerDB is a pretty special package from the point of view that the most important change we can +make to it is not on its API but rather on how data is stored on disk. + +This is why we follow a version naming schema that differs from Semantic Versioning. + +- New major versions are released when the data format on disk changes in an incompatible way. +- New minor versions are released whenever the API changes but data compatibility is maintained. + Note that the changes on the API could be backward-incompatible - unlike Semantic Versioning. +- New patch versions are released when there's no changes to the data format nor the API. + +Following these rules: + +- v1.5.0 and v1.6.0 can be used on top of the same files without any concerns, as their major + version is the same, therefore the data format on disk is compatible. +- v1.6.0 and v2.0.0 are data incompatible as their major version implies, so files created with + v1.6.0 will need to be converted into the new format before they can be used by v2.0.0. + - v2.x.x and v3.x.x are data incompatible as their major version implies, so files created with + v2.x.x will need to be converted into the new format before they can be used by v3.0.0. + + +For a longer explanation on the reasons behind using a new versioning naming schema, you can read +[VERSIONING](VERSIONING.md). + +## Badger Documentation + +Badger Documentation is available at https://dgraph.io/docs/badger + +## Resources + +### Blog Posts +1. [Introducing Badger: A fast key-value store written natively in +Go](https://open.dgraph.io/post/badger/) +2. [Make Badger crash resilient with ALICE](https://open.dgraph.io/post/alice/) +3. [Badger vs LMDB vs BoltDB: Benchmarking key-value databases in Go](https://open.dgraph.io/post/badger-lmdb-boltdb/) +4. [Concurrent ACID Transactions in Badger](https://open.dgraph.io/post/badger-txn/) + +## Design +Badger was written with these design goals in mind: + +- Write a key-value database in pure Go. +- Use latest research to build the fastest KV database for data sets spanning terabytes. +- Optimize for SSDs. + +Badger’s design is based on a paper titled _[WiscKey: Separating Keys from +Values in SSD-conscious Storage][wisckey]_. + +[wisckey]: https://www.usenix.org/system/files/conference/fast16/fast16-papers-lu.pdf + +### Comparisons +| Feature | Badger | RocksDB | BoltDB | +| ------- | ------ | ------- | ------ | +| Design | LSM tree with value log | LSM tree only | B+ tree | +| High Read throughput | Yes | No | Yes | +| High Write throughput | Yes | Yes | No | +| Designed for SSDs | Yes (with latest research 1) | Not specifically 2 | No | +| Embeddable | Yes | Yes | Yes | +| Sorted KV access | Yes | Yes | Yes | +| Pure Go (no Cgo) | Yes | No | Yes | +| Transactions | Yes, ACID, concurrent with SSI3 | Yes (but non-ACID) | Yes, ACID | +| Snapshots | Yes | Yes | Yes | +| TTL support | Yes | Yes | No | +| 3D access (key-value-version) | Yes4 | No | No | + +1 The [WISCKEY paper][wisckey] (on which Badger is based) saw big +wins with separating values from keys, significantly reducing the write +amplification compared to a typical LSM tree. + +2 RocksDB is an SSD optimized version of LevelDB, which was designed specifically for rotating disks. +As such RocksDB's design isn't aimed at SSDs. + +3 SSI: Serializable Snapshot Isolation. For more details, see the blog post [Concurrent ACID Transactions in Badger](https://blog.dgraph.io/post/badger-txn/) + +4 Badger provides direct access to value versions via its Iterator API. +Users can also specify how many versions to keep per key via Options. + +### Benchmarks +We have run comprehensive benchmarks against RocksDB, Bolt and LMDB. The +benchmarking code, and the detailed logs for the benchmarks can be found in the +[badger-bench] repo. More explanation, including graphs can be found the blog posts (linked +above). + +[badger-bench]: https://github.com/dgraph-io/badger-bench + +## Projects Using Badger +Below is a list of known projects that use Badger: + +* [Dgraph](https://github.com/dgraph-io/dgraph) - Distributed graph database. +* [Jaeger](https://github.com/jaegertracing/jaeger) - Distributed tracing platform. +* [go-ipfs](https://github.com/ipfs/go-ipfs) - Go client for the InterPlanetary File System (IPFS), a new hypermedia distribution protocol. +* [Riot](https://github.com/go-ego/riot) - An open-source, distributed search engine. +* [emitter](https://github.com/emitter-io/emitter) - Scalable, low latency, distributed pub/sub broker with message storage, uses MQTT, gossip and badger. +* [OctoSQL](https://github.com/cube2222/octosql) - Query tool that allows you to join, analyse and transform data from multiple databases using SQL. +* [Dkron](https://dkron.io/) - Distributed, fault tolerant job scheduling system. +* [smallstep/certificates](https://github.com/smallstep/certificates) - Step-ca is an online certificate authority for secure, automated certificate management. +* [Sandglass](https://github.com/celrenheit/sandglass) - distributed, horizontally scalable, persistent, time sorted message queue. +* [TalariaDB](https://github.com/grab/talaria) - Grab's Distributed, low latency time-series database. +* [Sloop](https://github.com/salesforce/sloop) - Salesforce's Kubernetes History Visualization Project. +* [Usenet Express](https://usenetexpress.com/) - Serving over 300TB of data with Badger. +* [gorush](https://github.com/appleboy/gorush) - A push notification server written in Go. +* [0-stor](https://github.com/zero-os/0-stor) - Single device object store. +* [Dispatch Protocol](https://github.com/dispatchlabs/disgo) - Blockchain protocol for distributed application data analytics. +* [GarageMQ](https://github.com/valinurovam/garagemq) - AMQP server written in Go. +* [RedixDB](https://alash3al.github.io/redix/) - A real-time persistent key-value store with the same redis protocol. +* [BBVA](https://github.com/BBVA/raft-badger) - Raft backend implementation using BadgerDB for Hashicorp raft. +* [Fantom](https://github.com/Fantom-foundation/go-lachesis) - aBFT Consensus platform for distributed applications. +* [decred](https://github.com/decred/dcrdata) - An open, progressive, and self-funding cryptocurrency with a system of community-based governance integrated into its blockchain. +* [OpenNetSys](https://github.com/opennetsys/c3-go) - Create useful dApps in any software language. +* [HoneyTrap](https://github.com/honeytrap/honeytrap) - An extensible and opensource system for running, monitoring and managing honeypots. +* [Insolar](https://github.com/insolar/insolar) - Enterprise-ready blockchain platform. +* [IoTeX](https://github.com/iotexproject/iotex-core) - The next generation of the decentralized network for IoT powered by scalability- and privacy-centric blockchains. +* [go-sessions](https://github.com/kataras/go-sessions) - The sessions manager for Go net/http and fasthttp. +* [Babble](https://github.com/mosaicnetworks/babble) - BFT Consensus platform for distributed applications. +* [Tormenta](https://github.com/jpincas/tormenta) - Embedded object-persistence layer / simple JSON database for Go projects. +* [BadgerHold](https://github.com/timshannon/badgerhold) - An embeddable NoSQL store for querying Go types built on Badger +* [Goblero](https://github.com/didil/goblero) - Pure Go embedded persistent job queue backed by BadgerDB +* [Surfline](https://www.surfline.com) - Serving global wave and weather forecast data with Badger. +* [Cete](https://github.com/mosuka/cete) - Simple and highly available distributed key-value store built on Badger. Makes it easy bringing up a cluster of Badger with Raft consensus algorithm by hashicorp/raft. +* [Volument](https://volument.com/) - A new take on website analytics backed by Badger. +* [KVdb](https://kvdb.io/) - Hosted key-value store and serverless platform built on top of Badger. +* [Terminotes](https://gitlab.com/asad-awadia/terminotes) - Self hosted notes storage and search server - storage powered by BadgerDB +* [Pyroscope](https://github.com/pyroscope-io/pyroscope) - Open source confinuous profiling platform built with BadgerDB +* [Veri](https://github.com/bgokden/veri) - A distributed feature store optimized for Search and Recommendation tasks. +* [bIter](https://github.com/MikkelHJuul/bIter) - A library and Iterator interface for working with the `badger.Iterator`, simplifying from-to, and prefix mechanics. +* [ld](https://github.com/MikkelHJuul/ld) - (Lean Database) A very simple gRPC-only key-value database, exposing BadgerDB with key-range scanning semantics. +* [Souin](https://github.com/darkweak/Souin) - A RFC compliant HTTP cache with lot of other features based on Badger for the storage. Compatible with all existing reverse-proxies. +* [Xuperchain](https://github.com/xuperchain/xupercore) - A highly flexible blockchain architecture with great transaction performance. +* [m2](https://github.com/qichengzx/m2) - A simple http key/value store based on the raft protocol. +* [chaindb](https://github.com/ChainSafe/chaindb) - A blockchain storage layer used by [Gossamer](https://chainsafe.github.io/gossamer/), a Go client for the [Polkadot Network](https://polkadot.network/). +* [vxdb](https://github.com/vitalvas/vxdb) - Simple schema-less Key-Value NoSQL database with simplest API interface. +* [Opacity](https://github.com/opacity/storage-node) - Backend implementation for the Opacity storage project +* [Vephar](https://github.com/vaccovecrana/vephar) - A minimal key/value store using hashicorp-raft for cluster coordination and Badger for data storage. +* [gowarcserver](https://github.com/nlnwa/gowarcserver) - Open-source server for warc files. Can be used in conjunction with pywb +* [flow-go](https://github.com/onflow/flow-go) - A fast, secure, and developer-friendly blockchain built to support the next generation of games, apps and the digital assets that power them. +* [Wrgl](https://www.wrgl.co) - A data version control system that works like Git but specialized to store and diff CSV. +* [Loggie](https://github.com/loggie-io/loggie) - A lightweight, cloud-native data transfer agent and aggregator. +* [raft-badger](https://github.com/rfyiamcool/raft-badger) - raft-badger implements LogStore and StableStore Interface of hashcorp/raft. it is used to store raft log and metadata of hashcorp/raft. +* [DVID](https://github.com/janelia-flyem/dvid) - A dataservice for branched versioning of a variety of data types. Originally created for large-scale brain reconstructions in Connectomics. + +If you are using Badger in a project please send a pull request to add it to the list. + +## Contributing + +If you're interested in contributing to Badger see [CONTRIBUTING](./CONTRIBUTING.md). + +## Contact +- Please use [Github issues](https://github.com/dgraph-io/badger/issues) for filing bugs. +- Please use [discuss.dgraph.io](https://discuss.dgraph.io) for questions, discussions, and feature requests. +- Follow us on Twitter [@dgraphlabs](https://twitter.com/dgraphlabs). diff --git a/vendor/github.com/dgraph-io/badger/v4/VERSIONING.md b/vendor/github.com/dgraph-io/badger/v4/VERSIONING.md new file mode 100644 index 00000000000..7741bd40451 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/VERSIONING.md @@ -0,0 +1,47 @@ +# Serialization Versioning: Semantic Versioning for databases + +Semantic Versioning, commonly known as SemVer, is a great idea that has been very widely adopted as +a way to decide how to name software versions. The whole concept is very well summarized on +semver.org with the following lines: + +> Given a version number MAJOR.MINOR.PATCH, increment the: +> +> 1. MAJOR version when you make incompatible API changes, +> 2. MINOR version when you add functionality in a backwards-compatible manner, and +> 3. PATCH version when you make backwards-compatible bug fixes. +> +> Additional labels for pre-release and build metadata are available as extensions to the +> MAJOR.MINOR.PATCH format. + +Unfortunately, API changes are not the most important changes for libraries that serialize data for +later consumption. For these libraries, such as BadgerDB, changes to the API are much easier to +handle than change to the data format used to store data on disk. + +## Serialization Version specification + +Serialization Versioning, like Semantic Versioning, uses 3 numbers and also calls them +MAJOR.MINOR.PATCH, but the semantics of the numbers are slightly modified: + +Given a version number MAJOR.MINOR.PATCH, increment the: + +- MAJOR version when you make changes that require a transformation of the dataset before it can be +used again. +- MINOR version when old datasets are still readable but the API might have changed in +backwards-compatible or incompatible ways. +- PATCH version when you make backwards-compatible bug fixes. + +Additional labels for pre-release and build metadata are available as extensions to the +MAJOR.MINOR.PATCH format. + +Following this naming strategy, migration from v1.x to v2.x requires a migration strategy for your +existing dataset, and as such has to be carefully planned. Migrations in between different minor +versions (e.g. v1.5.x and v1.6.x) might break your build, as the API *might* have changed, but once +your code compiles there's no need for any data migration. Lastly, changes in between two different +patch versions should never break your build or dataset. + +For more background on our decision to adopt Serialization Versioning, read the blog post +[Semantic Versioning, Go Modules, and Databases][blog] and the original proposal on +[this comment on Dgraph's Discuss forum][discuss]. + +[blog]: https://open.dgraph.io/post/serialization-versioning/ +[discuss]: https://discuss.dgraph.io/t/go-modules-on-badger-and-dgraph/4662/7 \ No newline at end of file diff --git a/vendor/github.com/dgraph-io/badger/v4/appveyor.yml b/vendor/github.com/dgraph-io/badger/v4/appveyor.yml new file mode 100644 index 00000000000..a3eaffdd43b --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/appveyor.yml @@ -0,0 +1,49 @@ +# version format +version: "{build}" + +# Operating system (build VM template) +os: Windows Server 2012 R2 + +# Platform. +platform: x64 + +clone_folder: c:\gopath\src\github.com\dgraph-io\badger + +# Environment variables +environment: + GOVERSION: 1.19 + GOPATH: c:\gopath + GO111MODULE: on + +# scripts that run after cloning repository +install: + - set PATH=%GOPATH%\bin;c:\go\bin;c:\msys64\mingw64\bin;%PATH% + - go version + - go env + - python --version + - gcc --version + +# To run your custom scripts instead of automatic MSBuild +build_script: + # We need to disable firewall - https://github.com/appveyor/ci/issues/1579#issuecomment-309830648 + - ps: Disable-NetFirewallRule -DisplayName 'File and Printer Sharing (SMB-Out)' + - cd c:\gopath\src\github.com\dgraph-io\badger + - git branch + - go get -t ./... + +# To run your custom scripts instead of automatic tests +test_script: + # Unit tests + - ps: Add-AppveyorTest "Unit Tests" -Outcome Running + - go test -v github.com/dgraph-io/badger/... + - ps: Update-AppveyorTest "Unit Tests" -Outcome Passed + +notifications: + - provider: Email + to: + - pawan@dgraph.io + on_build_failure: true + on_build_status_changed: true +# to disable deployment +deploy: off + diff --git a/vendor/github.com/dgraph-io/badger/v4/backup.go b/vendor/github.com/dgraph-io/badger/v4/backup.go new file mode 100644 index 00000000000..f9629064fd1 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/backup.go @@ -0,0 +1,289 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bufio" + "bytes" + "context" + "encoding/binary" + "io" + + "github.com/golang/protobuf/proto" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// flushThreshold determines when a buffer will be flushed. When performing a +// backup/restore, the entries will be batched up until the total size of batch +// is more than flushThreshold or entry size (without the value size) is more +// than the maxBatchSize. +const flushThreshold = 100 << 20 + +// Backup dumps a protobuf-encoded list of all entries in the database into the +// given writer, that are newer than or equal to the specified version. It +// returns a timestamp (version) indicating the version of last entry that is +// dumped, which after incrementing by 1 can be passed into later invocation to +// generate incremental backup of entries that have been added/modified since +// the last invocation of DB.Backup(). +// DB.Backup is a wrapper function over Stream.Backup to generate full and +// incremental backups of the DB. For more control over how many goroutines are +// used to generate the backup, or if you wish to backup only a certain range +// of keys, use Stream.Backup directly. +func (db *DB) Backup(w io.Writer, since uint64) (uint64, error) { + stream := db.NewStream() + stream.LogPrefix = "DB.Backup" + stream.SinceTs = since + return stream.Backup(w, since) +} + +// Backup dumps a protobuf-encoded list of all entries in the database into the +// given writer, that are newer than or equal to the specified version. It returns a +// timestamp(version) indicating the version of last entry that was dumped, which +// after incrementing by 1 can be passed into a later invocation to generate an +// incremental dump of entries that have been added/modified since the last +// invocation of Stream.Backup(). +// +// This can be used to backup the data in a database at a given point in time. +func (stream *Stream) Backup(w io.Writer, since uint64) (uint64, error) { + stream.KeyToList = func(key []byte, itr *Iterator) (*pb.KVList, error) { + list := &pb.KVList{} + a := itr.Alloc + for ; itr.Valid(); itr.Next() { + item := itr.Item() + if !bytes.Equal(item.Key(), key) { + return list, nil + } + if item.Version() < since { + return nil, errors.Errorf("Backup: Item Version: %d less than sinceTs: %d", + item.Version(), since) + } + + var valCopy []byte + if !item.IsDeletedOrExpired() { + // No need to copy value, if item is deleted or expired. + err := item.Value(func(val []byte) error { + valCopy = a.Copy(val) + return nil + }) + if err != nil { + stream.db.opt.Errorf("Key [%x, %d]. Error while fetching value [%v]\n", + item.Key(), item.Version(), err) + return nil, err + } + } + + // clear txn bits + meta := item.meta &^ (bitTxn | bitFinTxn) + kv := y.NewKV(a) + *kv = pb.KV{ + Key: a.Copy(item.Key()), + Value: valCopy, + UserMeta: a.Copy([]byte{item.UserMeta()}), + Version: item.Version(), + ExpiresAt: item.ExpiresAt(), + Meta: a.Copy([]byte{meta}), + } + list.Kv = append(list.Kv, kv) + + switch { + case item.DiscardEarlierVersions(): + // If we need to discard earlier versions of this item, add a delete + // marker just below the current version. + list.Kv = append(list.Kv, &pb.KV{ + Key: item.KeyCopy(nil), + Version: item.Version() - 1, + Meta: []byte{bitDelete}, + }) + return list, nil + + case item.IsDeletedOrExpired(): + return list, nil + } + } + return list, nil + } + + var maxVersion uint64 + stream.Send = func(buf *z.Buffer) error { + list, err := BufferToKVList(buf) + if err != nil { + return err + } + out := list.Kv[:0] + for _, kv := range list.Kv { + if maxVersion < kv.Version { + maxVersion = kv.Version + } + if !kv.StreamDone { + // Don't pick stream done changes. + out = append(out, kv) + } + } + list.Kv = out + return writeTo(list, w) + } + + if err := stream.Orchestrate(context.Background()); err != nil { + return 0, err + } + return maxVersion, nil +} + +func writeTo(list *pb.KVList, w io.Writer) error { + if err := binary.Write(w, binary.LittleEndian, uint64(proto.Size(list))); err != nil { + return err + } + buf, err := proto.Marshal(list) + if err != nil { + return err + } + _, err = w.Write(buf) + return err +} + +// KVLoader is used to write KVList objects in to badger. It can be used to restore a backup. +type KVLoader struct { + db *DB + throttle *y.Throttle + entries []*Entry + entriesSize int64 + totalSize int64 +} + +// NewKVLoader returns a new instance of KVLoader. +func (db *DB) NewKVLoader(maxPendingWrites int) *KVLoader { + return &KVLoader{ + db: db, + throttle: y.NewThrottle(maxPendingWrites), + entries: make([]*Entry, 0, db.opt.maxBatchCount), + } +} + +// Set writes the key-value pair to the database. +func (l *KVLoader) Set(kv *pb.KV) error { + var userMeta, meta byte + if len(kv.UserMeta) > 0 { + userMeta = kv.UserMeta[0] + } + if len(kv.Meta) > 0 { + meta = kv.Meta[0] + } + e := &Entry{ + Key: y.KeyWithTs(kv.Key, kv.Version), + Value: kv.Value, + UserMeta: userMeta, + ExpiresAt: kv.ExpiresAt, + meta: meta, + } + estimatedSize := e.estimateSizeAndSetThreshold(l.db.valueThreshold()) + // Flush entries if inserting the next entry would overflow the transactional limits. + if int64(len(l.entries))+1 >= l.db.opt.maxBatchCount || + l.entriesSize+estimatedSize >= l.db.opt.maxBatchSize || + l.totalSize >= flushThreshold { + if err := l.send(); err != nil { + return err + } + } + l.entries = append(l.entries, e) + l.entriesSize += estimatedSize + l.totalSize += estimatedSize + int64(len(e.Value)) + return nil +} + +func (l *KVLoader) send() error { + if err := l.throttle.Do(); err != nil { + return err + } + if err := l.db.batchSetAsync(l.entries, func(err error) { + l.throttle.Done(err) + }); err != nil { + return err + } + + l.entries = make([]*Entry, 0, l.db.opt.maxBatchCount) + l.entriesSize = 0 + l.totalSize = 0 + return nil +} + +// Finish is meant to be called after all the key-value pairs have been loaded. +func (l *KVLoader) Finish() error { + if len(l.entries) > 0 { + if err := l.send(); err != nil { + return err + } + } + return l.throttle.Finish() +} + +// Load reads a protobuf-encoded list of all entries from a reader and writes +// them to the database. This can be used to restore the database from a backup +// made by calling DB.Backup(). If more complex logic is needed to restore a badger +// backup, the KVLoader interface should be used instead. +// +// DB.Load() should be called on a database that is not running any other +// concurrent transactions while it is running. +func (db *DB) Load(r io.Reader, maxPendingWrites int) error { + br := bufio.NewReaderSize(r, 16<<10) + unmarshalBuf := make([]byte, 1<<10) + + ldr := db.NewKVLoader(maxPendingWrites) + for { + var sz uint64 + err := binary.Read(br, binary.LittleEndian, &sz) + if err == io.EOF { + break + } else if err != nil { + return err + } + + if cap(unmarshalBuf) < int(sz) { + unmarshalBuf = make([]byte, sz) + } + + if _, err = io.ReadFull(br, unmarshalBuf[:sz]); err != nil { + return err + } + + list := &pb.KVList{} + if err := proto.Unmarshal(unmarshalBuf[:sz], list); err != nil { + return err + } + + for _, kv := range list.Kv { + if err := ldr.Set(kv); err != nil { + return err + } + + // Update nextTxnTs, memtable stores this + // timestamp in badger head when flushed. + if kv.Version >= db.orc.nextTxnTs { + db.orc.nextTxnTs = kv.Version + 1 + } + } + } + + if err := ldr.Finish(); err != nil { + return err + } + db.orc.txnMark.Done(db.orc.nextTxnTs - 1) + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/batch.go b/vendor/github.com/dgraph-io/badger/v4/batch.go new file mode 100644 index 00000000000..885451fe9c3 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/batch.go @@ -0,0 +1,246 @@ +/* + * Copyright 2018 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "sync" + "sync/atomic" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// WriteBatch holds the necessary info to perform batched writes. +type WriteBatch struct { + sync.Mutex + txn *Txn + db *DB + throttle *y.Throttle + err atomic.Value + + isManaged bool + commitTs uint64 + finished bool +} + +// NewWriteBatch creates a new WriteBatch. This provides a way to conveniently do a lot of writes, +// batching them up as tightly as possible in a single transaction and using callbacks to avoid +// waiting for them to commit, thus achieving good performance. This API hides away the logic of +// creating and committing transactions. Due to the nature of SSI guaratees provided by Badger, +// blind writes can never encounter transaction conflicts (ErrConflict). +func (db *DB) NewWriteBatch() *WriteBatch { + if db.opt.managedTxns { + panic("cannot use NewWriteBatch in managed mode. Use NewWriteBatchAt instead") + } + return db.newWriteBatch(false) +} + +func (db *DB) newWriteBatch(isManaged bool) *WriteBatch { + return &WriteBatch{ + db: db, + isManaged: isManaged, + txn: db.newTransaction(true, isManaged), + throttle: y.NewThrottle(16), + } +} + +// SetMaxPendingTxns sets a limit on maximum number of pending transactions while writing batches. +// This function should be called before using WriteBatch. Default value of MaxPendingTxns is +// 16 to minimise memory usage. +func (wb *WriteBatch) SetMaxPendingTxns(max int) { + wb.throttle = y.NewThrottle(max) +} + +// Cancel function must be called if there's a chance that Flush might not get +// called. If neither Flush or Cancel is called, the transaction oracle would +// never get a chance to clear out the row commit timestamp map, thus causing an +// unbounded memory consumption. Typically, you can call Cancel as a defer +// statement right after NewWriteBatch is called. +// +// Note that any committed writes would still go through despite calling Cancel. +func (wb *WriteBatch) Cancel() { + wb.Lock() + defer wb.Unlock() + wb.finished = true + if err := wb.throttle.Finish(); err != nil { + wb.db.opt.Errorf("WatchBatch.Cancel error while finishing: %v", err) + } + wb.txn.Discard() +} + +func (wb *WriteBatch) callback(err error) { + // sync.WaitGroup is thread-safe, so it doesn't need to be run inside wb.Lock. + defer wb.throttle.Done(err) + if err == nil { + return + } + if err := wb.Error(); err != nil { + return + } + wb.err.Store(err) +} + +func (wb *WriteBatch) writeKV(kv *pb.KV) error { + e := Entry{Key: kv.Key, Value: kv.Value} + if len(kv.UserMeta) > 0 { + e.UserMeta = kv.UserMeta[0] + } + y.AssertTrue(kv.Version != 0) + e.version = kv.Version + return wb.handleEntry(&e) +} + +func (wb *WriteBatch) Write(buf *z.Buffer) error { + wb.Lock() + defer wb.Unlock() + + err := buf.SliceIterate(func(s []byte) error { + kv := &pb.KV{} + if err := kv.Unmarshal(s); err != nil { + return err + } + return wb.writeKV(kv) + }) + return err +} + +func (wb *WriteBatch) WriteList(kvList *pb.KVList) error { + wb.Lock() + defer wb.Unlock() + for _, kv := range kvList.Kv { + if err := wb.writeKV(kv); err != nil { + return err + } + } + return nil +} + +// SetEntryAt is the equivalent of Txn.SetEntry but it also allows setting version for the entry. +// SetEntryAt can be used only in managed mode. +func (wb *WriteBatch) SetEntryAt(e *Entry, ts uint64) error { + if !wb.db.opt.managedTxns { + return errors.New("SetEntryAt can only be used in managed mode. Use SetEntry instead") + } + e.version = ts + return wb.SetEntry(e) +} + +// Should be called with lock acquired. +func (wb *WriteBatch) handleEntry(e *Entry) error { + if err := wb.txn.SetEntry(e); err != ErrTxnTooBig { + return err + } + // Txn has reached it's zenith. Commit now. + if cerr := wb.commit(); cerr != nil { + return cerr + } + // This time the error must not be ErrTxnTooBig, otherwise, we make the + // error permanent. + if err := wb.txn.SetEntry(e); err != nil { + wb.err.Store(err) + return err + } + return nil +} + +// SetEntry is the equivalent of Txn.SetEntry. +func (wb *WriteBatch) SetEntry(e *Entry) error { + wb.Lock() + defer wb.Unlock() + return wb.handleEntry(e) +} + +// Set is equivalent of Txn.Set(). +func (wb *WriteBatch) Set(k, v []byte) error { + e := &Entry{Key: k, Value: v} + return wb.SetEntry(e) +} + +// DeleteAt is equivalent of Txn.Delete but accepts a delete timestamp. +func (wb *WriteBatch) DeleteAt(k []byte, ts uint64) error { + e := Entry{Key: k, meta: bitDelete, version: ts} + return wb.SetEntry(&e) +} + +// Delete is equivalent of Txn.Delete. +func (wb *WriteBatch) Delete(k []byte) error { + wb.Lock() + defer wb.Unlock() + + if err := wb.txn.Delete(k); err != ErrTxnTooBig { + return err + } + if err := wb.commit(); err != nil { + return err + } + if err := wb.txn.Delete(k); err != nil { + wb.err.Store(err) + return err + } + return nil +} + +// Caller to commit must hold a write lock. +func (wb *WriteBatch) commit() error { + if err := wb.Error(); err != nil { + return err + } + if wb.finished { + return y.ErrCommitAfterFinish + } + if err := wb.throttle.Do(); err != nil { + wb.err.Store(err) + return err + } + wb.txn.CommitWith(wb.callback) + wb.txn = wb.db.newTransaction(true, wb.isManaged) + wb.txn.commitTs = wb.commitTs + return wb.Error() +} + +// Flush must be called at the end to ensure that any pending writes get committed to Badger. Flush +// returns any error stored by WriteBatch. +func (wb *WriteBatch) Flush() error { + wb.Lock() + err := wb.commit() + if err != nil { + wb.Unlock() + return err + } + wb.finished = true + wb.txn.Discard() + wb.Unlock() + + if err := wb.throttle.Finish(); err != nil { + if wb.Error() != nil { + return errors.Errorf("wb.err: %s err: %s", wb.Error(), err) + } + return err + } + + return wb.Error() +} + +// Error returns any errors encountered so far. No commits would be run once an error is detected. +func (wb *WriteBatch) Error() error { + // If the interface conversion fails, the err will be nil. + err, _ := wb.err.Load().(error) + return err +} diff --git a/vendor/github.com/dgraph-io/badger/v4/changes.sh b/vendor/github.com/dgraph-io/badger/v4/changes.sh new file mode 100644 index 00000000000..e7cede91515 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/changes.sh @@ -0,0 +1,18 @@ +#!/bin/bash + +set -e +GHORG=${GHORG:-dgraph-io} +GHREPO=${GHREPO:-badger} +cat < 0 { + r.right = kr.right + } + if kr.inf { + r.inf = true + } +} + +func (r keyRange) overlapsWith(dst keyRange) bool { + // Empty keyRange always overlaps. + if r.isEmpty() { + return true + } + // TODO(ibrahim): Do you need this? + // Empty dst doesn't overlap with anything. + if dst.isEmpty() { + return false + } + if r.inf || dst.inf { + return true + } + + // [dst.left, dst.right] ... [r.left, r.right] + // If my left is greater than dst right, we have no overlap. + if y.CompareKeys(r.left, dst.right) > 0 { + return false + } + // [r.left, r.right] ... [dst.left, dst.right] + // If my right is less than dst left, we have no overlap. + if y.CompareKeys(r.right, dst.left) < 0 { + return false + } + // We have overlap. + return true +} + +// getKeyRange returns the smallest and the biggest in the list of tables. +// TODO(naman): Write a test for this. The smallest and the biggest should +// be the smallest of the leftmost table and the biggest of the right most table. +func getKeyRange(tables ...*table.Table) keyRange { + if len(tables) == 0 { + return keyRange{} + } + smallest := tables[0].Smallest() + biggest := tables[0].Biggest() + for i := 1; i < len(tables); i++ { + if y.CompareKeys(tables[i].Smallest(), smallest) < 0 { + smallest = tables[i].Smallest() + } + if y.CompareKeys(tables[i].Biggest(), biggest) > 0 { + biggest = tables[i].Biggest() + } + } + + // We pick all the versions of the smallest and the biggest key. Note that version zero would + // be the rightmost key, considering versions are default sorted in descending order. + return keyRange{ + left: y.KeyWithTs(y.ParseKey(smallest), math.MaxUint64), + right: y.KeyWithTs(y.ParseKey(biggest), 0), + } +} + +type levelCompactStatus struct { + ranges []keyRange + delSize int64 +} + +func (lcs *levelCompactStatus) debug() string { + var b bytes.Buffer + for _, r := range lcs.ranges { + b.WriteString(r.String()) + } + return b.String() +} + +func (lcs *levelCompactStatus) overlapsWith(dst keyRange) bool { + for _, r := range lcs.ranges { + if r.overlapsWith(dst) { + return true + } + } + return false +} + +func (lcs *levelCompactStatus) remove(dst keyRange) bool { + final := lcs.ranges[:0] + var found bool + for _, r := range lcs.ranges { + if !r.equals(dst) { + final = append(final, r) + } else { + found = true + } + } + lcs.ranges = final + return found +} + +type compactStatus struct { + sync.RWMutex + levels []*levelCompactStatus + tables map[uint64]struct{} +} + +func (cs *compactStatus) overlapsWith(level int, this keyRange) bool { + cs.RLock() + defer cs.RUnlock() + + thisLevel := cs.levels[level] + return thisLevel.overlapsWith(this) +} + +func (cs *compactStatus) delSize(l int) int64 { + cs.RLock() + defer cs.RUnlock() + return cs.levels[l].delSize +} + +type thisAndNextLevelRLocked struct{} + +// compareAndAdd will check whether we can run this compactDef. That it doesn't overlap with any +// other running compaction. If it can be run, it would store this run in the compactStatus state. +func (cs *compactStatus) compareAndAdd(_ thisAndNextLevelRLocked, cd compactDef) bool { + cs.Lock() + defer cs.Unlock() + + tl := cd.thisLevel.level + y.AssertTruef(tl < len(cs.levels), "Got level %d. Max levels: %d", tl, len(cs.levels)) + thisLevel := cs.levels[cd.thisLevel.level] + nextLevel := cs.levels[cd.nextLevel.level] + + if thisLevel.overlapsWith(cd.thisRange) { + return false + } + if nextLevel.overlapsWith(cd.nextRange) { + return false + } + // Check whether this level really needs compaction or not. Otherwise, we'll end up + // running parallel compactions for the same level. + // Update: We should not be checking size here. Compaction priority already did the size checks. + // Here we should just be executing the wish of others. + + thisLevel.ranges = append(thisLevel.ranges, cd.thisRange) + nextLevel.ranges = append(nextLevel.ranges, cd.nextRange) + thisLevel.delSize += cd.thisSize + for _, t := range append(cd.top, cd.bot...) { + cs.tables[t.ID()] = struct{}{} + } + return true +} + +func (cs *compactStatus) delete(cd compactDef) { + cs.Lock() + defer cs.Unlock() + + tl := cd.thisLevel.level + y.AssertTruef(tl < len(cs.levels), "Got level %d. Max levels: %d", tl, len(cs.levels)) + + thisLevel := cs.levels[cd.thisLevel.level] + nextLevel := cs.levels[cd.nextLevel.level] + + thisLevel.delSize -= cd.thisSize + found := thisLevel.remove(cd.thisRange) + // The following check makes sense only if we're compacting more than one + // table. In case of the max level, we might rewrite a single table to + // remove stale data. + if cd.thisLevel != cd.nextLevel && !cd.nextRange.isEmpty() { + found = nextLevel.remove(cd.nextRange) && found + } + + if !found { + this := cd.thisRange + next := cd.nextRange + fmt.Printf("Looking for: %s in this level %d.\n", this, tl) + fmt.Printf("This Level:\n%s\n", thisLevel.debug()) + fmt.Println() + fmt.Printf("Looking for: %s in next level %d.\n", next, cd.nextLevel.level) + fmt.Printf("Next Level:\n%s\n", nextLevel.debug()) + log.Fatal("keyRange not found") + } + for _, t := range append(cd.top, cd.bot...) { + _, ok := cs.tables[t.ID()] + y.AssertTrue(ok) + delete(cs.tables, t.ID()) + } +} diff --git a/vendor/github.com/dgraph-io/badger/v4/db.go b/vendor/github.com/dgraph-io/badger/v4/db.go new file mode 100644 index 00000000000..d30ac6c3d40 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/db.go @@ -0,0 +1,2087 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "context" + "encoding/binary" + "expvar" + "fmt" + "math" + "os" + "path/filepath" + "sort" + "strings" + "sync" + "sync/atomic" + "time" + + humanize "github.com/dustin/go-humanize" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/options" + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/skl" + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto" + "github.com/dgraph-io/ristretto/z" +) + +var ( + badgerPrefix = []byte("!badger!") // Prefix for internal keys used by badger. + txnKey = []byte("!badger!txn") // For indicating end of entries in txn. + bannedNsKey = []byte("!badger!banned") // For storing the banned namespaces. +) + +type closers struct { + updateSize *z.Closer + compactors *z.Closer + memtable *z.Closer + writes *z.Closer + valueGC *z.Closer + pub *z.Closer + cacheHealth *z.Closer +} + +type lockedKeys struct { + sync.RWMutex + keys map[uint64]struct{} +} + +func (lk *lockedKeys) add(key uint64) { + lk.Lock() + defer lk.Unlock() + lk.keys[key] = struct{}{} +} + +func (lk *lockedKeys) has(key uint64) bool { + lk.RLock() + defer lk.RUnlock() + _, ok := lk.keys[key] + return ok +} + +func (lk *lockedKeys) all() []uint64 { + lk.RLock() + defer lk.RUnlock() + keys := make([]uint64, 0, len(lk.keys)) + for key := range lk.keys { + keys = append(keys, key) + } + return keys +} + +// DB provides the various functions required to interact with Badger. +// DB is thread-safe. +type DB struct { + testOnlyDBExtensions + + lock sync.RWMutex // Guards list of inmemory tables, not individual reads and writes. + + dirLockGuard *directoryLockGuard + // nil if Dir and ValueDir are the same + valueDirGuard *directoryLockGuard + + closers closers + + mt *memTable // Our latest (actively written) in-memory table + imm []*memTable // Add here only AFTER pushing to flushChan. + + // Initialized via openMemTables. + nextMemFid int + + opt Options + manifest *manifestFile + lc *levelsController + vlog valueLog + writeCh chan *request + flushChan chan *memTable // For flushing memtables. + closeOnce sync.Once // For closing DB only once. + + blockWrites atomic.Int32 + isClosed atomic.Uint32 + + orc *oracle + bannedNamespaces *lockedKeys + threshold *vlogThreshold + + pub *publisher + registry *KeyRegistry + blockCache *ristretto.Cache + indexCache *ristretto.Cache + allocPool *z.AllocatorPool +} + +const ( + kvWriteChCapacity = 1000 +) + +func checkAndSetOptions(opt *Options) error { + // It's okay to have zero compactors which will disable all compactions but + // we cannot have just one compactor otherwise we will end up with all data + // on level 2. + if opt.NumCompactors == 1 { + return errors.New("Cannot have 1 compactor. Need at least 2") + } + + if opt.InMemory && (opt.Dir != "" || opt.ValueDir != "") { + return errors.New("Cannot use badger in Disk-less mode with Dir or ValueDir set") + } + opt.maxBatchSize = (15 * opt.MemTableSize) / 100 + opt.maxBatchCount = opt.maxBatchSize / int64(skl.MaxNodeSize) + + // This is the maximum value, vlogThreshold can have if dynamic thresholding is enabled. + opt.maxValueThreshold = math.Min(maxValueThreshold, float64(opt.maxBatchSize)) + if opt.VLogPercentile < 0.0 || opt.VLogPercentile > 1.0 { + return errors.New("vlogPercentile must be within range of 0.0-1.0") + } + + // We are limiting opt.ValueThreshold to maxValueThreshold for now. + if opt.ValueThreshold > maxValueThreshold { + return errors.Errorf("Invalid ValueThreshold, must be less or equal to %d", + maxValueThreshold) + } + + // If ValueThreshold is greater than opt.maxBatchSize, we won't be able to push any data using + // the transaction APIs. Transaction batches entries into batches of size opt.maxBatchSize. + if opt.ValueThreshold > opt.maxBatchSize { + return errors.Errorf("Valuethreshold %d greater than max batch size of %d. Either "+ + "reduce opt.ValueThreshold or increase opt.MaxTableSize.", + opt.ValueThreshold, opt.maxBatchSize) + } + // ValueLogFileSize should be stricly LESS than 2<<30 otherwise we will + // overflow the uint32 when we mmap it in OpenMemtable. + if !(opt.ValueLogFileSize < 2<<30 && opt.ValueLogFileSize >= 1<<20) { + return ErrValueLogSize + } + + if opt.ReadOnly { + // Do not perform compaction in read only mode. + opt.CompactL0OnClose = false + } + + needCache := (opt.Compression != options.None) || (len(opt.EncryptionKey) > 0) + if needCache && opt.BlockCacheSize == 0 { + panic("BlockCacheSize should be set since compression/encryption are enabled") + } + return nil +} + +// Open returns a new DB object. +func Open(opt Options) (*DB, error) { + if err := checkAndSetOptions(&opt); err != nil { + return nil, err + } + var dirLockGuard, valueDirLockGuard *directoryLockGuard + + // Create directories and acquire lock on it only if badger is not running in InMemory mode. + // We don't have any directories/files in InMemory mode so we don't need to acquire + // any locks on them. + if !opt.InMemory { + if err := createDirs(opt); err != nil { + return nil, err + } + var err error + if !opt.BypassLockGuard { + dirLockGuard, err = acquireDirectoryLock(opt.Dir, lockFile, opt.ReadOnly) + if err != nil { + return nil, err + } + defer func() { + if dirLockGuard != nil { + _ = dirLockGuard.release() + } + }() + absDir, err := filepath.Abs(opt.Dir) + if err != nil { + return nil, err + } + absValueDir, err := filepath.Abs(opt.ValueDir) + if err != nil { + return nil, err + } + if absValueDir != absDir { + valueDirLockGuard, err = acquireDirectoryLock(opt.ValueDir, lockFile, opt.ReadOnly) + if err != nil { + return nil, err + } + defer func() { + if valueDirLockGuard != nil { + _ = valueDirLockGuard.release() + } + }() + } + } + } + + manifestFile, manifest, err := openOrCreateManifestFile(opt) + if err != nil { + return nil, err + } + defer func() { + if manifestFile != nil { + _ = manifestFile.close() + } + }() + + db := &DB{ + imm: make([]*memTable, 0, opt.NumMemtables), + flushChan: make(chan *memTable, opt.NumMemtables), + writeCh: make(chan *request, kvWriteChCapacity), + opt: opt, + manifest: manifestFile, + dirLockGuard: dirLockGuard, + valueDirGuard: valueDirLockGuard, + orc: newOracle(opt), + pub: newPublisher(), + allocPool: z.NewAllocatorPool(8), + bannedNamespaces: &lockedKeys{keys: make(map[uint64]struct{})}, + threshold: initVlogThreshold(&opt), + } + + db.syncChan = opt.syncChan + + // Cleanup all the goroutines started by badger in case of an error. + defer func() { + if err != nil { + opt.Errorf("Received err: %v. Cleaning up...", err) + db.cleanup() + db = nil + } + }() + + if opt.BlockCacheSize > 0 { + numInCache := opt.BlockCacheSize / int64(opt.BlockSize) + if numInCache == 0 { + // Make the value of this variable at least one since the cache requires + // the number of counters to be greater than zero. + numInCache = 1 + } + + config := ristretto.Config{ + NumCounters: numInCache * 8, + MaxCost: opt.BlockCacheSize, + BufferItems: 64, + Metrics: true, + OnExit: table.BlockEvictHandler, + } + db.blockCache, err = ristretto.NewCache(&config) + if err != nil { + return nil, y.Wrap(err, "failed to create data cache") + } + } + + if opt.IndexCacheSize > 0 { + // Index size is around 5% of the table size. + indexSz := int64(float64(opt.MemTableSize) * 0.05) + numInCache := opt.IndexCacheSize / indexSz + if numInCache == 0 { + // Make the value of this variable at least one since the cache requires + // the number of counters to be greater than zero. + numInCache = 1 + } + + config := ristretto.Config{ + NumCounters: numInCache * 8, + MaxCost: opt.IndexCacheSize, + BufferItems: 64, + Metrics: true, + } + db.indexCache, err = ristretto.NewCache(&config) + if err != nil { + return nil, y.Wrap(err, "failed to create bf cache") + } + } + + db.closers.cacheHealth = z.NewCloser(1) + go db.monitorCache(db.closers.cacheHealth) + + if db.opt.InMemory { + db.opt.SyncWrites = false + // If badger is running in memory mode, push everything into the LSM Tree. + db.opt.ValueThreshold = math.MaxInt32 + } + krOpt := KeyRegistryOptions{ + ReadOnly: opt.ReadOnly, + Dir: opt.Dir, + EncryptionKey: opt.EncryptionKey, + EncryptionKeyRotationDuration: opt.EncryptionKeyRotationDuration, + InMemory: opt.InMemory, + } + + if db.registry, err = OpenKeyRegistry(krOpt); err != nil { + return db, err + } + db.calculateSize() + db.closers.updateSize = z.NewCloser(1) + go db.updateSize(db.closers.updateSize) + + if err := db.openMemTables(db.opt); err != nil { + return nil, y.Wrapf(err, "while opening memtables") + } + + if !db.opt.ReadOnly { + if db.mt, err = db.newMemTable(); err != nil { + return nil, y.Wrapf(err, "cannot create memtable") + } + } + + // newLevelsController potentially loads files in directory. + if db.lc, err = newLevelsController(db, &manifest); err != nil { + return db, err + } + + // Initialize vlog struct. + db.vlog.init(db) + + if !opt.ReadOnly { + db.closers.compactors = z.NewCloser(1) + db.lc.startCompact(db.closers.compactors) + + db.closers.memtable = z.NewCloser(1) + go func() { + db.flushMemtable(db.closers.memtable) // Need levels controller to be up. + }() + // Flush them to disk asap. + for _, mt := range db.imm { + db.flushChan <- mt + } + } + // We do increment nextTxnTs below. So, no need to do it here. + db.orc.nextTxnTs = db.MaxVersion() + db.opt.Infof("Set nextTxnTs to %d", db.orc.nextTxnTs) + + if err = db.vlog.open(db); err != nil { + return db, y.Wrapf(err, "During db.vlog.open") + } + + // Let's advance nextTxnTs to one more than whatever we observed via + // replaying the logs. + db.orc.txnMark.Done(db.orc.nextTxnTs) + // In normal mode, we must update readMark so older versions of keys can be removed during + // compaction when run in offline mode via the flatten tool. + db.orc.readMark.Done(db.orc.nextTxnTs) + db.orc.incrementNextTs() + + go db.threshold.listenForValueThresholdUpdate() + + if err := db.initBannedNamespaces(); err != nil { + return db, errors.Wrapf(err, "While setting banned keys") + } + + db.closers.writes = z.NewCloser(1) + go db.doWrites(db.closers.writes) + + if !db.opt.InMemory { + db.closers.valueGC = z.NewCloser(1) + go db.vlog.waitOnGC(db.closers.valueGC) + } + + db.closers.pub = z.NewCloser(1) + go db.pub.listenForUpdates(db.closers.pub) + + valueDirLockGuard = nil + dirLockGuard = nil + manifestFile = nil + return db, nil +} + +// initBannedNamespaces retrieves the banned namepsaces from the DB and updates in-memory structure. +func (db *DB) initBannedNamespaces() error { + if db.opt.NamespaceOffset < 0 { + return nil + } + return db.View(func(txn *Txn) error { + iopts := DefaultIteratorOptions + iopts.Prefix = bannedNsKey + iopts.PrefetchValues = false + iopts.InternalAccess = true + itr := txn.NewIterator(iopts) + defer itr.Close() + for itr.Rewind(); itr.Valid(); itr.Next() { + key := y.BytesToU64(itr.Item().Key()[len(bannedNsKey):]) + db.bannedNamespaces.add(key) + } + return nil + }) +} + +func (db *DB) MaxVersion() uint64 { + var maxVersion uint64 + update := func(a uint64) { + if a > maxVersion { + maxVersion = a + } + } + db.lock.Lock() + // In read only mode, we do not create new mem table. + if !db.opt.ReadOnly { + update(db.mt.maxVersion) + } + for _, mt := range db.imm { + update(mt.maxVersion) + } + db.lock.Unlock() + for _, ti := range db.Tables() { + update(ti.MaxVersion) + } + return maxVersion +} + +func (db *DB) monitorCache(c *z.Closer) { + defer c.Done() + count := 0 + analyze := func(name string, metrics *ristretto.Metrics) { + // If the mean life expectancy is less than 10 seconds, the cache + // might be too small. + le := metrics.LifeExpectancySeconds() + if le == nil { + return + } + lifeTooShort := le.Count > 0 && float64(le.Sum)/float64(le.Count) < 10 + hitRatioTooLow := metrics.Ratio() > 0 && metrics.Ratio() < 0.4 + if lifeTooShort && hitRatioTooLow { + db.opt.Warningf("%s might be too small. Metrics: %s\n", name, metrics) + db.opt.Warningf("Cache life expectancy (in seconds): %+v\n", le) + + } else if le.Count > 1000 && count%5 == 0 { + db.opt.Infof("%s metrics: %s\n", name, metrics) + } + } + + ticker := time.NewTicker(1 * time.Minute) + defer ticker.Stop() + for { + select { + case <-c.HasBeenClosed(): + return + case <-ticker.C: + } + + analyze("Block cache", db.BlockCacheMetrics()) + analyze("Index cache", db.IndexCacheMetrics()) + count++ + } +} + +// cleanup stops all the goroutines started by badger. This is used in open to +// cleanup goroutines in case of an error. +func (db *DB) cleanup() { + db.stopMemoryFlush() + db.stopCompactions() + + db.blockCache.Close() + db.indexCache.Close() + if db.closers.updateSize != nil { + db.closers.updateSize.Signal() + } + if db.closers.valueGC != nil { + db.closers.valueGC.Signal() + } + if db.closers.writes != nil { + db.closers.writes.Signal() + } + if db.closers.pub != nil { + db.closers.pub.Signal() + } + + db.orc.Stop() + + // Do not use vlog.Close() here. vlog.Close truncates the files. We don't + // want to truncate files unless the user has specified the truncate flag. +} + +// BlockCacheMetrics returns the metrics for the underlying block cache. +func (db *DB) BlockCacheMetrics() *ristretto.Metrics { + if db.blockCache != nil { + return db.blockCache.Metrics + } + return nil +} + +// IndexCacheMetrics returns the metrics for the underlying index cache. +func (db *DB) IndexCacheMetrics() *ristretto.Metrics { + if db.indexCache != nil { + return db.indexCache.Metrics + } + return nil +} + +// Close closes a DB. It's crucial to call it to ensure all the pending updates make their way to +// disk. Calling DB.Close() multiple times would still only close the DB once. +func (db *DB) Close() error { + var err error + db.closeOnce.Do(func() { + err = db.close() + }) + return err +} + +// IsClosed denotes if the badger DB is closed or not. A DB instance should not +// be used after closing it. +func (db *DB) IsClosed() bool { + return db.isClosed.Load() == 1 +} + +func (db *DB) close() (err error) { + defer db.allocPool.Release() + + db.opt.Debugf("Closing database") + db.opt.Infof("Lifetime L0 stalled for: %s\n", time.Duration(db.lc.l0stallsMs.Load())) + + db.blockWrites.Store(1) + db.isClosed.Store(1) + + if !db.opt.InMemory { + // Stop value GC first. + db.closers.valueGC.SignalAndWait() + } + + // Stop writes next. + db.closers.writes.SignalAndWait() + + // Don't accept any more write. + close(db.writeCh) + + db.closers.pub.SignalAndWait() + db.closers.cacheHealth.Signal() + + // Make sure that block writer is done pushing stuff into memtable! + // Otherwise, you will have a race condition: we are trying to flush memtables + // and remove them completely, while the block / memtable writer is still + // trying to push stuff into the memtable. This will also resolve the value + // offset problem: as we push into memtable, we update value offsets there. + if db.mt != nil { + if db.mt.sl.Empty() { + // Remove the memtable if empty. + db.mt.DecrRef() + } else { + db.opt.Debugf("Flushing memtable") + for { + pushedMemTable := func() bool { + db.lock.Lock() + defer db.lock.Unlock() + y.AssertTrue(db.mt != nil) + select { + case db.flushChan <- db.mt: + db.imm = append(db.imm, db.mt) // Flusher will attempt to remove this from s.imm. + db.mt = nil // Will segfault if we try writing! + db.opt.Debugf("pushed to flush chan\n") + return true + default: + // If we fail to push, we need to unlock and wait for a short while. + // The flushing operation needs to update s.imm. Otherwise, we have a + // deadlock. + // TODO: Think about how to do this more cleanly, maybe without any locks. + } + return false + }() + if pushedMemTable { + break + } + time.Sleep(10 * time.Millisecond) + } + } + } + db.stopMemoryFlush() + db.stopCompactions() + + // Force Compact L0 + // We don't need to care about cstatus since no parallel compaction is running. + if db.opt.CompactL0OnClose { + err := db.lc.doCompact(173, compactionPriority{level: 0, score: 1.73}) + switch err { + case errFillTables: + // This error only means that there might be enough tables to do a compaction. So, we + // should not report it to the end user to avoid confusing them. + case nil: + db.opt.Debugf("Force compaction on level 0 done") + default: + db.opt.Warningf("While forcing compaction on level 0: %v", err) + } + } + + // Now close the value log. + if vlogErr := db.vlog.Close(); vlogErr != nil { + err = y.Wrap(vlogErr, "DB.Close") + } + + db.opt.Infof(db.LevelsToString()) + if lcErr := db.lc.close(); err == nil { + err = y.Wrap(lcErr, "DB.Close") + } + db.opt.Debugf("Waiting for closer") + db.closers.updateSize.SignalAndWait() + db.orc.Stop() + db.blockCache.Close() + db.indexCache.Close() + + db.threshold.close() + + if db.opt.InMemory { + return + } + + if db.dirLockGuard != nil { + if guardErr := db.dirLockGuard.release(); err == nil { + err = y.Wrap(guardErr, "DB.Close") + } + } + if db.valueDirGuard != nil { + if guardErr := db.valueDirGuard.release(); err == nil { + err = y.Wrap(guardErr, "DB.Close") + } + } + if manifestErr := db.manifest.close(); err == nil { + err = y.Wrap(manifestErr, "DB.Close") + } + if registryErr := db.registry.Close(); err == nil { + err = y.Wrap(registryErr, "DB.Close") + } + + // Fsync directories to ensure that lock file, and any other removed files whose directory + // we haven't specifically fsynced, are guaranteed to have their directory entry removal + // persisted to disk. + if syncErr := db.syncDir(db.opt.Dir); err == nil { + err = y.Wrap(syncErr, "DB.Close") + } + if syncErr := db.syncDir(db.opt.ValueDir); err == nil { + err = y.Wrap(syncErr, "DB.Close") + } + + return err +} + +// VerifyChecksum verifies checksum for all tables on all levels. +// This method can be used to verify checksum, if opt.ChecksumVerificationMode is NoVerification. +func (db *DB) VerifyChecksum() error { + return db.lc.verifyChecksum() +} + +const ( + lockFile = "LOCK" +) + +// Sync syncs database content to disk. This function provides +// more control to user to sync data whenever required. +func (db *DB) Sync() error { + /** + Make an attempt to sync both the logs, the active memtable's WAL and the vLog (1847). + Cases: + - All_ok :: If both the logs sync successfully. + + - Entry_Lost :: If an entry with a value pointer was present in the active memtable's WAL, + :: and the WAL was synced but there was an error in syncing the vLog. + :: The entry will be considered lost and this case will need to be handled during recovery. + + - Entries_Lost :: If there were errors in syncing both the logs, multiple entries would be lost. + + - Entries_Lost :: If the active memtable's WAL is not synced but the vLog is synced, it will + :: result in entries being lost because recovery of the active memtable is done from its WAL. + :: Check `UpdateSkipList` in memtable.go. + + - Nothing_lost :: If an entry with its value was present in the active memtable's WAL, and the WAL was synced, + :: but there was an error in syncing the vLog. + :: Nothing is lost for this very specific entry because the entry is completely present in the memtable's WAL. + + - Partially_lost :: If entries were written partially in either of the logs, + :: the logs will be truncated during recovery. + :: As a result of truncation, some entries might be lost. + :: Assume that 4KB of data is to be synced and invoking `Sync` results only in syncing 3KB + :: of data and then the machine shuts down or the disk failure happens, + :: this will result in partial writes. [[This case needs verification]] + */ + db.lock.RLock() + memtableSyncError := db.mt.SyncWAL() + db.lock.RUnlock() + + vLogSyncError := db.vlog.sync() + return y.CombineErrors(memtableSyncError, vLogSyncError) +} + +// getMemtables returns the current memtables and get references. +func (db *DB) getMemTables() ([]*memTable, func()) { + db.lock.RLock() + defer db.lock.RUnlock() + + var tables []*memTable + + // Mutable memtable does not exist in read-only mode. + if !db.opt.ReadOnly { + // Get mutable memtable. + tables = append(tables, db.mt) + db.mt.IncrRef() + } + + // Get immutable memtables. + last := len(db.imm) - 1 + for i := range db.imm { + tables = append(tables, db.imm[last-i]) + db.imm[last-i].IncrRef() + } + return tables, func() { + for _, tbl := range tables { + tbl.DecrRef() + } + } +} + +// get returns the value in memtable or disk for given key. +// Note that value will include meta byte. +// +// IMPORTANT: We should never write an entry with an older timestamp for the same key, We need to +// maintain this invariant to search for the latest value of a key, or else we need to search in all +// tables and find the max version among them. To maintain this invariant, we also need to ensure +// that all versions of a key are always present in the same table from level 1, because compaction +// can push any table down. +// +// Update(23/09/2020) - We have dropped the move key implementation. Earlier we +// were inserting move keys to fix the invalid value pointers but we no longer +// do that. For every get("fooX") call where X is the version, we will search +// for "fooX" in all the levels of the LSM tree. This is expensive but it +// removes the overhead of handling move keys completely. +func (db *DB) get(key []byte) (y.ValueStruct, error) { + if db.IsClosed() { + return y.ValueStruct{}, ErrDBClosed + } + tables, decr := db.getMemTables() // Lock should be released. + defer decr() + + var maxVs y.ValueStruct + version := y.ParseTs(key) + + y.NumGetsAdd(db.opt.MetricsEnabled, 1) + for i := 0; i < len(tables); i++ { + vs := tables[i].sl.Get(key) + y.NumMemtableGetsAdd(db.opt.MetricsEnabled, 1) + if vs.Meta == 0 && vs.Value == nil { + continue + } + // Found the required version of the key, return immediately. + if vs.Version == version { + y.NumGetsWithResultsAdd(db.opt.MetricsEnabled, 1) + return vs, nil + } + if maxVs.Version < vs.Version { + maxVs = vs + } + } + return db.lc.get(key, maxVs, 0) +} + +var requestPool = sync.Pool{ + New: func() interface{} { + return new(request) + }, +} + +func (db *DB) writeToLSM(b *request) error { + // We should check the length of b.Prts and b.Entries only when badger is not + // running in InMemory mode. In InMemory mode, we don't write anything to the + // value log and that's why the length of b.Ptrs will always be zero. + if !db.opt.InMemory && len(b.Ptrs) != len(b.Entries) { + return errors.Errorf("Ptrs and Entries don't match: %+v", b) + } + + for i, entry := range b.Entries { + var err error + if entry.skipVlogAndSetThreshold(db.valueThreshold()) { + // Will include deletion / tombstone case. + err = db.mt.Put(entry.Key, + y.ValueStruct{ + Value: entry.Value, + // Ensure value pointer flag is removed. Otherwise, the value will fail + // to be retrieved during iterator prefetch. `bitValuePointer` is only + // known to be set in write to LSM when the entry is loaded from a backup + // with lower ValueThreshold and its value was stored in the value log. + Meta: entry.meta &^ bitValuePointer, + UserMeta: entry.UserMeta, + ExpiresAt: entry.ExpiresAt, + }) + } else { + // Write pointer to Memtable. + err = db.mt.Put(entry.Key, + y.ValueStruct{ + Value: b.Ptrs[i].Encode(), + Meta: entry.meta | bitValuePointer, + UserMeta: entry.UserMeta, + ExpiresAt: entry.ExpiresAt, + }) + } + if err != nil { + return y.Wrapf(err, "while writing to memTable") + } + } + if db.opt.SyncWrites { + return db.mt.SyncWAL() + } + return nil +} + +// writeRequests is called serially by only one goroutine. +func (db *DB) writeRequests(reqs []*request) error { + if len(reqs) == 0 { + return nil + } + + done := func(err error) { + for _, r := range reqs { + r.Err = err + r.Wg.Done() + } + } + db.opt.Debugf("writeRequests called. Writing to value log") + err := db.vlog.write(reqs) + if err != nil { + done(err) + return err + } + + db.opt.Debugf("Writing to memtable") + var count int + for _, b := range reqs { + if len(b.Entries) == 0 { + continue + } + count += len(b.Entries) + var i uint64 + var err error + for err = db.ensureRoomForWrite(); err == errNoRoom; err = db.ensureRoomForWrite() { + i++ + if i%100 == 0 { + db.opt.Debugf("Making room for writes") + } + // We need to poll a bit because both hasRoomForWrite and the flusher need access to s.imm. + // When flushChan is full and you are blocked there, and the flusher is trying to update s.imm, + // you will get a deadlock. + time.Sleep(10 * time.Millisecond) + } + if err != nil { + done(err) + return y.Wrap(err, "writeRequests") + } + if err := db.writeToLSM(b); err != nil { + done(err) + return y.Wrap(err, "writeRequests") + } + } + + db.opt.Debugf("Sending updates to subscribers") + db.pub.sendUpdates(reqs) + + done(nil) + db.opt.Debugf("%d entries written", count) + return nil +} + +func (db *DB) sendToWriteCh(entries []*Entry) (*request, error) { + if db.blockWrites.Load() == 1 { + return nil, ErrBlockedWrites + } + var count, size int64 + for _, e := range entries { + size += e.estimateSizeAndSetThreshold(db.valueThreshold()) + count++ + } + y.NumBytesWrittenUserAdd(db.opt.MetricsEnabled, size) + if count >= db.opt.maxBatchCount || size >= db.opt.maxBatchSize { + return nil, ErrTxnTooBig + } + + // We can only service one request because we need each txn to be stored in a contigous section. + // Txns should not interleave among other txns or rewrites. + req := requestPool.Get().(*request) + req.reset() + req.Entries = entries + req.Wg.Add(1) + req.IncrRef() // for db write + db.writeCh <- req // Handled in doWrites. + y.NumPutsAdd(db.opt.MetricsEnabled, int64(len(entries))) + + return req, nil +} + +func (db *DB) doWrites(lc *z.Closer) { + defer lc.Done() + pendingCh := make(chan struct{}, 1) + + writeRequests := func(reqs []*request) { + if err := db.writeRequests(reqs); err != nil { + db.opt.Errorf("writeRequests: %v", err) + } + <-pendingCh + } + + // This variable tracks the number of pending writes. + reqLen := new(expvar.Int) + y.PendingWritesSet(db.opt.MetricsEnabled, db.opt.Dir, reqLen) + + reqs := make([]*request, 0, 10) + for { + var r *request + select { + case r = <-db.writeCh: + case <-lc.HasBeenClosed(): + goto closedCase + } + + for { + reqs = append(reqs, r) + reqLen.Set(int64(len(reqs))) + + if len(reqs) >= 3*kvWriteChCapacity { + pendingCh <- struct{}{} // blocking. + goto writeCase + } + + select { + // Either push to pending, or continue to pick from writeCh. + case r = <-db.writeCh: + case pendingCh <- struct{}{}: + goto writeCase + case <-lc.HasBeenClosed(): + goto closedCase + } + } + + closedCase: + // All the pending request are drained. + // Don't close the writeCh, because it has be used in several places. + for { + select { + case r = <-db.writeCh: + reqs = append(reqs, r) + default: + pendingCh <- struct{}{} // Push to pending before doing a write. + writeRequests(reqs) + return + } + } + + writeCase: + go writeRequests(reqs) + reqs = make([]*request, 0, 10) + reqLen.Set(0) + } +} + +// batchSet applies a list of badger.Entry. If a request level error occurs it +// will be returned. +// +// Check(kv.BatchSet(entries)) +func (db *DB) batchSet(entries []*Entry) error { + req, err := db.sendToWriteCh(entries) + if err != nil { + return err + } + + return req.Wait() +} + +// batchSetAsync is the asynchronous version of batchSet. It accepts a callback +// function which is called when all the sets are complete. If a request level +// error occurs, it will be passed back via the callback. +// +// err := kv.BatchSetAsync(entries, func(err error)) { +// Check(err) +// } +func (db *DB) batchSetAsync(entries []*Entry, f func(error)) error { + req, err := db.sendToWriteCh(entries) + if err != nil { + return err + } + go func() { + err := req.Wait() + // Write is complete. Let's call the callback function now. + f(err) + }() + return nil +} + +var errNoRoom = errors.New("No room for write") + +// ensureRoomForWrite is always called serially. +func (db *DB) ensureRoomForWrite() error { + var err error + db.lock.Lock() + defer db.lock.Unlock() + + y.AssertTrue(db.mt != nil) // A nil mt indicates that DB is being closed. + if !db.mt.isFull() { + return nil + } + + select { + case db.flushChan <- db.mt: + db.opt.Debugf("Flushing memtable, mt.size=%d size of flushChan: %d\n", + db.mt.sl.MemSize(), len(db.flushChan)) + // We manage to push this task. Let's modify imm. + db.imm = append(db.imm, db.mt) + db.mt, err = db.newMemTable() + if err != nil { + return y.Wrapf(err, "cannot create new mem table") + } + // New memtable is empty. We certainly have room. + return nil + default: + // We need to do this to unlock and allow the flusher to modify imm. + return errNoRoom + } +} + +func arenaSize(opt Options) int64 { + return opt.MemTableSize + opt.maxBatchSize + opt.maxBatchCount*int64(skl.MaxNodeSize) +} + +// buildL0Table builds a new table from the memtable. +func buildL0Table(iter y.Iterator, dropPrefixes [][]byte, bopts table.Options) *table.Builder { + defer iter.Close() + + b := table.NewTableBuilder(bopts) + for iter.Rewind(); iter.Valid(); iter.Next() { + if len(dropPrefixes) > 0 && hasAnyPrefixes(iter.Key(), dropPrefixes) { + continue + } + vs := iter.Value() + var vp valuePointer + if vs.Meta&bitValuePointer > 0 { + vp.Decode(vs.Value) + } + b.Add(iter.Key(), iter.Value(), vp.Len) + } + + return b +} + +// handleMemTableFlush must be run serially. +func (db *DB) handleMemTableFlush(mt *memTable, dropPrefixes [][]byte) error { + bopts := buildTableOptions(db) + itr := mt.sl.NewUniIterator(false) + builder := buildL0Table(itr, nil, bopts) + defer builder.Close() + + // buildL0Table can return nil if the none of the items in the skiplist are + // added to the builder. This can happen when drop prefix is set and all + // the items are skipped. + if builder.Empty() { + builder.Finish() + return nil + } + + fileID := db.lc.reserveFileID() + var tbl *table.Table + var err error + if db.opt.InMemory { + data := builder.Finish() + tbl, err = table.OpenInMemoryTable(data, fileID, &bopts) + } else { + tbl, err = table.CreateTable(table.NewFilename(fileID, db.opt.Dir), builder) + } + if err != nil { + return y.Wrap(err, "error while creating table") + } + // We own a ref on tbl. + err = db.lc.addLevel0Table(tbl) // This will incrRef + _ = tbl.DecrRef() // Releases our ref. + return err +} + +// flushMemtable must keep running until we send it an empty memtable. If there +// are errors during handling the memtable flush, we'll retry indefinitely. +func (db *DB) flushMemtable(lc *z.Closer) { + defer lc.Done() + + for mt := range db.flushChan { + if mt == nil { + continue + } + + for { + if err := db.handleMemTableFlush(mt, nil); err != nil { + // Encountered error. Retry indefinitely. + db.opt.Errorf("error flushing memtable to disk: %v, retrying", err) + time.Sleep(time.Second) + continue + } + + // Update s.imm. Need a lock. + db.lock.Lock() + // This is a single-threaded operation. mt corresponds to the head of + // db.imm list. Once we flush it, we advance db.imm. The next mt + // which would arrive here would match db.imm[0], because we acquire a + // lock over DB when pushing to flushChan. + // TODO: This logic is dirty AF. Any change and this could easily break. + y.AssertTrue(mt == db.imm[0]) + db.imm = db.imm[1:] + mt.DecrRef() // Return memory. + // unlock + db.lock.Unlock() + break + } + } +} + +func exists(path string) (bool, error) { + _, err := os.Stat(path) + if err == nil { + return true, nil + } + if os.IsNotExist(err) { + return false, nil + } + return true, err +} + +// This function does a filewalk, calculates the size of vlog and sst files and stores it in +// y.LSMSize and y.VlogSize. +func (db *DB) calculateSize() { + if db.opt.InMemory { + return + } + newInt := func(val int64) *expvar.Int { + v := new(expvar.Int) + v.Add(val) + return v + } + + totalSize := func(dir string) (int64, int64) { + var lsmSize, vlogSize int64 + err := filepath.Walk(dir, func(path string, info os.FileInfo, err error) error { + if err != nil { + return err + } + ext := filepath.Ext(path) + switch ext { + case ".sst": + lsmSize += info.Size() + case ".vlog": + vlogSize += info.Size() + } + return nil + }) + if err != nil { + db.opt.Debugf("Got error while calculating total size of directory: %s", dir) + } + return lsmSize, vlogSize + } + + lsmSize, vlogSize := totalSize(db.opt.Dir) + y.LSMSizeSet(db.opt.MetricsEnabled, db.opt.Dir, newInt(lsmSize)) + // If valueDir is different from dir, we'd have to do another walk. + if db.opt.ValueDir != db.opt.Dir { + _, vlogSize = totalSize(db.opt.ValueDir) + } + y.VlogSizeSet(db.opt.MetricsEnabled, db.opt.ValueDir, newInt(vlogSize)) +} + +func (db *DB) updateSize(lc *z.Closer) { + defer lc.Done() + if db.opt.InMemory { + return + } + + metricsTicker := time.NewTicker(time.Minute) + defer metricsTicker.Stop() + + for { + select { + case <-metricsTicker.C: + db.calculateSize() + case <-lc.HasBeenClosed(): + return + } + } +} + +// RunValueLogGC triggers a value log garbage collection. +// +// It picks value log files to perform GC based on statistics that are collected +// during compactions. If no such statistics are available, then log files are +// picked in random order. The process stops as soon as the first log file is +// encountered which does not result in garbage collection. +// +// When a log file is picked, it is first sampled. If the sample shows that we +// can discard at least discardRatio space of that file, it would be rewritten. +// +// If a call to RunValueLogGC results in no rewrites, then an ErrNoRewrite is +// thrown indicating that the call resulted in no file rewrites. +// +// We recommend setting discardRatio to 0.5, thus indicating that a file be +// rewritten if half the space can be discarded. This results in a lifetime +// value log write amplification of 2 (1 from original write + 0.5 rewrite + +// 0.25 + 0.125 + ... = 2). Setting it to higher value would result in fewer +// space reclaims, while setting it to a lower value would result in more space +// reclaims at the cost of increased activity on the LSM tree. discardRatio +// must be in the range (0.0, 1.0), both endpoints excluded, otherwise an +// ErrInvalidRequest is returned. +// +// Only one GC is allowed at a time. If another value log GC is running, or DB +// has been closed, this would return an ErrRejected. +// +// Note: Every time GC is run, it would produce a spike of activity on the LSM +// tree. +func (db *DB) RunValueLogGC(discardRatio float64) error { + if db.opt.InMemory { + return ErrGCInMemoryMode + } + if discardRatio >= 1.0 || discardRatio <= 0.0 { + return ErrInvalidRequest + } + + // Pick a log file and run GC + return db.vlog.runGC(discardRatio) +} + +// Size returns the size of lsm and value log files in bytes. It can be used to decide how often to +// call RunValueLogGC. +func (db *DB) Size() (lsm, vlog int64) { + if y.LSMSizeGet(db.opt.MetricsEnabled, db.opt.Dir) == nil { + lsm, vlog = 0, 0 + return + } + lsm = y.LSMSizeGet(db.opt.MetricsEnabled, db.opt.Dir).(*expvar.Int).Value() + vlog = y.VlogSizeGet(db.opt.MetricsEnabled, db.opt.ValueDir).(*expvar.Int).Value() + return +} + +// Sequence represents a Badger sequence. +type Sequence struct { + lock sync.Mutex + db *DB + key []byte + next uint64 + leased uint64 + bandwidth uint64 +} + +// Next would return the next integer in the sequence, updating the lease by running a transaction +// if needed. +func (seq *Sequence) Next() (uint64, error) { + seq.lock.Lock() + defer seq.lock.Unlock() + if seq.next >= seq.leased { + if err := seq.updateLease(); err != nil { + return 0, err + } + } + val := seq.next + seq.next++ + return val, nil +} + +// Release the leased sequence to avoid wasted integers. This should be done right +// before closing the associated DB. However it is valid to use the sequence after +// it was released, causing a new lease with full bandwidth. +func (seq *Sequence) Release() error { + seq.lock.Lock() + defer seq.lock.Unlock() + err := seq.db.Update(func(txn *Txn) error { + item, err := txn.Get(seq.key) + if err != nil { + return err + } + + var num uint64 + if err := item.Value(func(v []byte) error { + num = binary.BigEndian.Uint64(v) + return nil + }); err != nil { + return err + } + + if num == seq.leased { + var buf [8]byte + binary.BigEndian.PutUint64(buf[:], seq.next) + return txn.SetEntry(NewEntry(seq.key, buf[:])) + } + + return nil + }) + if err != nil { + return err + } + seq.leased = seq.next + return nil +} + +func (seq *Sequence) updateLease() error { + return seq.db.Update(func(txn *Txn) error { + item, err := txn.Get(seq.key) + switch { + case err == ErrKeyNotFound: + seq.next = 0 + case err != nil: + return err + default: + var num uint64 + if err := item.Value(func(v []byte) error { + num = binary.BigEndian.Uint64(v) + return nil + }); err != nil { + return err + } + seq.next = num + } + + lease := seq.next + seq.bandwidth + var buf [8]byte + binary.BigEndian.PutUint64(buf[:], lease) + if err = txn.SetEntry(NewEntry(seq.key, buf[:])); err != nil { + return err + } + seq.leased = lease + return nil + }) +} + +// GetSequence would initiate a new sequence object, generating it from the stored lease, if +// available, in the database. Sequence can be used to get a list of monotonically increasing +// integers. Multiple sequences can be created by providing different keys. Bandwidth sets the +// size of the lease, determining how many Next() requests can be served from memory. +// +// GetSequence is not supported on ManagedDB. Calling this would result in a panic. +func (db *DB) GetSequence(key []byte, bandwidth uint64) (*Sequence, error) { + if db.opt.managedTxns { + panic("Cannot use GetSequence with managedDB=true.") + } + + switch { + case len(key) == 0: + return nil, ErrEmptyKey + case bandwidth == 0: + return nil, ErrZeroBandwidth + } + seq := &Sequence{ + db: db, + key: key, + next: 0, + leased: 0, + bandwidth: bandwidth, + } + err := seq.updateLease() + return seq, err +} + +// Tables gets the TableInfo objects from the level controller. If withKeysCount +// is true, TableInfo objects also contain counts of keys for the tables. +func (db *DB) Tables() []TableInfo { + return db.lc.getTableInfo() +} + +// Levels gets the LevelInfo. +func (db *DB) Levels() []LevelInfo { + return db.lc.getLevelInfo() +} + +// EstimateSize can be used to get rough estimate of data size for a given prefix. +func (db *DB) EstimateSize(prefix []byte) (uint64, uint64) { + var onDiskSize, uncompressedSize uint64 + tables := db.Tables() + for _, ti := range tables { + if bytes.HasPrefix(ti.Left, prefix) && bytes.HasPrefix(ti.Right, prefix) { + onDiskSize += uint64(ti.OnDiskSize) + uncompressedSize += uint64(ti.UncompressedSize) + } + } + return onDiskSize, uncompressedSize +} + +// Ranges can be used to get rough key ranges to divide up iteration over the DB. The ranges here +// would consider the prefix, but would not necessarily start or end with the prefix. In fact, the +// first range would have nil as left key, and the last range would have nil as the right key. +func (db *DB) Ranges(prefix []byte, numRanges int) []*keyRange { + var splits []string + tables := db.Tables() + + // We just want table ranges here and not keys count. + for _, ti := range tables { + // We don't use ti.Left, because that has a tendency to store !badger keys. Skip over tables + // at upper levels. Only choose tables from the last level. + if ti.Level != db.opt.MaxLevels-1 { + continue + } + if bytes.HasPrefix(ti.Right, prefix) { + splits = append(splits, string(ti.Right)) + } + } + + // If the number of splits is low, look at the offsets inside the + // tables to generate more splits. + if len(splits) < 32 { + numTables := len(tables) + if numTables == 0 { + numTables = 1 + } + numPerTable := 32 / numTables + if numPerTable == 0 { + numPerTable = 1 + } + splits = db.lc.keySplits(numPerTable, prefix) + } + + // If the number of splits is still < 32, then look at the memtables. + if len(splits) < 32 { + maxPerSplit := 10000 + mtSplits := func(mt *memTable) { + if mt == nil { + return + } + count := 0 + iter := mt.sl.NewIterator() + for iter.SeekToFirst(); iter.Valid(); iter.Next() { + if count%maxPerSplit == 0 { + // Add a split every maxPerSplit keys. + if bytes.HasPrefix(iter.Key(), prefix) { + splits = append(splits, string(iter.Key())) + } + } + count += 1 + } + _ = iter.Close() + } + + db.lock.Lock() + defer db.lock.Unlock() + var memTables []*memTable + memTables = append(memTables, db.imm...) + for _, mt := range memTables { + mtSplits(mt) + } + mtSplits(db.mt) + } + + // We have our splits now. Let's convert them to ranges. + sort.Strings(splits) + var ranges []*keyRange + var start []byte + for _, key := range splits { + ranges = append(ranges, &keyRange{left: start, right: y.SafeCopy(nil, []byte(key))}) + start = y.SafeCopy(nil, []byte(key)) + } + ranges = append(ranges, &keyRange{left: start}) + + // Figure out the approximate table size this range has to deal with. + for _, t := range tables { + tr := keyRange{left: t.Left, right: t.Right} + for _, r := range ranges { + if len(r.left) == 0 || len(r.right) == 0 { + continue + } + if r.overlapsWith(tr) { + r.size += int64(t.UncompressedSize) + } + } + } + + var total int64 + for _, r := range ranges { + total += r.size + } + if total == 0 { + return ranges + } + // Figure out the average size, so we know how to bin the ranges together. + avg := total / int64(numRanges) + + var out []*keyRange + var i int + for i < len(ranges) { + r := ranges[i] + cur := &keyRange{left: r.left, size: r.size, right: r.right} + i++ + for ; i < len(ranges); i++ { + next := ranges[i] + if cur.size+next.size > avg { + break + } + cur.right = next.right + cur.size += next.size + } + out = append(out, cur) + } + return out +} + +// MaxBatchCount returns max possible entries in batch +func (db *DB) MaxBatchCount() int64 { + return db.opt.maxBatchCount +} + +// MaxBatchSize returns max possible batch size +func (db *DB) MaxBatchSize() int64 { + return db.opt.maxBatchSize +} + +func (db *DB) stopMemoryFlush() { + // Stop memtable flushes. + if db.closers.memtable != nil { + close(db.flushChan) + db.closers.memtable.SignalAndWait() + } +} + +func (db *DB) stopCompactions() { + // Stop compactions. + if db.closers.compactors != nil { + db.closers.compactors.SignalAndWait() + } +} + +func (db *DB) startCompactions() { + // Resume compactions. + if db.closers.compactors != nil { + db.closers.compactors = z.NewCloser(1) + db.lc.startCompact(db.closers.compactors) + } +} + +func (db *DB) startMemoryFlush() { + // Start memory fluhser. + if db.closers.memtable != nil { + db.flushChan = make(chan *memTable, db.opt.NumMemtables) + db.closers.memtable = z.NewCloser(1) + go func() { + db.flushMemtable(db.closers.memtable) + }() + } +} + +// Flatten can be used to force compactions on the LSM tree so all the tables fall on the same +// level. This ensures that all the versions of keys are colocated and not split across multiple +// levels, which is necessary after a restore from backup. During Flatten, live compactions are +// stopped. Ideally, no writes are going on during Flatten. Otherwise, it would create competition +// between flattening the tree and new tables being created at level zero. +func (db *DB) Flatten(workers int) error { + + db.stopCompactions() + defer db.startCompactions() + + compactAway := func(cp compactionPriority) error { + db.opt.Infof("Attempting to compact with %+v\n", cp) + errCh := make(chan error, 1) + for i := 0; i < workers; i++ { + go func() { + errCh <- db.lc.doCompact(175, cp) + }() + } + var success int + var rerr error + for i := 0; i < workers; i++ { + err := <-errCh + if err != nil { + rerr = err + db.opt.Warningf("While running doCompact with %+v. Error: %v\n", cp, err) + } else { + success++ + } + } + if success == 0 { + return rerr + } + // We could do at least one successful compaction. So, we'll consider this a success. + db.opt.Infof("%d compactor(s) succeeded. One or more tables from level %d compacted.\n", + success, cp.level) + return nil + } + + hbytes := func(sz int64) string { + return humanize.IBytes(uint64(sz)) + } + + t := db.lc.levelTargets() + for { + db.opt.Infof("\n") + var levels []int + for i, l := range db.lc.levels { + sz := l.getTotalSize() + db.opt.Infof("Level: %d. %8s Size. %8s Max.\n", + i, hbytes(l.getTotalSize()), hbytes(t.targetSz[i])) + if sz > 0 { + levels = append(levels, i) + } + } + if len(levels) <= 1 { + prios := db.lc.pickCompactLevels() + if len(prios) == 0 || prios[0].score <= 1.0 { + db.opt.Infof("All tables consolidated into one level. Flattening done.\n") + return nil + } + if err := compactAway(prios[0]); err != nil { + return err + } + continue + } + // Create an artificial compaction priority, to ensure that we compact the level. + cp := compactionPriority{level: levels[0], score: 1.71} + if err := compactAway(cp); err != nil { + return err + } + } +} + +func (db *DB) blockWrite() error { + // Stop accepting new writes. + if !db.blockWrites.CompareAndSwap(0, 1) { + return ErrBlockedWrites + } + + // Make all pending writes finish. The following will also close writeCh. + db.closers.writes.SignalAndWait() + db.opt.Infof("Writes flushed. Stopping compactions now...") + return nil +} + +func (db *DB) unblockWrite() { + db.closers.writes = z.NewCloser(1) + go db.doWrites(db.closers.writes) + + // Resume writes. + db.blockWrites.Store(0) +} + +func (db *DB) prepareToDrop() (func(), error) { + if db.opt.ReadOnly { + panic("Attempting to drop data in read-only mode.") + } + // In order prepare for drop, we need to block the incoming writes and + // write it to db. Then, flush all the pending memtable. So that, we + // don't miss any entries. + if err := db.blockWrite(); err != nil { + return func() {}, err + } + reqs := make([]*request, 0, 10) + for { + select { + case r := <-db.writeCh: + reqs = append(reqs, r) + default: + if err := db.writeRequests(reqs); err != nil { + db.opt.Errorf("writeRequests: %v", err) + } + db.stopMemoryFlush() + return func() { + db.opt.Infof("Resuming writes") + db.startMemoryFlush() + db.unblockWrite() + }, nil + } + } +} + +// DropAll would drop all the data stored in Badger. It does this in the following way. +// - Stop accepting new writes. +// - Pause memtable flushes and compactions. +// - Pick all tables from all levels, create a changeset to delete all these +// tables and apply it to manifest. +// - Pick all log files from value log, and delete all of them. Restart value log files from zero. +// - Resume memtable flushes and compactions. +// +// NOTE: DropAll is resilient to concurrent writes, but not to reads. It is up to the user to not do +// any reads while DropAll is going on, otherwise they may result in panics. Ideally, both reads and +// writes are paused before running DropAll, and resumed after it is finished. +func (db *DB) DropAll() error { + f, err := db.dropAll() + if f != nil { + f() + } + return err +} + +func (db *DB) dropAll() (func(), error) { + db.opt.Infof("DropAll called. Blocking writes...") + f, err := db.prepareToDrop() + if err != nil { + return f, err + } + // prepareToDrop will stop all the incomming write and flushes any pending memtables. + // Before we drop, we'll stop the compaction because anyways all the datas are going to + // be deleted. + db.stopCompactions() + resume := func() { + db.startCompactions() + f() + } + // Block all foreign interactions with memory tables. + db.lock.Lock() + defer db.lock.Unlock() + + // Remove inmemory tables. Calling DecrRef for safety. Not sure if they're absolutely needed. + db.mt.DecrRef() + for _, mt := range db.imm { + mt.DecrRef() + } + db.imm = db.imm[:0] + db.mt, err = db.newMemTable() // Set it up for future writes. + if err != nil { + return resume, y.Wrapf(err, "cannot open new memtable") + } + + num, err := db.lc.dropTree() + if err != nil { + return resume, err + } + db.opt.Infof("Deleted %d SSTables. Now deleting value logs...\n", num) + + num, err = db.vlog.dropAll() + if err != nil { + return resume, err + } + db.lc.nextFileID.Store(1) + db.opt.Infof("Deleted %d value log files. DropAll done.\n", num) + db.blockCache.Clear() + db.indexCache.Clear() + db.threshold.Clear(db.opt) + return resume, nil +} + +// DropPrefix would drop all the keys with the provided prefix. It does this in the following way: +// - Stop accepting new writes. +// - Stop memtable flushes before acquiring lock. Because we're acquring lock here +// and memtable flush stalls for lock, which leads to deadlock +// - Flush out all memtables, skipping over keys with the given prefix, Kp. +// - Write out the value log header to memtables when flushing, so we don't accidentally bring Kp +// back after a restart. +// - Stop compaction. +// - Compact L0->L1, skipping over Kp. +// - Compact rest of the levels, Li->Li, picking tables which have Kp. +// - Resume memtable flushes, compactions and writes. +func (db *DB) DropPrefix(prefixes ...[]byte) error { + if len(prefixes) == 0 { + return nil + } + db.opt.Infof("DropPrefix called for %s", prefixes) + f, err := db.prepareToDrop() + if err != nil { + return err + } + defer f() + + var filtered [][]byte + if filtered, err = db.filterPrefixesToDrop(prefixes); err != nil { + return err + } + // If there is no prefix for which the data already exist, do not do anything. + if len(filtered) == 0 { + db.opt.Infof("No prefixes to drop") + return nil + } + // Block all foreign interactions with memory tables. + db.lock.Lock() + defer db.lock.Unlock() + + db.imm = append(db.imm, db.mt) + for _, memtable := range db.imm { + if memtable.sl.Empty() { + memtable.DecrRef() + continue + } + db.opt.Debugf("Flushing memtable") + if err := db.handleMemTableFlush(memtable, filtered); err != nil { + db.opt.Errorf("While trying to flush memtable: %v", err) + return err + } + memtable.DecrRef() + } + db.stopCompactions() + defer db.startCompactions() + db.imm = db.imm[:0] + db.mt, err = db.newMemTable() + if err != nil { + return y.Wrapf(err, "cannot create new mem table") + } + + // Drop prefixes from the levels. + if err := db.lc.dropPrefixes(filtered); err != nil { + return err + } + db.opt.Infof("DropPrefix done") + return nil +} + +func (db *DB) filterPrefixesToDrop(prefixes [][]byte) ([][]byte, error) { + var filtered [][]byte + for _, prefix := range prefixes { + err := db.View(func(txn *Txn) error { + iopts := DefaultIteratorOptions + iopts.Prefix = prefix + iopts.PrefetchValues = false + itr := txn.NewIterator(iopts) + defer itr.Close() + itr.Rewind() + if itr.ValidForPrefix(prefix) { + filtered = append(filtered, prefix) + } + return nil + }) + if err != nil { + return filtered, err + } + } + return filtered, nil +} + +// Checks if the key is banned. Returns the respective error if the key belongs to any of the banned +// namepspaces. Else it returns nil. +func (db *DB) isBanned(key []byte) error { + if db.opt.NamespaceOffset < 0 { + return nil + } + if len(key) <= db.opt.NamespaceOffset+8 { + return nil + } + if db.bannedNamespaces.has(y.BytesToU64(key[db.opt.NamespaceOffset:])) { + return ErrBannedKey + } + return nil +} + +// BanNamespace bans a namespace. Read/write to keys belonging to any of such namespace is denied. +func (db *DB) BanNamespace(ns uint64) error { + if db.opt.NamespaceOffset < 0 { + return ErrNamespaceMode + } + db.opt.Infof("Banning namespace: %d", ns) + // First set the banned namespaces in DB and then update the in-memory structure. + key := y.KeyWithTs(append(bannedNsKey, y.U64ToBytes(ns)...), 1) + entry := []*Entry{{ + Key: key, + Value: nil, + }} + req, err := db.sendToWriteCh(entry) + if err != nil { + return err + } + if err := req.Wait(); err != nil { + return err + } + db.bannedNamespaces.add(ns) + return nil +} + +// BannedNamespaces returns the list of prefixes banned for DB. +func (db *DB) BannedNamespaces() []uint64 { + return db.bannedNamespaces.all() +} + +// KVList contains a list of key-value pairs. +type KVList = pb.KVList + +// Subscribe can be used to watch key changes for the given key prefixes and the ignore string. +// At least one prefix should be passed, or an error will be returned. +// You can use an empty prefix to monitor all changes to the DB. +// Ignore string is the byte ranges for which prefix matching will be ignored. +// For example: ignore = "2-3", and prefix = "abc" will match for keys "abxxc", "abdfc" etc. +// This function blocks until the given context is done or an error occurs. +// The given function will be called with a new KVList containing the modified keys and the +// corresponding values. +func (db *DB) Subscribe(ctx context.Context, cb func(kv *KVList) error, matches []pb.Match) error { + if cb == nil { + return ErrNilCallback + } + + c := z.NewCloser(1) + s, err := db.pub.newSubscriber(c, matches) + if err != nil { + return y.Wrapf(err, "while creating a new subscriber") + } + slurp := func(batch *pb.KVList) error { + for { + select { + case kvs := <-s.sendCh: + batch.Kv = append(batch.Kv, kvs.Kv...) + default: + if len(batch.GetKv()) > 0 { + return cb(batch) + } + return nil + } + } + } + + drain := func() { + for { + select { + case _, ok := <-s.sendCh: + if !ok { + // Channel is closed. + return + } + default: + return + } + } + } + for { + select { + case <-c.HasBeenClosed(): + // No need to delete here. Closer will be called only while + // closing DB. Subscriber will be deleted by cleanSubscribers. + err := slurp(new(pb.KVList)) + // Drain if any pending updates. + c.Done() + return err + case <-ctx.Done(): + c.Done() + s.active.Store(0) + drain() + db.pub.deleteSubscriber(s.id) + // Delete the subscriber to avoid further updates. + return ctx.Err() + case batch := <-s.sendCh: + err := slurp(batch) + if err != nil { + c.Done() + s.active.Store(0) + drain() + // Delete the subscriber if there is an error by the callback. + db.pub.deleteSubscriber(s.id) + return err + } + } + } +} + +func (db *DB) syncDir(dir string) error { + if db.opt.InMemory { + return nil + } + return syncDir(dir) +} + +func createDirs(opt Options) error { + for _, path := range []string{opt.Dir, opt.ValueDir} { + dirExists, err := exists(path) + if err != nil { + return y.Wrapf(err, "Invalid Dir: %q", path) + } + if !dirExists { + if opt.ReadOnly { + return errors.Errorf("Cannot find directory %q for read-only open", path) + } + // Try to create the directory + err = os.MkdirAll(path, 0700) + if err != nil { + return y.Wrapf(err, "Error Creating Dir: %q", path) + } + } + } + return nil +} + +// Stream the contents of this DB to a new DB with options outOptions that will be +// created in outDir. +func (db *DB) StreamDB(outOptions Options) error { + outDir := outOptions.Dir + + // Open output DB. + outDB, err := OpenManaged(outOptions) + if err != nil { + return y.Wrapf(err, "cannot open out DB at %s", outDir) + } + defer outDB.Close() + writer := outDB.NewStreamWriter() + if err := writer.Prepare(); err != nil { + return y.Wrapf(err, "cannot create stream writer in out DB at %s", outDir) + } + + // Stream contents of DB to the output DB. + stream := db.NewStreamAt(math.MaxUint64) + stream.LogPrefix = fmt.Sprintf("Streaming DB to new DB at %s", outDir) + + stream.Send = func(buf *z.Buffer) error { + return writer.Write(buf) + } + if err := stream.Orchestrate(context.Background()); err != nil { + return y.Wrapf(err, "cannot stream DB to out DB at %s", outDir) + } + if err := writer.Flush(); err != nil { + return y.Wrapf(err, "cannot flush writer") + } + return nil +} + +// Opts returns a copy of the DB options. +func (db *DB) Opts() Options { + return db.opt +} + +type CacheType int + +const ( + BlockCache CacheType = iota + IndexCache +) + +// CacheMaxCost updates the max cost of the given cache (either block or index cache). +// The call will have an effect only if the DB was created with the cache. Otherwise it is +// a no-op. If you pass a negative value, the function will return the current value +// without updating it. +func (db *DB) CacheMaxCost(cache CacheType, maxCost int64) (int64, error) { + if db == nil { + return 0, nil + } + + if maxCost < 0 { + switch cache { + case BlockCache: + return db.blockCache.MaxCost(), nil + case IndexCache: + return db.indexCache.MaxCost(), nil + default: + return 0, errors.Errorf("invalid cache type") + } + } + + switch cache { + case BlockCache: + db.blockCache.UpdateMaxCost(maxCost) + return maxCost, nil + case IndexCache: + db.indexCache.UpdateMaxCost(maxCost) + return maxCost, nil + default: + return 0, errors.Errorf("invalid cache type") + } +} + +func (db *DB) LevelsToString() string { + levels := db.Levels() + h := func(sz int64) string { + return humanize.IBytes(uint64(sz)) + } + base := func(b bool) string { + if b { + return "B" + } + return " " + } + + var b strings.Builder + b.WriteRune('\n') + for _, li := range levels { + b.WriteString(fmt.Sprintf( + "Level %d [%s]: NumTables: %02d. Size: %s of %s. Score: %.2f->%.2f"+ + " StaleData: %s Target FileSize: %s\n", + li.Level, base(li.IsBaseLevel), li.NumTables, + h(li.Size), h(li.TargetSize), li.Score, li.Adjusted, h(li.StaleDatSize), + h(li.TargetFileSize))) + } + b.WriteString("Level Done\n") + return b.String() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/dir_plan9.go b/vendor/github.com/dgraph-io/badger/v4/dir_plan9.go new file mode 100644 index 00000000000..7f7f194bcd2 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/dir_plan9.go @@ -0,0 +1,150 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "os" + "path/filepath" + "strings" + + "github.com/dgraph-io/badger/v4/y" +) + +// directoryLockGuard holds a lock on a directory and a pid file inside. The pid file isn't part +// of the locking mechanism, it's just advisory. +type directoryLockGuard struct { + // File handle on the directory, which we've locked. + f *os.File + // The absolute path to our pid file. + path string +} + +// acquireDirectoryLock gets a lock on the directory. +// It will also write our pid to dirPath/pidFileName for convenience. +// readOnly is not supported on Plan 9. +func acquireDirectoryLock(dirPath string, pidFileName string, readOnly bool) ( + *directoryLockGuard, error) { + if readOnly { + return nil, ErrPlan9NotSupported + } + + // Convert to absolute path so that Release still works even if we do an unbalanced + // chdir in the meantime. + absPidFilePath, err := filepath.Abs(filepath.Join(dirPath, pidFileName)) + if err != nil { + return nil, y.Wrap(err, "cannot get absolute path for pid lock file") + } + + // If the file was unpacked or created by some other program, it might not + // have the ModeExclusive bit set. Set it before we call OpenFile, so that we + // can be confident that a successful OpenFile implies exclusive use. + // + // OpenFile fails if the file ModeExclusive bit set *and* the file is already open. + // So, if the file is closed when the DB crashed, we're fine. When the process + // that was managing the DB crashes, the OS will close the file for us. + // + // This bit of code is copied from Go's lockedfile internal package: + // https://github.com/golang/go/blob/go1.15rc1/src/cmd/go/internal/lockedfile/lockedfile_plan9.go#L58 + if fi, err := os.Stat(absPidFilePath); err == nil { + if fi.Mode()&os.ModeExclusive == 0 { + if err := os.Chmod(absPidFilePath, fi.Mode()|os.ModeExclusive); err != nil { + return nil, y.Wrapf(err, "could not set exclusive mode bit") + } + } + } else if !os.IsNotExist(err) { + return nil, err + } + f, err := os.OpenFile(absPidFilePath, os.O_WRONLY|os.O_TRUNC|os.O_CREATE, 0666|os.ModeExclusive) + if err != nil { + if isLocked(err) { + return nil, y.Wrapf(err, + "Cannot open pid lock file %q. Another process is using this Badger database", + absPidFilePath) + } + return nil, y.Wrapf(err, "Cannot open pid lock file %q", absPidFilePath) + } + + if _, err = fmt.Fprintf(f, "%d\n", os.Getpid()); err != nil { + f.Close() + return nil, y.Wrapf(err, "could not write pid") + } + return &directoryLockGuard{f, absPidFilePath}, nil +} + +// Release deletes the pid file and releases our lock on the directory. +func (guard *directoryLockGuard) release() error { + // It's important that we remove the pid file first. + err := os.Remove(guard.path) + + if closeErr := guard.f.Close(); err == nil { + err = closeErr + } + guard.path = "" + guard.f = nil + + return err +} + +// openDir opens a directory for syncing. +func openDir(path string) (*os.File, error) { return os.Open(path) } + +// When you create or delete a file, you have to ensure the directory entry for the file is synced +// in order to guarantee the file is visible (if the system crashes). (See the man page for fsync, +// or see https://github.com/coreos/etcd/issues/6368 for an example.) +func syncDir(dir string) error { + f, err := openDir(dir) + if err != nil { + return y.Wrapf(err, "While opening directory: %s.", dir) + } + + err = f.Sync() + closeErr := f.Close() + if err != nil { + return y.Wrapf(err, "While syncing directory: %s.", dir) + } + return y.Wrapf(closeErr, "While closing directory: %s.", dir) +} + +// Opening an exclusive-use file returns an error. +// The expected error strings are: +// +// - "open/create -- file is locked" (cwfs, kfs) +// - "exclusive lock" (fossil) +// - "exclusive use file already open" (ramfs) +// +// See https://github.com/golang/go/blob/go1.15rc1/src/cmd/go/internal/lockedfile/lockedfile_plan9.go#L16 +var lockedErrStrings = [...]string{ + "file is locked", + "exclusive lock", + "exclusive use file already open", +} + +// Even though plan9 doesn't support the Lock/RLock/Unlock functions to +// manipulate already-open files, IsLocked is still meaningful: os.OpenFile +// itself may return errors that indicate that a file with the ModeExclusive bit +// set is already open. +func isLocked(err error) bool { + s := err.Error() + + for _, frag := range lockedErrStrings { + if strings.Contains(s, frag) { + return true + } + } + return false +} diff --git a/vendor/github.com/dgraph-io/badger/v4/dir_unix.go b/vendor/github.com/dgraph-io/badger/v4/dir_unix.go new file mode 100644 index 00000000000..ecaa5ced78b --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/dir_unix.go @@ -0,0 +1,119 @@ +//go:build !windows && !plan9 +// +build !windows,!plan9 + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "os" + "path/filepath" + + "golang.org/x/sys/unix" + + "github.com/dgraph-io/badger/v4/y" +) + +// directoryLockGuard holds a lock on a directory and a pid file inside. The pid file isn't part +// of the locking mechanism, it's just advisory. +type directoryLockGuard struct { + // File handle on the directory, which we've flocked. + f *os.File + // The absolute path to our pid file. + path string + // Was this a shared lock for a read-only database? + readOnly bool +} + +// acquireDirectoryLock gets a lock on the directory (using flock). If +// this is not read-only, it will also write our pid to +// dirPath/pidFileName for convenience. +func acquireDirectoryLock(dirPath string, pidFileName string, readOnly bool) ( + *directoryLockGuard, error) { + // Convert to absolute path so that Release still works even if we do an unbalanced + // chdir in the meantime. + absPidFilePath, err := filepath.Abs(filepath.Join(dirPath, pidFileName)) + if err != nil { + return nil, y.Wrapf(err, "cannot get absolute path for pid lock file") + } + f, err := os.Open(dirPath) + if err != nil { + return nil, y.Wrapf(err, "cannot open directory %q", dirPath) + } + opts := unix.LOCK_EX | unix.LOCK_NB + if readOnly { + opts = unix.LOCK_SH | unix.LOCK_NB + } + + err = unix.Flock(int(f.Fd()), opts) + if err != nil { + f.Close() + return nil, y.Wrapf(err, + "Cannot acquire directory lock on %q. Another process is using this Badger database.", + dirPath) + } + + if !readOnly { + // Yes, we happily overwrite a pre-existing pid file. We're the + // only read-write badger process using this directory. + err = os.WriteFile(absPidFilePath, []byte(fmt.Sprintf("%d\n", os.Getpid())), 0666) + if err != nil { + f.Close() + return nil, y.Wrapf(err, + "Cannot write pid file %q", absPidFilePath) + } + } + return &directoryLockGuard{f, absPidFilePath, readOnly}, nil +} + +// Release deletes the pid file and releases our lock on the directory. +func (guard *directoryLockGuard) release() error { + var err error + if !guard.readOnly { + // It's important that we remove the pid file first. + err = os.Remove(guard.path) + } + + if closeErr := guard.f.Close(); err == nil { + err = closeErr + } + guard.path = "" + guard.f = nil + + return err +} + +// openDir opens a directory for syncing. +func openDir(path string) (*os.File, error) { return os.Open(path) } + +// When you create or delete a file, you have to ensure the directory entry for the file is synced +// in order to guarantee the file is visible (if the system crashes). (See the man page for fsync, +// or see https://github.com/coreos/etcd/issues/6368 for an example.) +func syncDir(dir string) error { + f, err := openDir(dir) + if err != nil { + return y.Wrapf(err, "While opening directory: %s.", dir) + } + + err = f.Sync() + closeErr := f.Close() + if err != nil { + return y.Wrapf(err, "While syncing directory: %s.", dir) + } + return y.Wrapf(closeErr, "While closing directory: %s.", dir) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/dir_windows.go b/vendor/github.com/dgraph-io/badger/v4/dir_windows.go new file mode 100644 index 00000000000..bd8d2f9d2f8 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/dir_windows.go @@ -0,0 +1,111 @@ +//go:build windows +// +build windows + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +// OpenDir opens a directory in windows with write access for syncing. +import ( + "os" + "path/filepath" + "syscall" + + "github.com/dgraph-io/badger/v4/y" +) + +// FILE_ATTRIBUTE_TEMPORARY - A file that is being used for temporary storage. +// FILE_FLAG_DELETE_ON_CLOSE - The file is to be deleted immediately after all of its handles are +// closed, which includes the specified handle and any other open or duplicated handles. +// See: https://docs.microsoft.com/en-us/windows/desktop/FileIO/file-attribute-constants +// NOTE: Added here to avoid importing golang.org/x/sys/windows +const ( + FILE_ATTRIBUTE_TEMPORARY = 0x00000100 + FILE_FLAG_DELETE_ON_CLOSE = 0x04000000 +) + +func openDir(path string) (*os.File, error) { + fd, err := openDirWin(path) + if err != nil { + return nil, err + } + return os.NewFile(uintptr(fd), path), nil +} + +func openDirWin(path string) (fd syscall.Handle, err error) { + if len(path) == 0 { + return syscall.InvalidHandle, syscall.ERROR_FILE_NOT_FOUND + } + pathp, err := syscall.UTF16PtrFromString(path) + if err != nil { + return syscall.InvalidHandle, err + } + access := uint32(syscall.GENERIC_READ | syscall.GENERIC_WRITE) + sharemode := uint32(syscall.FILE_SHARE_READ | syscall.FILE_SHARE_WRITE) + createmode := uint32(syscall.OPEN_EXISTING) + fl := uint32(syscall.FILE_FLAG_BACKUP_SEMANTICS) + return syscall.CreateFile(pathp, access, sharemode, nil, createmode, fl, 0) +} + +// DirectoryLockGuard holds a lock on the directory. +type directoryLockGuard struct { + h syscall.Handle + path string +} + +// AcquireDirectoryLock acquires exclusive access to a directory. +func acquireDirectoryLock(dirPath string, pidFileName string, readOnly bool) (*directoryLockGuard, error) { + if readOnly { + return nil, ErrWindowsNotSupported + } + + // Convert to absolute path so that Release still works even if we do an unbalanced + // chdir in the meantime. + absLockFilePath, err := filepath.Abs(filepath.Join(dirPath, pidFileName)) + if err != nil { + return nil, y.Wrap(err, "Cannot get absolute path for pid lock file") + } + + // This call creates a file handler in memory that only one process can use at a time. When + // that process ends, the file is deleted by the system. + // FILE_ATTRIBUTE_TEMPORARY is used to tell Windows to try to create the handle in memory. + // FILE_FLAG_DELETE_ON_CLOSE is not specified in syscall_windows.go but tells Windows to delete + // the file when all processes holding the handler are closed. + // XXX: this works but it's a bit klunky. i'd prefer to use LockFileEx but it needs unsafe pkg. + h, err := syscall.CreateFile( + syscall.StringToUTF16Ptr(absLockFilePath), 0, 0, nil, + syscall.OPEN_ALWAYS, + uint32(FILE_ATTRIBUTE_TEMPORARY|FILE_FLAG_DELETE_ON_CLOSE), + 0) + if err != nil { + return nil, y.Wrapf(err, + "Cannot create lock file %q. Another process is using this Badger database", + absLockFilePath) + } + + return &directoryLockGuard{h: h, path: absLockFilePath}, nil +} + +// Release removes the directory lock. +func (g *directoryLockGuard) release() error { + g.path = "" + return syscall.CloseHandle(g.h) +} + +// Windows doesn't support syncing directories to the file system. See +// https://github.com/dgraph-io/badger/issues/699#issuecomment-504133587 for more details. +func syncDir(dir string) error { return nil } diff --git a/vendor/github.com/dgraph-io/badger/v4/discard.go b/vendor/github.com/dgraph-io/badger/v4/discard.go new file mode 100644 index 00000000000..439897cbb0e --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/discard.go @@ -0,0 +1,168 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "encoding/binary" + "os" + "path/filepath" + "sort" + "sync" + + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// discardStats keeps track of the amount of data that could be discarded for +// a given logfile. +type discardStats struct { + sync.Mutex + + *z.MmapFile + opt Options + nextEmptySlot int +} + +const discardFname string = "DISCARD" + +func InitDiscardStats(opt Options) (*discardStats, error) { + fname := filepath.Join(opt.ValueDir, discardFname) + + // 1MB file can store 65.536 discard entries. Each entry is 16 bytes. + mf, err := z.OpenMmapFile(fname, os.O_CREATE|os.O_RDWR, 1<<20) + lf := &discardStats{ + MmapFile: mf, + opt: opt, + } + if err == z.NewFile { + // We don't need to zero out the entire 1MB. + lf.zeroOut() + + } else if err != nil { + return nil, y.Wrapf(err, "while opening file: %s\n", discardFname) + } + + for slot := 0; slot < lf.maxSlot(); slot++ { + if lf.get(16*slot) == 0 { + lf.nextEmptySlot = slot + break + } + } + sort.Sort(lf) + opt.Infof("Discard stats nextEmptySlot: %d\n", lf.nextEmptySlot) + return lf, nil +} + +func (lf *discardStats) Len() int { + return lf.nextEmptySlot +} +func (lf *discardStats) Less(i, j int) bool { + return lf.get(16*i) < lf.get(16*j) +} +func (lf *discardStats) Swap(i, j int) { + left := lf.Data[16*i : 16*i+16] + right := lf.Data[16*j : 16*j+16] + var tmp [16]byte + copy(tmp[:], left) + copy(left, right) + copy(right, tmp[:]) +} + +// offset is not slot. +func (lf *discardStats) get(offset int) uint64 { + return binary.BigEndian.Uint64(lf.Data[offset : offset+8]) +} +func (lf *discardStats) set(offset int, val uint64) { + binary.BigEndian.PutUint64(lf.Data[offset:offset+8], val) +} + +// zeroOut would zero out the next slot. +func (lf *discardStats) zeroOut() { + lf.set(lf.nextEmptySlot*16, 0) + lf.set(lf.nextEmptySlot*16+8, 0) +} + +func (lf *discardStats) maxSlot() int { + return len(lf.Data) / 16 +} + +// Update would update the discard stats for the given file id. If discard is +// 0, it would return the current value of discard for the file. If discard is +// < 0, it would set the current value of discard to zero for the file. +func (lf *discardStats) Update(fidu uint32, discard int64) int64 { + fid := uint64(fidu) + lf.Lock() + defer lf.Unlock() + + idx := sort.Search(lf.nextEmptySlot, func(slot int) bool { + return lf.get(slot*16) >= fid + }) + if idx < lf.nextEmptySlot && lf.get(idx*16) == fid { + off := idx*16 + 8 + curDisc := lf.get(off) + if discard == 0 { + return int64(curDisc) + } + if discard < 0 { + lf.set(off, 0) + return 0 + } + lf.set(off, curDisc+uint64(discard)) + return int64(curDisc + uint64(discard)) + } + if discard <= 0 { + // No need to add a new entry. + return 0 + } + + // Could not find the fid. Add the entry. + idx = lf.nextEmptySlot + lf.set(idx*16, fid) + lf.set(idx*16+8, uint64(discard)) + + // Move to next slot. + lf.nextEmptySlot++ + for lf.nextEmptySlot >= lf.maxSlot() { + y.Check(lf.Truncate(2 * int64(len(lf.Data)))) + } + lf.zeroOut() + + sort.Sort(lf) + return discard +} + +func (lf *discardStats) Iterate(f func(fid, stats uint64)) { + for slot := 0; slot < lf.nextEmptySlot; slot++ { + idx := 16 * slot + f(lf.get(idx), lf.get(idx+8)) + } +} + +// MaxDiscard returns the file id with maximum discard bytes. +func (lf *discardStats) MaxDiscard() (uint32, int64) { + lf.Lock() + defer lf.Unlock() + + var maxFid, maxVal uint64 + lf.Iterate(func(fid, val uint64) { + if maxVal < val { + maxVal = val + maxFid = fid + } + }) + return uint32(maxFid), int64(maxVal) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/doc.go b/vendor/github.com/dgraph-io/badger/v4/doc.go new file mode 100644 index 00000000000..128f2fccb98 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/doc.go @@ -0,0 +1,27 @@ +/* +Package badger implements an embeddable, simple and fast key-value database, +written in pure Go. It is designed to be highly performant for both reads and +writes simultaneously. Badger uses Multi-Version Concurrency Control (MVCC), and +supports transactions. It runs transactions concurrently, with serializable +snapshot isolation guarantees. + +Badger uses an LSM tree along with a value log to separate keys from values, +hence reducing both write amplification and the size of the LSM tree. This +allows LSM tree to be served entirely from RAM, while the values are served +from SSD. + +# Usage + +Badger has the following main types: DB, Txn, Item and Iterator. DB contains +keys that are associated with values. It must be opened with the appropriate +options before it can be accessed. + +All operations happen inside a Txn. Txn represents a transaction, which can +be read-only or read-write. Read-only transactions can read values for a +given key (which are returned inside an Item), or iterate over a set of +key-value pairs using an Iterator (which are returned as Item type values as +well). Read-write transactions can also update and delete keys from the DB. + +See the examples for more usage details. +*/ +package badger diff --git a/vendor/github.com/dgraph-io/badger/v4/errors.go b/vendor/github.com/dgraph-io/badger/v4/errors.go new file mode 100644 index 00000000000..f5df6d5118c --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/errors.go @@ -0,0 +1,126 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "math" + + "github.com/pkg/errors" +) + +const ( + // ValueThresholdLimit is the maximum permissible value of opt.ValueThreshold. + ValueThresholdLimit = math.MaxUint16 - 16 + 1 +) + +var ( + // ErrValueLogSize is returned when opt.ValueLogFileSize option is not within the valid + // range. + ErrValueLogSize = errors.New("Invalid ValueLogFileSize, must be in range [1MB, 2GB)") + + // ErrKeyNotFound is returned when key isn't found on a txn.Get. + ErrKeyNotFound = errors.New("Key not found") + + // ErrTxnTooBig is returned if too many writes are fit into a single transaction. + ErrTxnTooBig = errors.New("Txn is too big to fit into one request") + + // ErrConflict is returned when a transaction conflicts with another transaction. This can + // happen if the read rows had been updated concurrently by another transaction. + ErrConflict = errors.New("Transaction Conflict. Please retry") + + // ErrReadOnlyTxn is returned if an update function is called on a read-only transaction. + ErrReadOnlyTxn = errors.New("No sets or deletes are allowed in a read-only transaction") + + // ErrDiscardedTxn is returned if a previously discarded transaction is re-used. + ErrDiscardedTxn = errors.New("This transaction has been discarded. Create a new one") + + // ErrEmptyKey is returned if an empty key is passed on an update function. + ErrEmptyKey = errors.New("Key cannot be empty") + + // ErrInvalidKey is returned if the key has a special !badger! prefix, + // reserved for internal usage. + ErrInvalidKey = errors.New("Key is using a reserved !badger! prefix") + + // ErrBannedKey is returned if the read/write key belongs to any banned namespace. + ErrBannedKey = errors.New("Key is using the banned prefix") + + // ErrThresholdZero is returned if threshold is set to zero, and value log GC is called. + // In such a case, GC can't be run. + ErrThresholdZero = errors.New( + "Value log GC can't run because threshold is set to zero") + + // ErrNoRewrite is returned if a call for value log GC doesn't result in a log file rewrite. + ErrNoRewrite = errors.New( + "Value log GC attempt didn't result in any cleanup") + + // ErrRejected is returned if a value log GC is called either while another GC is running, or + // after DB::Close has been called. + ErrRejected = errors.New("Value log GC request rejected") + + // ErrInvalidRequest is returned if the user request is invalid. + ErrInvalidRequest = errors.New("Invalid request") + + // ErrManagedTxn is returned if the user tries to use an API which isn't + // allowed due to external management of transactions, when using ManagedDB. + ErrManagedTxn = errors.New( + "Invalid API request. Not allowed to perform this action using ManagedDB") + + // ErrNamespaceMode is returned if the user tries to use an API which is allowed only when + // NamespaceOffset is non-negative. + ErrNamespaceMode = errors.New( + "Invalid API request. Not allowed to perform this action when NamespaceMode is not set.") + + // ErrInvalidDump if a data dump made previously cannot be loaded into the database. + ErrInvalidDump = errors.New("Data dump cannot be read") + + // ErrZeroBandwidth is returned if the user passes in zero bandwidth for sequence. + ErrZeroBandwidth = errors.New("Bandwidth must be greater than zero") + + // ErrWindowsNotSupported is returned when opt.ReadOnly is used on Windows + ErrWindowsNotSupported = errors.New("Read-only mode is not supported on Windows") + + // ErrPlan9NotSupported is returned when opt.ReadOnly is used on Plan 9 + ErrPlan9NotSupported = errors.New("Read-only mode is not supported on Plan 9") + + // ErrTruncateNeeded is returned when the value log gets corrupt, and requires truncation of + // corrupt data to allow Badger to run properly. + ErrTruncateNeeded = errors.New( + "Log truncate required to run DB. This might result in data loss") + + // ErrBlockedWrites is returned if the user called DropAll. During the process of dropping all + // data from Badger, we stop accepting new writes, by returning this error. + ErrBlockedWrites = errors.New("Writes are blocked, possibly due to DropAll or Close") + + // ErrNilCallback is returned when subscriber's callback is nil. + ErrNilCallback = errors.New("Callback cannot be nil") + + // ErrEncryptionKeyMismatch is returned when the storage key is not + // matched with the key previously given. + ErrEncryptionKeyMismatch = errors.New("Encryption key mismatch") + + // ErrInvalidDataKeyID is returned if the datakey id is invalid. + ErrInvalidDataKeyID = errors.New("Invalid datakey id") + + // ErrInvalidEncryptionKey is returned if length of encryption keys is invalid. + ErrInvalidEncryptionKey = errors.New("Encryption key's length should be" + + "either 16, 24, or 32 bytes") + // ErrGCInMemoryMode is returned when db.RunValueLogGC is called in in-memory mode. + ErrGCInMemoryMode = errors.New("Cannot run value log GC when DB is opened in InMemory mode") + + // ErrDBClosed is returned when a get operation is performed after closing the DB. + ErrDBClosed = errors.New("DB Closed") +) diff --git a/vendor/github.com/dgraph-io/badger/v4/fb/BlockOffset.go b/vendor/github.com/dgraph-io/badger/v4/fb/BlockOffset.go new file mode 100644 index 00000000000..6ef437e2505 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/fb/BlockOffset.go @@ -0,0 +1,104 @@ +// Code generated by the FlatBuffers compiler. DO NOT EDIT. + +package fb + +import ( + flatbuffers "github.com/google/flatbuffers/go" +) + +type BlockOffset struct { + _tab flatbuffers.Table +} + +func GetRootAsBlockOffset(buf []byte, offset flatbuffers.UOffsetT) *BlockOffset { + n := flatbuffers.GetUOffsetT(buf[offset:]) + x := &BlockOffset{} + x.Init(buf, n+offset) + return x +} + +func (rcv *BlockOffset) Init(buf []byte, i flatbuffers.UOffsetT) { + rcv._tab.Bytes = buf + rcv._tab.Pos = i +} + +func (rcv *BlockOffset) Table() flatbuffers.Table { + return rcv._tab +} + +func (rcv *BlockOffset) Key(j int) byte { + o := flatbuffers.UOffsetT(rcv._tab.Offset(4)) + if o != 0 { + a := rcv._tab.Vector(o) + return rcv._tab.GetByte(a + flatbuffers.UOffsetT(j*1)) + } + return 0 +} + +func (rcv *BlockOffset) KeyLength() int { + o := flatbuffers.UOffsetT(rcv._tab.Offset(4)) + if o != 0 { + return rcv._tab.VectorLen(o) + } + return 0 +} + +func (rcv *BlockOffset) KeyBytes() []byte { + o := flatbuffers.UOffsetT(rcv._tab.Offset(4)) + if o != 0 { + return rcv._tab.ByteVector(o + rcv._tab.Pos) + } + return nil +} + +func (rcv *BlockOffset) MutateKey(j int, n byte) bool { + o := flatbuffers.UOffsetT(rcv._tab.Offset(4)) + if o != 0 { + a := rcv._tab.Vector(o) + return rcv._tab.MutateByte(a+flatbuffers.UOffsetT(j*1), n) + } + return false +} + +func (rcv *BlockOffset) Offset() uint32 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(6)) + if o != 0 { + return rcv._tab.GetUint32(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *BlockOffset) MutateOffset(n uint32) bool { + return rcv._tab.MutateUint32Slot(6, n) +} + +func (rcv *BlockOffset) Len() uint32 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(8)) + if o != 0 { + return rcv._tab.GetUint32(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *BlockOffset) MutateLen(n uint32) bool { + return rcv._tab.MutateUint32Slot(8, n) +} + +func BlockOffsetStart(builder *flatbuffers.Builder) { + builder.StartObject(3) +} +func BlockOffsetAddKey(builder *flatbuffers.Builder, key flatbuffers.UOffsetT) { + builder.PrependUOffsetTSlot(0, flatbuffers.UOffsetT(key), 0) +} +func BlockOffsetStartKeyVector(builder *flatbuffers.Builder, numElems int) flatbuffers.UOffsetT { + return builder.StartVector(1, numElems, 1) +} +func BlockOffsetAddOffset(builder *flatbuffers.Builder, offset uint32) { + builder.PrependUint32Slot(1, offset, 0) +} +func BlockOffsetAddLen(builder *flatbuffers.Builder, len uint32) { + builder.PrependUint32Slot(2, len, 0) +} +func BlockOffsetEnd(builder *flatbuffers.Builder) flatbuffers.UOffsetT { + return builder.EndObject() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/fb/TableIndex.go b/vendor/github.com/dgraph-io/badger/v4/fb/TableIndex.go new file mode 100644 index 00000000000..7b4074b57a3 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/fb/TableIndex.go @@ -0,0 +1,175 @@ +// Code generated by the FlatBuffers compiler. DO NOT EDIT. + +package fb + +import ( + flatbuffers "github.com/google/flatbuffers/go" +) + +type TableIndex struct { + _tab flatbuffers.Table +} + +func GetRootAsTableIndex(buf []byte, offset flatbuffers.UOffsetT) *TableIndex { + n := flatbuffers.GetUOffsetT(buf[offset:]) + x := &TableIndex{} + x.Init(buf, n+offset) + return x +} + +func (rcv *TableIndex) Init(buf []byte, i flatbuffers.UOffsetT) { + rcv._tab.Bytes = buf + rcv._tab.Pos = i +} + +func (rcv *TableIndex) Table() flatbuffers.Table { + return rcv._tab +} + +func (rcv *TableIndex) Offsets(obj *BlockOffset, j int) bool { + o := flatbuffers.UOffsetT(rcv._tab.Offset(4)) + if o != 0 { + x := rcv._tab.Vector(o) + x += flatbuffers.UOffsetT(j) * 4 + x = rcv._tab.Indirect(x) + obj.Init(rcv._tab.Bytes, x) + return true + } + return false +} + +func (rcv *TableIndex) OffsetsLength() int { + o := flatbuffers.UOffsetT(rcv._tab.Offset(4)) + if o != 0 { + return rcv._tab.VectorLen(o) + } + return 0 +} + +func (rcv *TableIndex) BloomFilter(j int) byte { + o := flatbuffers.UOffsetT(rcv._tab.Offset(6)) + if o != 0 { + a := rcv._tab.Vector(o) + return rcv._tab.GetByte(a + flatbuffers.UOffsetT(j*1)) + } + return 0 +} + +func (rcv *TableIndex) BloomFilterLength() int { + o := flatbuffers.UOffsetT(rcv._tab.Offset(6)) + if o != 0 { + return rcv._tab.VectorLen(o) + } + return 0 +} + +func (rcv *TableIndex) BloomFilterBytes() []byte { + o := flatbuffers.UOffsetT(rcv._tab.Offset(6)) + if o != 0 { + return rcv._tab.ByteVector(o + rcv._tab.Pos) + } + return nil +} + +func (rcv *TableIndex) MutateBloomFilter(j int, n byte) bool { + o := flatbuffers.UOffsetT(rcv._tab.Offset(6)) + if o != 0 { + a := rcv._tab.Vector(o) + return rcv._tab.MutateByte(a+flatbuffers.UOffsetT(j*1), n) + } + return false +} + +func (rcv *TableIndex) MaxVersion() uint64 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(8)) + if o != 0 { + return rcv._tab.GetUint64(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *TableIndex) MutateMaxVersion(n uint64) bool { + return rcv._tab.MutateUint64Slot(8, n) +} + +func (rcv *TableIndex) KeyCount() uint32 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(10)) + if o != 0 { + return rcv._tab.GetUint32(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *TableIndex) MutateKeyCount(n uint32) bool { + return rcv._tab.MutateUint32Slot(10, n) +} + +func (rcv *TableIndex) UncompressedSize() uint32 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(12)) + if o != 0 { + return rcv._tab.GetUint32(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *TableIndex) MutateUncompressedSize(n uint32) bool { + return rcv._tab.MutateUint32Slot(12, n) +} + +func (rcv *TableIndex) OnDiskSize() uint32 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(14)) + if o != 0 { + return rcv._tab.GetUint32(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *TableIndex) MutateOnDiskSize(n uint32) bool { + return rcv._tab.MutateUint32Slot(14, n) +} + +func (rcv *TableIndex) StaleDataSize() uint32 { + o := flatbuffers.UOffsetT(rcv._tab.Offset(16)) + if o != 0 { + return rcv._tab.GetUint32(o + rcv._tab.Pos) + } + return 0 +} + +func (rcv *TableIndex) MutateStaleDataSize(n uint32) bool { + return rcv._tab.MutateUint32Slot(16, n) +} + +func TableIndexStart(builder *flatbuffers.Builder) { + builder.StartObject(7) +} +func TableIndexAddOffsets(builder *flatbuffers.Builder, offsets flatbuffers.UOffsetT) { + builder.PrependUOffsetTSlot(0, flatbuffers.UOffsetT(offsets), 0) +} +func TableIndexStartOffsetsVector(builder *flatbuffers.Builder, numElems int) flatbuffers.UOffsetT { + return builder.StartVector(4, numElems, 4) +} +func TableIndexAddBloomFilter(builder *flatbuffers.Builder, bloomFilter flatbuffers.UOffsetT) { + builder.PrependUOffsetTSlot(1, flatbuffers.UOffsetT(bloomFilter), 0) +} +func TableIndexStartBloomFilterVector(builder *flatbuffers.Builder, numElems int) flatbuffers.UOffsetT { + return builder.StartVector(1, numElems, 1) +} +func TableIndexAddMaxVersion(builder *flatbuffers.Builder, maxVersion uint64) { + builder.PrependUint64Slot(2, maxVersion, 0) +} +func TableIndexAddKeyCount(builder *flatbuffers.Builder, keyCount uint32) { + builder.PrependUint32Slot(3, keyCount, 0) +} +func TableIndexAddUncompressedSize(builder *flatbuffers.Builder, uncompressedSize uint32) { + builder.PrependUint32Slot(4, uncompressedSize, 0) +} +func TableIndexAddOnDiskSize(builder *flatbuffers.Builder, onDiskSize uint32) { + builder.PrependUint32Slot(5, onDiskSize, 0) +} +func TableIndexAddStaleDataSize(builder *flatbuffers.Builder, staleDataSize uint32) { + builder.PrependUint32Slot(6, staleDataSize, 0) +} +func TableIndexEnd(builder *flatbuffers.Builder) flatbuffers.UOffsetT { + return builder.EndObject() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/fb/flatbuffer.fbs b/vendor/github.com/dgraph-io/badger/v4/fb/flatbuffer.fbs new file mode 100644 index 00000000000..af559dd4a49 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/fb/flatbuffer.fbs @@ -0,0 +1,36 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +namespace fb; + +table TableIndex { + offsets:[BlockOffset]; + bloom_filter:[ubyte]; + max_version:uint64; + key_count:uint32; + uncompressed_size:uint32; + on_disk_size:uint32; + stale_data_size:uint32; +} + +table BlockOffset { + key:[ubyte]; + offset:uint; + len:uint; +} + +root_type TableIndex; +root_type BlockOffset; diff --git a/vendor/github.com/dgraph-io/badger/v4/fb/gen.sh b/vendor/github.com/dgraph-io/badger/v4/fb/gen.sh new file mode 100644 index 00000000000..90d88b4a3ae --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/fb/gen.sh @@ -0,0 +1,12 @@ +#!/usr/bin/env bash + +set -e + +## Install flatc if not present +## ref. https://google.github.io/flatbuffers/flatbuffers_guide_building.html +command -v flatc > /dev/null || { ./install_flatbuffers.sh ; } + +flatc --go flatbuffer.fbs +# Move files to the correct directory. +mv fb/* ./ +rmdir fb diff --git a/vendor/github.com/dgraph-io/badger/v4/fb/install_flatbuffers.sh b/vendor/github.com/dgraph-io/badger/v4/fb/install_flatbuffers.sh new file mode 100644 index 00000000000..ae07da7e245 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/fb/install_flatbuffers.sh @@ -0,0 +1,47 @@ +#!/usr/bin/env bash + +set -e + +install_mac() { + command -v brew > /dev/null || \ + { echo "[ERROR]: 'brew' command not not found. Exiting" 1>&2; exit 1; } + brew install flatbuffers +} + +install_linux() { + for CMD in curl cmake g++ make; do + command -v $CMD > /dev/null || \ + { echo "[ERROR]: '$CMD' command not not found. Exiting" 1>&2; exit 1; } + done + + ## Create Temp Build Directory + BUILD_DIR=$(mktemp -d) + pushd $BUILD_DIR + + ## Fetch Latest Tarball + LATEST_VERSION=$(curl -s https://api.github.com/repos/google/flatbuffers/releases/latest | grep -oP '(?<=tag_name": ")[^"]+') + curl -sLO https://github.com/google/flatbuffers/archive/$LATEST_VERSION.tar.gz + tar xf $LATEST_VERSION.tar.gz + + ## Build Binaries + cd flatbuffers-${LATEST_VERSION#v} + cmake -G "Unix Makefiles" -DCMAKE_BUILD_TYPE=Release + make + ./flattests + cp flatc /usr/local/bin/flatc + + ## Cleanup Temp Build Directory + popd + rm -rf $BUILD_DIR +} + +SYSTEM=$(uname -s) + +case ${SYSTEM,,} in + linux) + sudo bash -c "$(declare -f install_linux); install_linux" + ;; + darwin) + install_mac + ;; +esac diff --git a/vendor/github.com/dgraph-io/badger/v4/histogram.go b/vendor/github.com/dgraph-io/badger/v4/histogram.go new file mode 100644 index 00000000000..ad6002123de --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/histogram.go @@ -0,0 +1,169 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "math" +) + +// PrintHistogram builds and displays the key-value size histogram. +// When keyPrefix is set, only the keys that have prefix "keyPrefix" are +// considered for creating the histogram +func (db *DB) PrintHistogram(keyPrefix []byte) { + if db == nil { + fmt.Println("\nCannot build histogram: DB is nil.") + return + } + histogram := db.buildHistogram(keyPrefix) + fmt.Printf("Histogram of key sizes (in bytes)\n") + histogram.keySizeHistogram.printHistogram() + fmt.Printf("Histogram of value sizes (in bytes)\n") + histogram.valueSizeHistogram.printHistogram() +} + +// histogramData stores information about a histogram +type histogramData struct { + bins []int64 + countPerBin []int64 + totalCount int64 + min int64 + max int64 + sum int64 +} + +// sizeHistogram contains keySize histogram and valueSize histogram +type sizeHistogram struct { + keySizeHistogram, valueSizeHistogram histogramData +} + +// newSizeHistogram returns a new instance of keyValueSizeHistogram with +// properly initialized fields. +func newSizeHistogram() *sizeHistogram { + // TODO(ibrahim): find appropriate bin size. + keyBins := createHistogramBins(1, 16) + valueBins := createHistogramBins(1, 30) + return &sizeHistogram{ + keySizeHistogram: histogramData{ + bins: keyBins, + countPerBin: make([]int64, len(keyBins)+1), + max: math.MinInt64, + min: math.MaxInt64, + sum: 0, + }, + valueSizeHistogram: histogramData{ + bins: valueBins, + countPerBin: make([]int64, len(valueBins)+1), + max: math.MinInt64, + min: math.MaxInt64, + sum: 0, + }, + } +} + +// createHistogramBins creates bins for an histogram. The bin sizes are powers +// of two of the form [2^min_exponent, ..., 2^max_exponent]. +func createHistogramBins(minExponent, maxExponent uint32) []int64 { + var bins []int64 + for i := minExponent; i <= maxExponent; i++ { + bins = append(bins, int64(1)< histogram.max { + histogram.max = value + } + if value < histogram.min { + histogram.min = value + } + + histogram.sum += value + histogram.totalCount++ + + for index := 0; index <= len(histogram.bins); index++ { + // Allocate value in the last buckets if we reached the end of the Bounds array. + if index == len(histogram.bins) { + histogram.countPerBin[index]++ + break + } + + // Check if the value should be added to the "index" bin + if value < histogram.bins[index] { + histogram.countPerBin[index]++ + break + } + } +} + +// buildHistogram builds the key-value size histogram. +// When keyPrefix is set, only the keys that have prefix "keyPrefix" are +// considered for creating the histogram +func (db *DB) buildHistogram(keyPrefix []byte) *sizeHistogram { + txn := db.NewTransaction(false) + defer txn.Discard() + + itr := txn.NewIterator(DefaultIteratorOptions) + defer itr.Close() + + badgerHistogram := newSizeHistogram() + + // Collect key and value sizes. + for itr.Seek(keyPrefix); itr.ValidForPrefix(keyPrefix); itr.Next() { + item := itr.Item() + badgerHistogram.keySizeHistogram.Update(item.KeySize()) + badgerHistogram.valueSizeHistogram.Update(item.ValueSize()) + } + return badgerHistogram +} + +// printHistogram prints the histogram data in a human-readable format. +func (histogram histogramData) printHistogram() { + fmt.Printf("Total count: %d\n", histogram.totalCount) + fmt.Printf("Min value: %d\n", histogram.min) + fmt.Printf("Max value: %d\n", histogram.max) + fmt.Printf("Mean: %.2f\n", float64(histogram.sum)/float64(histogram.totalCount)) + fmt.Printf("%24s %9s\n", "Range", "Count") + + numBins := len(histogram.bins) + for index, count := range histogram.countPerBin { + if count == 0 { + continue + } + + // The last bin represents the bin that contains the range from + // the last bin up to infinity so it's processed differently than the + // other bins. + if index == len(histogram.countPerBin)-1 { + lowerBound := int(histogram.bins[numBins-1]) + fmt.Printf("[%10d, %10s) %9d\n", lowerBound, "infinity", count) + continue + } + + upperBound := int(histogram.bins[index]) + lowerBound := 0 + if index > 0 { + lowerBound = int(histogram.bins[index-1]) + } + + fmt.Printf("[%10d, %10d) %9d\n", lowerBound, upperBound, count) + } + fmt.Println() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/iterator.go b/vendor/github.com/dgraph-io/badger/v4/iterator.go new file mode 100644 index 00000000000..2f69db6b24f --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/iterator.go @@ -0,0 +1,787 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "fmt" + "hash/crc32" + "math" + "sort" + "sync" + "time" + + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +type prefetchStatus uint8 + +const ( + prefetched prefetchStatus = iota + 1 +) + +// Item is returned during iteration. Both the Key() and Value() output is only valid until +// iterator.Next() is called. +type Item struct { + key []byte + vptr []byte + val []byte + version uint64 + expiresAt uint64 + + slice *y.Slice // Used only during prefetching. + next *Item + txn *Txn + + err error + wg sync.WaitGroup + status prefetchStatus + meta byte // We need to store meta to know about bitValuePointer. + userMeta byte +} + +// String returns a string representation of Item +func (item *Item) String() string { + return fmt.Sprintf("key=%q, version=%d, meta=%x", item.Key(), item.Version(), item.meta) +} + +// Key returns the key. +// +// Key is only valid as long as item is valid, or transaction is valid. If you need to use it +// outside its validity, please use KeyCopy. +func (item *Item) Key() []byte { + return item.key +} + +// KeyCopy returns a copy of the key of the item, writing it to dst slice. +// If nil is passed, or capacity of dst isn't sufficient, a new slice would be allocated and +// returned. +func (item *Item) KeyCopy(dst []byte) []byte { + return y.SafeCopy(dst, item.key) +} + +// Version returns the commit timestamp of the item. +func (item *Item) Version() uint64 { + return item.version +} + +// Value retrieves the value of the item from the value log. +// +// This method must be called within a transaction. Calling it outside a +// transaction is considered undefined behavior. If an iterator is being used, +// then Item.Value() is defined in the current iteration only, because items are +// reused. +// +// If you need to use a value outside a transaction, please use Item.ValueCopy +// instead, or copy it yourself. Value might change once discard or commit is called. +// Use ValueCopy if you want to do a Set after Get. +func (item *Item) Value(fn func(val []byte) error) error { + item.wg.Wait() + if item.status == prefetched { + if item.err == nil && fn != nil { + if err := fn(item.val); err != nil { + return err + } + } + return item.err + } + buf, cb, err := item.yieldItemValue() + defer runCallback(cb) + if err != nil { + return err + } + if fn != nil { + return fn(buf) + } + return nil +} + +// ValueCopy returns a copy of the value of the item from the value log, writing it to dst slice. +// If nil is passed, or capacity of dst isn't sufficient, a new slice would be allocated and +// returned. Tip: It might make sense to reuse the returned slice as dst argument for the next call. +// +// This function is useful in long running iterate/update transactions to avoid a write deadlock. +// See Github issue: https://github.com/dgraph-io/badger/issues/315 +func (item *Item) ValueCopy(dst []byte) ([]byte, error) { + item.wg.Wait() + if item.status == prefetched { + return y.SafeCopy(dst, item.val), item.err + } + buf, cb, err := item.yieldItemValue() + defer runCallback(cb) + return y.SafeCopy(dst, buf), err +} + +func (item *Item) hasValue() bool { + if item.meta == 0 && item.vptr == nil { + // key not found + return false + } + return true +} + +// IsDeletedOrExpired returns true if item contains deleted or expired value. +func (item *Item) IsDeletedOrExpired() bool { + return isDeletedOrExpired(item.meta, item.expiresAt) +} + +// DiscardEarlierVersions returns whether the item was created with the +// option to discard earlier versions of a key when multiple are available. +func (item *Item) DiscardEarlierVersions() bool { + return item.meta&bitDiscardEarlierVersions > 0 +} + +func (item *Item) yieldItemValue() ([]byte, func(), error) { + key := item.Key() // No need to copy. + if !item.hasValue() { + return nil, nil, nil + } + + if item.slice == nil { + item.slice = new(y.Slice) + } + + if (item.meta & bitValuePointer) == 0 { + val := item.slice.Resize(len(item.vptr)) + copy(val, item.vptr) + return val, nil, nil + } + + var vp valuePointer + vp.Decode(item.vptr) + db := item.txn.db + result, cb, err := db.vlog.Read(vp, item.slice) + if err != nil { + db.opt.Errorf("Unable to read: Key: %v, Version : %v, meta: %v, userMeta: %v"+ + " Error: %v", key, item.version, item.meta, item.userMeta, err) + var txn *Txn + if db.opt.managedTxns { + txn = db.NewTransactionAt(math.MaxUint64, false) + } else { + txn = db.NewTransaction(false) + } + defer txn.Discard() + + iopt := DefaultIteratorOptions + iopt.AllVersions = true + iopt.InternalAccess = true + iopt.PrefetchValues = false + + it := txn.NewKeyIterator(item.Key(), iopt) + defer it.Close() + for it.Rewind(); it.Valid(); it.Next() { + item := it.Item() + var vp valuePointer + if item.meta&bitValuePointer > 0 { + vp.Decode(item.vptr) + } + db.opt.Errorf("Key: %v, Version : %v, meta: %v, userMeta: %v valuePointer: %+v", + item.Key(), item.version, item.meta, item.userMeta, vp) + } + } + // Don't return error if we cannot read the value. Just log the error. + return result, cb, nil +} + +func runCallback(cb func()) { + if cb != nil { + cb() + } +} + +func (item *Item) prefetchValue() { + val, cb, err := item.yieldItemValue() + defer runCallback(cb) + + item.err = err + item.status = prefetched + if val == nil { + return + } + buf := item.slice.Resize(len(val)) + copy(buf, val) + item.val = buf +} + +// EstimatedSize returns the approximate size of the key-value pair. +// +// This can be called while iterating through a store to quickly estimate the +// size of a range of key-value pairs (without fetching the corresponding +// values). +func (item *Item) EstimatedSize() int64 { + if !item.hasValue() { + return 0 + } + if (item.meta & bitValuePointer) == 0 { + return int64(len(item.key) + len(item.vptr)) + } + var vp valuePointer + vp.Decode(item.vptr) + return int64(vp.Len) // includes key length. +} + +// KeySize returns the size of the key. +// Exact size of the key is key + 8 bytes of timestamp +func (item *Item) KeySize() int64 { + return int64(len(item.key)) +} + +// ValueSize returns the approximate size of the value. +// +// This can be called to quickly estimate the size of a value without fetching +// it. +func (item *Item) ValueSize() int64 { + if !item.hasValue() { + return 0 + } + if (item.meta & bitValuePointer) == 0 { + return int64(len(item.vptr)) + } + var vp valuePointer + vp.Decode(item.vptr) + + klen := int64(len(item.key) + 8) // 8 bytes for timestamp. + // 6 bytes are for the approximate length of the header. Since header is encoded in varint, we + // cannot find the exact length of header without fetching it. + return int64(vp.Len) - klen - 6 - crc32.Size +} + +// UserMeta returns the userMeta set by the user. Typically, this byte, optionally set by the user +// is used to interpret the value. +func (item *Item) UserMeta() byte { + return item.userMeta +} + +// ExpiresAt returns a Unix time value indicating when the item will be +// considered expired. 0 indicates that the item will never expire. +func (item *Item) ExpiresAt() uint64 { + return item.expiresAt +} + +// TODO: Switch this to use linked list container in Go. +type list struct { + head *Item + tail *Item +} + +func (l *list) push(i *Item) { + i.next = nil + if l.tail == nil { + l.head = i + l.tail = i + return + } + l.tail.next = i + l.tail = i +} + +func (l *list) pop() *Item { + if l.head == nil { + return nil + } + i := l.head + if l.head == l.tail { + l.tail = nil + l.head = nil + } else { + l.head = i.next + } + i.next = nil + return i +} + +// IteratorOptions is used to set options when iterating over Badger key-value +// stores. +// +// This package provides DefaultIteratorOptions which contains options that +// should work for most applications. Consider using that as a starting point +// before customizing it for your own needs. +type IteratorOptions struct { + // PrefetchSize is the number of KV pairs to prefetch while iterating. + // Valid only if PrefetchValues is true. + PrefetchSize int + // PrefetchValues Indicates whether we should prefetch values during + // iteration and store them. + PrefetchValues bool + Reverse bool // Direction of iteration. False is forward, true is backward. + AllVersions bool // Fetch all valid versions of the same key. + InternalAccess bool // Used to allow internal access to badger keys. + + // The following option is used to narrow down the SSTables that iterator + // picks up. If Prefix is specified, only tables which could have this + // prefix are picked based on their range of keys. + prefixIsKey bool // If set, use the prefix for bloom filter lookup. + Prefix []byte // Only iterate over this given prefix. + SinceTs uint64 // Only read data that has version > SinceTs. +} + +func (opt *IteratorOptions) compareToPrefix(key []byte) int { + // We should compare key without timestamp. For example key - a[TS] might be > "aa" prefix. + key = y.ParseKey(key) + if len(key) > len(opt.Prefix) { + key = key[:len(opt.Prefix)] + } + return bytes.Compare(key, opt.Prefix) +} + +func (opt *IteratorOptions) pickTable(t table.TableInterface) bool { + // Ignore this table if its max version is less than the sinceTs. + if t.MaxVersion() < opt.SinceTs { + return false + } + if len(opt.Prefix) == 0 { + return true + } + if opt.compareToPrefix(t.Smallest()) > 0 { + return false + } + if opt.compareToPrefix(t.Biggest()) < 0 { + return false + } + // Bloom filter lookup would only work if opt.Prefix does NOT have the read + // timestamp as part of the key. + if opt.prefixIsKey && t.DoesNotHave(y.Hash(opt.Prefix)) { + return false + } + return true +} + +// pickTables picks the necessary table for the iterator. This function also assumes +// that the tables are sorted in the right order. +func (opt *IteratorOptions) pickTables(all []*table.Table) []*table.Table { + filterTables := func(tables []*table.Table) []*table.Table { + if opt.SinceTs > 0 { + tmp := tables[:0] + for _, t := range tables { + if t.MaxVersion() < opt.SinceTs { + continue + } + tmp = append(tmp, t) + } + tables = tmp + } + return tables + } + + if len(opt.Prefix) == 0 { + out := make([]*table.Table, len(all)) + copy(out, all) + return filterTables(out) + } + sIdx := sort.Search(len(all), func(i int) bool { + // table.Biggest >= opt.prefix + // if opt.Prefix < table.Biggest, then surely it is not in any of the preceding tables. + return opt.compareToPrefix(all[i].Biggest()) >= 0 + }) + if sIdx == len(all) { + // Not found. + return []*table.Table{} + } + + filtered := all[sIdx:] + if !opt.prefixIsKey { + eIdx := sort.Search(len(filtered), func(i int) bool { + return opt.compareToPrefix(filtered[i].Smallest()) > 0 + }) + out := make([]*table.Table, len(filtered[:eIdx])) + copy(out, filtered[:eIdx]) + return filterTables(out) + } + + // opt.prefixIsKey == true. This code is optimizing for opt.prefixIsKey part. + var out []*table.Table + hash := y.Hash(opt.Prefix) + for _, t := range filtered { + // When we encounter the first table whose smallest key is higher than opt.Prefix, we can + // stop. This is an IMPORTANT optimization, just considering how often we call + // NewKeyIterator. + if opt.compareToPrefix(t.Smallest()) > 0 { + // if table.Smallest > opt.Prefix, then this and all tables after this can be ignored. + break + } + // opt.Prefix is actually the key. So, we can run bloom filter checks + // as well. + if t.DoesNotHave(hash) { + continue + } + out = append(out, t) + } + return filterTables(out) +} + +// DefaultIteratorOptions contains default options when iterating over Badger key-value stores. +var DefaultIteratorOptions = IteratorOptions{ + PrefetchValues: true, + PrefetchSize: 100, + Reverse: false, + AllVersions: false, +} + +// Iterator helps iterating over the KV pairs in a lexicographically sorted order. +type Iterator struct { + iitr y.Iterator + txn *Txn + readTs uint64 + + opt IteratorOptions + item *Item + data list + waste list + + lastKey []byte // Used to skip over multiple versions of the same key. + + closed bool + scanned int // Used to estimate the size of data scanned by iterator. + + // ThreadId is an optional value that can be set to identify which goroutine created + // the iterator. It can be used, for example, to uniquely identify each of the + // iterators created by the stream interface + ThreadId int + + Alloc *z.Allocator +} + +// NewIterator returns a new iterator. Depending upon the options, either only keys, or both +// key-value pairs would be fetched. The keys are returned in lexicographically sorted order. +// Using prefetch is recommended if you're doing a long running iteration, for performance. +// +// Multiple Iterators: +// For a read-only txn, multiple iterators can be running simultaneously. However, for a read-write +// txn, iterators have the nuance of being a snapshot of the writes for the transaction at the time +// iterator was created. If writes are performed after an iterator is created, then that iterator +// will not be able to see those writes. Only writes performed before an iterator was created can be +// viewed. +func (txn *Txn) NewIterator(opt IteratorOptions) *Iterator { + if txn.discarded { + panic(ErrDiscardedTxn) + } + if txn.db.IsClosed() { + panic(ErrDBClosed) + } + + y.NumIteratorsCreatedAdd(txn.db.opt.MetricsEnabled, 1) + + // Keep track of the number of active iterators. + txn.numIterators.Add(1) + + // TODO: If Prefix is set, only pick those memtables which have keys with the prefix. + tables, decr := txn.db.getMemTables() + defer decr() + txn.db.vlog.incrIteratorCount() + var iters []y.Iterator + if itr := txn.newPendingWritesIterator(opt.Reverse); itr != nil { + iters = append(iters, itr) + } + for i := 0; i < len(tables); i++ { + iters = append(iters, tables[i].sl.NewUniIterator(opt.Reverse)) + } + iters = txn.db.lc.appendIterators(iters, &opt) // This will increment references. + res := &Iterator{ + txn: txn, + iitr: table.NewMergeIterator(iters, opt.Reverse), + opt: opt, + readTs: txn.readTs, + } + return res +} + +// NewKeyIterator is just like NewIterator, but allows the user to iterate over all versions of a +// single key. Internally, it sets the Prefix option in provided opt, and uses that prefix to +// additionally run bloom filter lookups before picking tables from the LSM tree. +func (txn *Txn) NewKeyIterator(key []byte, opt IteratorOptions) *Iterator { + if len(opt.Prefix) > 0 { + panic("opt.Prefix should be nil for NewKeyIterator.") + } + opt.Prefix = key // This key must be without the timestamp. + opt.prefixIsKey = true + opt.AllVersions = true + return txn.NewIterator(opt) +} + +func (it *Iterator) newItem() *Item { + item := it.waste.pop() + if item == nil { + item = &Item{slice: new(y.Slice), txn: it.txn} + } + return item +} + +// Item returns pointer to the current key-value pair. +// This item is only valid until it.Next() gets called. +func (it *Iterator) Item() *Item { + tx := it.txn + tx.addReadKey(it.item.Key()) + return it.item +} + +// Valid returns false when iteration is done. +func (it *Iterator) Valid() bool { + if it.item == nil { + return false + } + if it.opt.prefixIsKey { + return bytes.Equal(it.item.key, it.opt.Prefix) + } + return bytes.HasPrefix(it.item.key, it.opt.Prefix) +} + +// ValidForPrefix returns false when iteration is done +// or when the current key is not prefixed by the specified prefix. +func (it *Iterator) ValidForPrefix(prefix []byte) bool { + return it.Valid() && bytes.HasPrefix(it.item.key, prefix) +} + +// Close would close the iterator. It is important to call this when you're done with iteration. +func (it *Iterator) Close() { + if it.closed { + return + } + it.closed = true + if it.iitr == nil { + it.txn.numIterators.Add(-1) + return + } + + it.iitr.Close() + // It is important to wait for the fill goroutines to finish. Otherwise, we might leave zombie + // goroutines behind, which are waiting to acquire file read locks after DB has been closed. + waitFor := func(l list) { + item := l.pop() + for item != nil { + item.wg.Wait() + item = l.pop() + } + } + waitFor(it.waste) + waitFor(it.data) + + // TODO: We could handle this error. + _ = it.txn.db.vlog.decrIteratorCount() + it.txn.numIterators.Add(-1) +} + +// Next would advance the iterator by one. Always check it.Valid() after a Next() +// to ensure you have access to a valid it.Item(). +func (it *Iterator) Next() { + if it.iitr == nil { + return + } + // Reuse current item + it.item.wg.Wait() // Just cleaner to wait before pushing to avoid doing ref counting. + it.scanned += len(it.item.key) + len(it.item.val) + len(it.item.vptr) + 2 + it.waste.push(it.item) + + // Set next item to current + it.item = it.data.pop() + for it.iitr.Valid() { + if it.parseItem() { + // parseItem calls one extra next. + // This is used to deal with the complexity of reverse iteration. + break + } + } +} + +func isDeletedOrExpired(meta byte, expiresAt uint64) bool { + if meta&bitDelete > 0 { + return true + } + if expiresAt == 0 { + return false + } + return expiresAt <= uint64(time.Now().Unix()) +} + +// parseItem is a complex function because it needs to handle both forward and reverse iteration +// implementation. We store keys such that their versions are sorted in descending order. This makes +// forward iteration efficient, but revese iteration complicated. This tradeoff is better because +// forward iteration is more common than reverse. It returns true, if either the iterator is invalid +// or it has pushed an item into it.data list, else it returns false. +// +// This function advances the iterator. +func (it *Iterator) parseItem() bool { + mi := it.iitr + key := mi.Key() + + setItem := func(item *Item) { + if it.item == nil { + it.item = item + } else { + it.data.push(item) + } + } + + isInternalKey := bytes.HasPrefix(key, badgerPrefix) + // Skip badger keys. + if !it.opt.InternalAccess && isInternalKey { + mi.Next() + return false + } + + // Skip any versions which are beyond the readTs. + version := y.ParseTs(key) + // Ignore everything that is above the readTs and below or at the sinceTs. + if version > it.readTs || (it.opt.SinceTs > 0 && version <= it.opt.SinceTs) { + mi.Next() + return false + } + + // Skip banned keys only if it does not have badger internal prefix. + if !isInternalKey && it.txn.db.isBanned(key) != nil { + mi.Next() + return false + } + + if it.opt.AllVersions { + // Return deleted or expired values also, otherwise user can't figure out + // whether the key was deleted. + item := it.newItem() + it.fill(item) + setItem(item) + mi.Next() + return true + } + + // If iterating in forward direction, then just checking the last key against current key would + // be sufficient. + if !it.opt.Reverse { + if y.SameKey(it.lastKey, key) { + mi.Next() + return false + } + // Only track in forward direction. + // We should update lastKey as soon as we find a different key in our snapshot. + // Consider keys: a 5, b 7 (del), b 5. When iterating, lastKey = a. + // Then we see b 7, which is deleted. If we don't store lastKey = b, we'll then return b 5, + // which is wrong. Therefore, update lastKey here. + it.lastKey = y.SafeCopy(it.lastKey, mi.Key()) + } + +FILL: + // If deleted, advance and return. + vs := mi.Value() + if isDeletedOrExpired(vs.Meta, vs.ExpiresAt) { + mi.Next() + return false + } + + item := it.newItem() + it.fill(item) + // fill item based on current cursor position. All Next calls have returned, so reaching here + // means no Next was called. + + mi.Next() // Advance but no fill item yet. + if !it.opt.Reverse || !mi.Valid() { // Forward direction, or invalid. + setItem(item) + return true + } + + // Reverse direction. + nextTs := y.ParseTs(mi.Key()) + mik := y.ParseKey(mi.Key()) + if nextTs <= it.readTs && bytes.Equal(mik, item.key) { + // This is a valid potential candidate. + goto FILL + } + // Ignore the next candidate. Return the current one. + setItem(item) + return true +} + +func (it *Iterator) fill(item *Item) { + vs := it.iitr.Value() + item.meta = vs.Meta + item.userMeta = vs.UserMeta + item.expiresAt = vs.ExpiresAt + + item.version = y.ParseTs(it.iitr.Key()) + item.key = y.SafeCopy(item.key, y.ParseKey(it.iitr.Key())) + + item.vptr = y.SafeCopy(item.vptr, vs.Value) + item.val = nil + if it.opt.PrefetchValues { + item.wg.Add(1) + go func() { + // FIXME we are not handling errors here. + item.prefetchValue() + item.wg.Done() + }() + } +} + +func (it *Iterator) prefetch() { + prefetchSize := 2 + if it.opt.PrefetchValues && it.opt.PrefetchSize > 1 { + prefetchSize = it.opt.PrefetchSize + } + + i := it.iitr + var count int + it.item = nil + for i.Valid() { + if !it.parseItem() { + continue + } + count++ + if count == prefetchSize { + break + } + } +} + +// Seek would seek to the provided key if present. If absent, it would seek to the next +// smallest key greater than the provided key if iterating in the forward direction. +// Behavior would be reversed if iterating backwards. +func (it *Iterator) Seek(key []byte) { + if it.iitr == nil { + return + } + if len(key) > 0 { + it.txn.addReadKey(key) + } + for i := it.data.pop(); i != nil; i = it.data.pop() { + i.wg.Wait() + it.waste.push(i) + } + + it.lastKey = it.lastKey[:0] + if len(key) == 0 { + key = it.opt.Prefix + } + if len(key) == 0 { + it.iitr.Rewind() + it.prefetch() + return + } + + if !it.opt.Reverse { + key = y.KeyWithTs(key, it.txn.readTs) + } else { + key = y.KeyWithTs(key, 0) + } + it.iitr.Seek(key) + it.prefetch() +} + +// Rewind would rewind the iterator cursor all the way to zero-th position, which would be the +// smallest key if iterating forward, and largest if iterating backward. It does not keep track of +// whether the cursor started with a Seek(). +func (it *Iterator) Rewind() { + it.Seek(nil) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/key_registry.go b/vendor/github.com/dgraph-io/badger/v4/key_registry.go new file mode 100644 index 00000000000..a1be0435da7 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/key_registry.go @@ -0,0 +1,424 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "crypto/aes" + "crypto/rand" + "encoding/binary" + "hash/crc32" + "io" + "os" + "path/filepath" + "sync" + "time" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" +) + +const ( + // KeyRegistryFileName is the file name for the key registry file. + KeyRegistryFileName = "KEYREGISTRY" + // KeyRegistryRewriteFileName is the file name for the rewrite key registry file. + KeyRegistryRewriteFileName = "REWRITE-KEYREGISTRY" +) + +// SanityText is used to check whether the given user provided storage key is valid or not +var sanityText = []byte("Hello Badger") + +// KeyRegistry used to maintain all the data keys. +type KeyRegistry struct { + sync.RWMutex + dataKeys map[uint64]*pb.DataKey + lastCreated int64 //lastCreated is the timestamp(seconds) of the last data key generated. + nextKeyID uint64 + fp *os.File + opt KeyRegistryOptions +} + +type KeyRegistryOptions struct { + Dir string + ReadOnly bool + EncryptionKey []byte + EncryptionKeyRotationDuration time.Duration + InMemory bool +} + +// newKeyRegistry returns KeyRegistry. +func newKeyRegistry(opt KeyRegistryOptions) *KeyRegistry { + return &KeyRegistry{ + dataKeys: make(map[uint64]*pb.DataKey), + nextKeyID: 0, + opt: opt, + } +} + +// OpenKeyRegistry opens key registry if it exists, otherwise it'll create key registry +// and returns key registry. +func OpenKeyRegistry(opt KeyRegistryOptions) (*KeyRegistry, error) { + // sanity check the encryption key length. + if len(opt.EncryptionKey) > 0 { + switch len(opt.EncryptionKey) { + default: + return nil, y.Wrapf(ErrInvalidEncryptionKey, "During OpenKeyRegistry") + case 16, 24, 32: + break + } + } + // If db is opened in InMemory mode, we don't need to write key registry to the disk. + if opt.InMemory { + return newKeyRegistry(opt), nil + } + path := filepath.Join(opt.Dir, KeyRegistryFileName) + var flags y.Flags + if opt.ReadOnly { + flags |= y.ReadOnly + } else { + flags |= y.Sync + } + fp, err := y.OpenExistingFile(path, flags) + // OpenExistingFile just open file. + // So checking whether the file exist or not. If not + // We'll create new keyregistry. + if os.IsNotExist(err) { + // Creating new registry file if not exist. + kr := newKeyRegistry(opt) + if opt.ReadOnly { + return kr, nil + } + // Writing the key registry to the file. + if err := WriteKeyRegistry(kr, opt); err != nil { + return nil, y.Wrapf(err, "Error while writing key registry.") + } + fp, err = y.OpenExistingFile(path, flags) + if err != nil { + return nil, y.Wrapf(err, "Error while opening newly created key registry.") + } + } else if err != nil { + return nil, y.Wrapf(err, "Error while opening key registry.") + } + kr, err := readKeyRegistry(fp, opt) + if err != nil { + // This case happens only if the file is opened properly and + // not able to read. + fp.Close() + return nil, err + } + if opt.ReadOnly { + // We'll close the file in readonly mode. + return kr, fp.Close() + } + kr.fp = fp + return kr, nil +} + +// keyRegistryIterator reads all the datakey from the key registry +type keyRegistryIterator struct { + encryptionKey []byte + fp *os.File + // lenCrcBuf contains crc buf and data length to move forward. + lenCrcBuf [8]byte +} + +// newKeyRegistryIterator returns iterator which will allow you to iterate +// over the data key of the key registry. +func newKeyRegistryIterator(fp *os.File, encryptionKey []byte) (*keyRegistryIterator, error) { + return &keyRegistryIterator{ + encryptionKey: encryptionKey, + fp: fp, + lenCrcBuf: [8]byte{}, + }, validRegistry(fp, encryptionKey) +} + +// validRegistry checks that given encryption key is valid or not. +func validRegistry(fp *os.File, encryptionKey []byte) error { + iv := make([]byte, aes.BlockSize) + var err error + if _, err = fp.Read(iv); err != nil { + return y.Wrapf(err, "Error while reading IV for key registry.") + } + eSanityText := make([]byte, len(sanityText)) + if _, err = fp.Read(eSanityText); err != nil { + return y.Wrapf(err, "Error while reading sanity text.") + } + if len(encryptionKey) > 0 { + // Decrypting sanity text. + if eSanityText, err = y.XORBlockAllocate(eSanityText, encryptionKey, iv); err != nil { + return y.Wrapf(err, "During validRegistry") + } + } + // Check the given key is valid or not. + if !bytes.Equal(eSanityText, sanityText) { + return ErrEncryptionKeyMismatch + } + return nil +} + +func (kri *keyRegistryIterator) next() (*pb.DataKey, error) { + var err error + // Read crc buf and data length. + if _, err = kri.fp.Read(kri.lenCrcBuf[:]); err != nil { + // EOF means end of the iteration. + if err != io.EOF { + return nil, y.Wrapf(err, "While reading crc in keyRegistryIterator.next") + } + return nil, err + } + l := int64(binary.BigEndian.Uint32(kri.lenCrcBuf[0:4])) + // Read protobuf data. + data := make([]byte, l) + if _, err = kri.fp.Read(data); err != nil { + // EOF means end of the iteration. + if err != io.EOF { + return nil, y.Wrapf(err, "While reading protobuf in keyRegistryIterator.next") + } + return nil, err + } + // Check checksum. + if crc32.Checksum(data, y.CastagnoliCrcTable) != binary.BigEndian.Uint32(kri.lenCrcBuf[4:]) { + return nil, y.Wrapf(y.ErrChecksumMismatch, "Error while checking checksum for data key.") + } + dataKey := &pb.DataKey{} + if err = dataKey.Unmarshal(data); err != nil { + return nil, y.Wrapf(err, "While unmarshal of datakey in keyRegistryIterator.next") + } + if len(kri.encryptionKey) > 0 { + // Decrypt the key if the storage key exists. + if dataKey.Data, err = y.XORBlockAllocate(dataKey.Data, kri.encryptionKey, dataKey.Iv); err != nil { + return nil, y.Wrapf(err, "While decrypting datakey in keyRegistryIterator.next") + } + } + return dataKey, nil +} + +// readKeyRegistry will read the key registry file and build the key registry struct. +func readKeyRegistry(fp *os.File, opt KeyRegistryOptions) (*KeyRegistry, error) { + itr, err := newKeyRegistryIterator(fp, opt.EncryptionKey) + if err != nil { + return nil, err + } + kr := newKeyRegistry(opt) + var dk *pb.DataKey + dk, err = itr.next() + for err == nil && dk != nil { + if dk.KeyId > kr.nextKeyID { + // Set the maximum key ID for next key ID generation. + kr.nextKeyID = dk.KeyId + } + if dk.CreatedAt > kr.lastCreated { + // Set the last generated key timestamp. + kr.lastCreated = dk.CreatedAt + } + // No need to lock since we are building the initial state. + kr.dataKeys[dk.KeyId] = dk + // Forward the iterator. + dk, err = itr.next() + } + // We read all the key. So, Ignoring this error. + if err == io.EOF { + err = nil + } + return kr, err +} + +/* +Structure of Key Registry. ++-------------------+---------------------+--------------------+--------------+------------------+ +| IV | Sanity Text | DataKey1 | DataKey2 | ... | ++-------------------+---------------------+--------------------+--------------+------------------+ +*/ + +// WriteKeyRegistry will rewrite the existing key registry file with new one. +// It is okay to give closed key registry. Since, it's using only the datakey. +func WriteKeyRegistry(reg *KeyRegistry, opt KeyRegistryOptions) error { + buf := &bytes.Buffer{} + iv, err := y.GenerateIV() + y.Check(err) + // Encrypt sanity text if the encryption key is presents. + eSanity := sanityText + if len(opt.EncryptionKey) > 0 { + var err error + eSanity, err = y.XORBlockAllocate(eSanity, opt.EncryptionKey, iv) + if err != nil { + return y.Wrapf(err, "Error while encrpting sanity text in WriteKeyRegistry") + } + } + y.Check2(buf.Write(iv)) + y.Check2(buf.Write(eSanity)) + // Write all the datakeys to the buf. + for _, k := range reg.dataKeys { + // Writing the datakey to the given buffer. + if err := storeDataKey(buf, opt.EncryptionKey, k); err != nil { + return y.Wrapf(err, "Error while storing datakey in WriteKeyRegistry") + } + } + tmpPath := filepath.Join(opt.Dir, KeyRegistryRewriteFileName) + // Open temporary file to write the data and do atomic rename. + fp, err := y.OpenTruncFile(tmpPath, true) + if err != nil { + return y.Wrapf(err, "Error while opening tmp file in WriteKeyRegistry") + } + // Write buf to the disk. + if _, err = fp.Write(buf.Bytes()); err != nil { + // close the fd before returning error. We're not using defer + // because, for windows we need to close the fd explicitly before + // renaming. + fp.Close() + return y.Wrapf(err, "Error while writing buf in WriteKeyRegistry") + } + // In Windows the files should be closed before doing a Rename. + if err = fp.Close(); err != nil { + return y.Wrapf(err, "Error while closing tmp file in WriteKeyRegistry") + } + // Rename to the original file. + if err = os.Rename(tmpPath, filepath.Join(opt.Dir, KeyRegistryFileName)); err != nil { + return y.Wrapf(err, "Error while renaming file in WriteKeyRegistry") + } + // Sync Dir. + return syncDir(opt.Dir) +} + +// DataKey returns datakey of the given key id. +func (kr *KeyRegistry) DataKey(id uint64) (*pb.DataKey, error) { + kr.RLock() + defer kr.RUnlock() + if id == 0 { + // nil represent plain text. + return nil, nil + } + dk, ok := kr.dataKeys[id] + if !ok { + return nil, y.Wrapf(ErrInvalidDataKeyID, "Error for the KEY ID %d", id) + } + return dk, nil +} + +// LatestDataKey will give you the latest generated datakey based on the rotation +// period. If the last generated datakey lifetime exceeds the rotation period. +// It'll create new datakey. +func (kr *KeyRegistry) LatestDataKey() (*pb.DataKey, error) { + if len(kr.opt.EncryptionKey) == 0 { + // nil is for no encryption. + return nil, nil + } + // validKey return datakey if the last generated key duration less than + // rotation duration. + validKey := func() (*pb.DataKey, bool) { + // Time diffrence from the last generated time. + diff := time.Since(time.Unix(kr.lastCreated, 0)) + if diff < kr.opt.EncryptionKeyRotationDuration { + return kr.dataKeys[kr.nextKeyID], true + } + return nil, false + } + kr.RLock() + key, valid := validKey() + kr.RUnlock() + if valid { + // If less than EncryptionKeyRotationDuration, returns the last generated key. + return key, nil + } + kr.Lock() + defer kr.Unlock() + // Key might have generated by another go routine. So, + // checking once again. + key, valid = validKey() + if valid { + return key, nil + } + k := make([]byte, len(kr.opt.EncryptionKey)) + iv, err := y.GenerateIV() + if err != nil { + return nil, err + } + _, err = rand.Read(k) + if err != nil { + return nil, err + } + // Otherwise Increment the KeyID and generate new datakey. + kr.nextKeyID++ + dk := &pb.DataKey{ + KeyId: kr.nextKeyID, + Data: k, + CreatedAt: time.Now().Unix(), + Iv: iv, + } + // Don't store the datakey on file if badger is running in InMemory mode. + if !kr.opt.InMemory { + // Store the datekey. + buf := &bytes.Buffer{} + if err = storeDataKey(buf, kr.opt.EncryptionKey, dk); err != nil { + return nil, err + } + // Persist the datakey to the disk + if _, err = kr.fp.Write(buf.Bytes()); err != nil { + return nil, err + } + } + // storeDatakey encrypts the datakey So, placing un-encrypted key in the memory. + dk.Data = k + kr.lastCreated = dk.CreatedAt + kr.dataKeys[kr.nextKeyID] = dk + return dk, nil +} + +// Close closes the key registry. +func (kr *KeyRegistry) Close() error { + if !(kr.opt.ReadOnly || kr.opt.InMemory) { + return kr.fp.Close() + } + return nil +} + +// storeDataKey stores datakey in an encrypted format in the given buffer. If storage key preset. +func storeDataKey(buf *bytes.Buffer, storageKey []byte, k *pb.DataKey) error { + // xor will encrypt the IV and xor with the given data. + // It'll used for both encryption and decryption. + xor := func() error { + if len(storageKey) == 0 { + return nil + } + var err error + k.Data, err = y.XORBlockAllocate(k.Data, storageKey, k.Iv) + return err + } + // In memory datakey will be plain text so encrypting before storing to the disk. + var err error + if err = xor(); err != nil { + return y.Wrapf(err, "Error while encrypting datakey in storeDataKey") + } + var data []byte + if data, err = k.Marshal(); err != nil { + err = y.Wrapf(err, "Error while marshaling datakey in storeDataKey") + var err2 error + // decrypting the datakey back. + if err2 = xor(); err2 != nil { + return y.Wrapf(err, + y.Wrapf(err2, "Error while decrypting datakey in storeDataKey").Error()) + } + return err + } + var lenCrcBuf [8]byte + binary.BigEndian.PutUint32(lenCrcBuf[0:4], uint32(len(data))) + binary.BigEndian.PutUint32(lenCrcBuf[4:8], crc32.Checksum(data, y.CastagnoliCrcTable)) + y.Check2(buf.Write(lenCrcBuf[:])) + y.Check2(buf.Write(data)) + // Decrypting the datakey back since we're using the pointer. + return xor() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/level_handler.go b/vendor/github.com/dgraph-io/badger/v4/level_handler.go new file mode 100644 index 00000000000..31673f15b2a --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/level_handler.go @@ -0,0 +1,353 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "sort" + "sync" + + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" +) + +type levelHandler struct { + // Guards tables, totalSize. + sync.RWMutex + + // For level >= 1, tables are sorted by key ranges, which do not overlap. + // For level 0, tables are sorted by time. + // For level 0, newest table are at the back. Compact the oldest one first, which is at the front. + tables []*table.Table + totalSize int64 + totalStaleSize int64 + + // The following are initialized once and const. + level int + strLevel string + db *DB +} + +func (s *levelHandler) isLastLevel() bool { + return s.level == s.db.opt.MaxLevels-1 +} + +func (s *levelHandler) getTotalStaleSize() int64 { + s.RLock() + defer s.RUnlock() + return s.totalStaleSize +} + +func (s *levelHandler) getTotalSize() int64 { + s.RLock() + defer s.RUnlock() + return s.totalSize +} + +// initTables replaces s.tables with given tables. This is done during loading. +func (s *levelHandler) initTables(tables []*table.Table) { + s.Lock() + defer s.Unlock() + + s.tables = tables + s.totalSize = 0 + s.totalStaleSize = 0 + for _, t := range tables { + s.addSize(t) + } + + if s.level == 0 { + // Key range will overlap. Just sort by fileID in ascending order + // because newer tables are at the end of level 0. + sort.Slice(s.tables, func(i, j int) bool { + return s.tables[i].ID() < s.tables[j].ID() + }) + } else { + // Sort tables by keys. + sort.Slice(s.tables, func(i, j int) bool { + return y.CompareKeys(s.tables[i].Smallest(), s.tables[j].Smallest()) < 0 + }) + } +} + +// deleteTables remove tables idx0, ..., idx1-1. +func (s *levelHandler) deleteTables(toDel []*table.Table) error { + s.Lock() // s.Unlock() below + + toDelMap := make(map[uint64]struct{}) + for _, t := range toDel { + toDelMap[t.ID()] = struct{}{} + } + + // Make a copy as iterators might be keeping a slice of tables. + var newTables []*table.Table + for _, t := range s.tables { + _, found := toDelMap[t.ID()] + if !found { + newTables = append(newTables, t) + continue + } + s.subtractSize(t) + } + s.tables = newTables + + s.Unlock() // Unlock s _before_ we DecrRef our tables, which can be slow. + + return decrRefs(toDel) +} + +// replaceTables will replace tables[left:right] with newTables. Note this EXCLUDES tables[right]. +// You must call decr() to delete the old tables _after_ writing the update to the manifest. +func (s *levelHandler) replaceTables(toDel, toAdd []*table.Table) error { + // Need to re-search the range of tables in this level to be replaced as other goroutines might + // be changing it as well. (They can't touch our tables, but if they add/remove other tables, + // the indices get shifted around.) + s.Lock() // We s.Unlock() below. + + toDelMap := make(map[uint64]struct{}) + for _, t := range toDel { + toDelMap[t.ID()] = struct{}{} + } + var newTables []*table.Table + for _, t := range s.tables { + _, found := toDelMap[t.ID()] + if !found { + newTables = append(newTables, t) + continue + } + s.subtractSize(t) + } + + // Increase totalSize first. + for _, t := range toAdd { + s.addSize(t) + t.IncrRef() + newTables = append(newTables, t) + } + + // Assign tables. + s.tables = newTables + sort.Slice(s.tables, func(i, j int) bool { + return y.CompareKeys(s.tables[i].Smallest(), s.tables[j].Smallest()) < 0 + }) + s.Unlock() // s.Unlock before we DecrRef tables -- that can be slow. + return decrRefs(toDel) +} + +// addTable adds toAdd table to levelHandler. Normally when we add tables to levelHandler, we sort +// tables based on table.Smallest. This is required for correctness of the system. But in case of +// stream writer this can be avoided. We can just add tables to levelHandler's table list +// and after all addTable calls, we can sort table list(check sortTable method). +// NOTE: levelHandler.sortTables() should be called after call addTable calls are done. +func (s *levelHandler) addTable(t *table.Table) { + s.Lock() + defer s.Unlock() + + s.addSize(t) // Increase totalSize first. + t.IncrRef() + s.tables = append(s.tables, t) +} + +// sortTables sorts tables of levelHandler based on table.Smallest. +// Normally it should be called after all addTable calls. +func (s *levelHandler) sortTables() { + s.RLock() + defer s.RUnlock() + + sort.Slice(s.tables, func(i, j int) bool { + return y.CompareKeys(s.tables[i].Smallest(), s.tables[j].Smallest()) < 0 + }) +} + +func decrRefs(tables []*table.Table) error { + for _, table := range tables { + if err := table.DecrRef(); err != nil { + return err + } + } + return nil +} + +func newLevelHandler(db *DB, level int) *levelHandler { + return &levelHandler{ + level: level, + strLevel: fmt.Sprintf("l%d", level), + db: db, + } +} + +// tryAddLevel0Table returns true if ok and no stalling. +func (s *levelHandler) tryAddLevel0Table(t *table.Table) bool { + y.AssertTrue(s.level == 0) + // Need lock as we may be deleting the first table during a level 0 compaction. + s.Lock() + defer s.Unlock() + // Stall (by returning false) if we are above the specified stall setting for L0. + if len(s.tables) >= s.db.opt.NumLevelZeroTablesStall { + return false + } + + s.tables = append(s.tables, t) + t.IncrRef() + s.addSize(t) + + return true +} + +// This should be called while holding the lock on the level. +func (s *levelHandler) addSize(t *table.Table) { + s.totalSize += t.Size() + s.totalStaleSize += int64(t.StaleDataSize()) +} + +// This should be called while holding the lock on the level. +func (s *levelHandler) subtractSize(t *table.Table) { + s.totalSize -= t.Size() + s.totalStaleSize -= int64(t.StaleDataSize()) +} +func (s *levelHandler) numTables() int { + s.RLock() + defer s.RUnlock() + return len(s.tables) +} + +func (s *levelHandler) close() error { + s.RLock() + defer s.RUnlock() + var err error + for _, t := range s.tables { + if closeErr := t.Close(-1); closeErr != nil && err == nil { + err = closeErr + } + } + return y.Wrap(err, "levelHandler.close") +} + +// getTableForKey acquires a read-lock to access s.tables. It returns a list of tableHandlers. +func (s *levelHandler) getTableForKey(key []byte) ([]*table.Table, func() error) { + s.RLock() + defer s.RUnlock() + + if s.level == 0 { + // For level 0, we need to check every table. Remember to make a copy as s.tables may change + // once we exit this function, and we don't want to lock s.tables while seeking in tables. + // CAUTION: Reverse the tables. + out := make([]*table.Table, 0, len(s.tables)) + for i := len(s.tables) - 1; i >= 0; i-- { + out = append(out, s.tables[i]) + s.tables[i].IncrRef() + } + return out, func() error { + for _, t := range out { + if err := t.DecrRef(); err != nil { + return err + } + } + return nil + } + } + // For level >= 1, we can do a binary search as key range does not overlap. + idx := sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(s.tables[i].Biggest(), key) >= 0 + }) + if idx >= len(s.tables) { + // Given key is strictly > than every element we have. + return nil, func() error { return nil } + } + tbl := s.tables[idx] + tbl.IncrRef() + return []*table.Table{tbl}, tbl.DecrRef +} + +// get returns value for a given key or the key after that. If not found, return nil. +func (s *levelHandler) get(key []byte) (y.ValueStruct, error) { + tables, decr := s.getTableForKey(key) + keyNoTs := y.ParseKey(key) + + hash := y.Hash(keyNoTs) + var maxVs y.ValueStruct + for _, th := range tables { + if th.DoesNotHave(hash) { + y.NumLSMBloomHitsAdd(s.db.opt.MetricsEnabled, s.strLevel, 1) + continue + } + + it := th.NewIterator(0) + defer it.Close() + + y.NumLSMGetsAdd(s.db.opt.MetricsEnabled, s.strLevel, 1) + it.Seek(key) + if !it.Valid() { + continue + } + if y.SameKey(key, it.Key()) { + if version := y.ParseTs(it.Key()); maxVs.Version < version { + maxVs = it.ValueCopy() + maxVs.Version = version + } + } + } + return maxVs, decr() +} + +// appendIterators appends iterators to an array of iterators, for merging. +// Note: This obtains references for the table handlers. Remember to close these iterators. +func (s *levelHandler) appendIterators(iters []y.Iterator, opt *IteratorOptions) []y.Iterator { + s.RLock() + defer s.RUnlock() + + var topt int + if opt.Reverse { + topt = table.REVERSED + } + if s.level == 0 { + // Remember to add in reverse order! + // The newer table at the end of s.tables should be added first as it takes precedence. + // Level 0 tables are not in key sorted order, so we need to consider them one by one. + var out []*table.Table + for _, t := range s.tables { + if opt.pickTable(t) { + out = append(out, t) + } + } + return appendIteratorsReversed(iters, out, topt) + } + + tables := opt.pickTables(s.tables) + if len(tables) == 0 { + return iters + } + return append(iters, table.NewConcatIterator(tables, topt)) +} + +type levelHandlerRLocked struct{} + +// overlappingTables returns the tables that intersect with key range. Returns a half-interval. +// This function should already have acquired a read lock, and this is so important the caller must +// pass an empty parameter declaring such. +func (s *levelHandler) overlappingTables(_ levelHandlerRLocked, kr keyRange) (int, int) { + if len(kr.left) == 0 || len(kr.right) == 0 { + return 0, 0 + } + left := sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(kr.left, s.tables[i].Biggest()) <= 0 + }) + right := sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(kr.right, s.tables[i].Smallest()) < 0 + }) + return left, right +} diff --git a/vendor/github.com/dgraph-io/badger/v4/levels.go b/vendor/github.com/dgraph-io/badger/v4/levels.go new file mode 100644 index 00000000000..3e397e704ed --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/levels.go @@ -0,0 +1,1774 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "context" + "encoding/hex" + "fmt" + "math" + "math/rand" + "os" + "sort" + "strings" + "sync" + "sync/atomic" + "time" + + "github.com/pkg/errors" + otrace "go.opencensus.io/trace" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +type levelsController struct { + nextFileID atomic.Uint64 + l0stallsMs atomic.Int64 + + // The following are initialized once and const. + levels []*levelHandler + kv *DB + + cstatus compactStatus +} + +// revertToManifest checks that all necessary table files exist and removes all table files not +// referenced by the manifest. idMap is a set of table file id's that were read from the directory +// listing. +func revertToManifest(kv *DB, mf *Manifest, idMap map[uint64]struct{}) error { + // 1. Check all files in manifest exist. + for id := range mf.Tables { + if _, ok := idMap[id]; !ok { + return fmt.Errorf("file does not exist for table %d", id) + } + } + + // 2. Delete files that shouldn't exist. + for id := range idMap { + if _, ok := mf.Tables[id]; !ok { + kv.opt.Debugf("Table file %d not referenced in MANIFEST\n", id) + filename := table.NewFilename(id, kv.opt.Dir) + if err := os.Remove(filename); err != nil { + return y.Wrapf(err, "While removing table %d", id) + } + } + } + + return nil +} + +func newLevelsController(db *DB, mf *Manifest) (*levelsController, error) { + y.AssertTrue(db.opt.NumLevelZeroTablesStall > db.opt.NumLevelZeroTables) + s := &levelsController{ + kv: db, + levels: make([]*levelHandler, db.opt.MaxLevels), + } + s.cstatus.tables = make(map[uint64]struct{}) + s.cstatus.levels = make([]*levelCompactStatus, db.opt.MaxLevels) + + for i := 0; i < db.opt.MaxLevels; i++ { + s.levels[i] = newLevelHandler(db, i) + s.cstatus.levels[i] = new(levelCompactStatus) + } + + if db.opt.InMemory { + return s, nil + } + // Compare manifest against directory, check for existent/non-existent files, and remove. + if err := revertToManifest(db, mf, getIDMap(db.opt.Dir)); err != nil { + return nil, err + } + + var mu sync.Mutex + tables := make([][]*table.Table, db.opt.MaxLevels) + var maxFileID uint64 + + // We found that using 3 goroutines allows disk throughput to be utilized to its max. + // Disk utilization is the main thing we should focus on, while trying to read the data. That's + // the one factor that remains constant between HDD and SSD. + throttle := y.NewThrottle(3) + + start := time.Now() + var numOpened atomic.Int32 + tick := time.NewTicker(3 * time.Second) + defer tick.Stop() + + for fileID, tf := range mf.Tables { + fname := table.NewFilename(fileID, db.opt.Dir) + select { + case <-tick.C: + db.opt.Infof("%d tables out of %d opened in %s\n", numOpened.Load(), + len(mf.Tables), time.Since(start).Round(time.Millisecond)) + default: + } + if err := throttle.Do(); err != nil { + closeAllTables(tables) + return nil, err + } + if fileID > maxFileID { + maxFileID = fileID + } + go func(fname string, tf TableManifest) { + var rerr error + defer func() { + throttle.Done(rerr) + numOpened.Add(1) + }() + dk, err := db.registry.DataKey(tf.KeyID) + if err != nil { + rerr = y.Wrapf(err, "Error while reading datakey") + return + } + topt := buildTableOptions(db) + // Explicitly set Compression and DataKey based on how the table was generated. + topt.Compression = tf.Compression + topt.DataKey = dk + + mf, err := z.OpenMmapFile(fname, db.opt.getFileFlags(), 0) + if err != nil { + rerr = y.Wrapf(err, "Opening file: %q", fname) + return + } + t, err := table.OpenTable(mf, topt) + if err != nil { + if strings.HasPrefix(err.Error(), "CHECKSUM_MISMATCH:") { + db.opt.Errorf(err.Error()) + db.opt.Errorf("Ignoring table %s", mf.Fd.Name()) + // Do not set rerr. We will continue without this table. + } else { + rerr = y.Wrapf(err, "Opening table: %q", fname) + } + return + } + + mu.Lock() + tables[tf.Level] = append(tables[tf.Level], t) + mu.Unlock() + }(fname, tf) + } + if err := throttle.Finish(); err != nil { + closeAllTables(tables) + return nil, err + } + db.opt.Infof("All %d tables opened in %s\n", numOpened.Load(), + time.Since(start).Round(time.Millisecond)) + s.nextFileID.Store(maxFileID + 1) + for i, tbls := range tables { + s.levels[i].initTables(tbls) + } + + // Make sure key ranges do not overlap etc. + if err := s.validate(); err != nil { + _ = s.cleanupLevels() + return nil, y.Wrap(err, "Level validation") + } + + // Sync directory (because we have at least removed some files, or previously created the + // manifest file). + if err := syncDir(db.opt.Dir); err != nil { + _ = s.close() + return nil, err + } + + return s, nil +} + +// Closes the tables, for cleanup in newLevelsController. (We Close() instead of using DecrRef() +// because that would delete the underlying files.) We ignore errors, which is OK because tables +// are read-only. +func closeAllTables(tables [][]*table.Table) { + for _, tableSlice := range tables { + for _, table := range tableSlice { + _ = table.Close(-1) + } + } +} + +func (s *levelsController) cleanupLevels() error { + var firstErr error + for _, l := range s.levels { + if err := l.close(); err != nil && firstErr == nil { + firstErr = err + } + } + return firstErr +} + +// dropTree picks all tables from all levels, creates a manifest changeset, +// applies it, and then decrements the refs of these tables, which would result +// in their deletion. +func (s *levelsController) dropTree() (int, error) { + // First pick all tables, so we can create a manifest changelog. + var all []*table.Table + for _, l := range s.levels { + l.RLock() + all = append(all, l.tables...) + l.RUnlock() + } + if len(all) == 0 { + return 0, nil + } + + // Generate the manifest changes. + changes := []*pb.ManifestChange{} + for _, table := range all { + // Add a delete change only if the table is not in memory. + if !table.IsInmemory { + changes = append(changes, newDeleteChange(table.ID())) + } + } + changeSet := pb.ManifestChangeSet{Changes: changes} + if err := s.kv.manifest.addChanges(changeSet.Changes); err != nil { + return 0, err + } + + // Now that manifest has been successfully written, we can delete the tables. + for _, l := range s.levels { + l.Lock() + l.totalSize = 0 + l.tables = l.tables[:0] + l.Unlock() + } + for _, table := range all { + if err := table.DecrRef(); err != nil { + return 0, err + } + } + return len(all), nil +} + +// dropPrefix runs a L0->L1 compaction, and then runs same level compaction on the rest of the +// levels. For L0->L1 compaction, it runs compactions normally, but skips over +// all the keys with the provided prefix. +// For Li->Li compactions, it picks up the tables which would have the prefix. The +// tables who only have keys with this prefix are quickly dropped. The ones which have other keys +// are run through MergeIterator and compacted to create new tables. All the mechanisms of +// compactions apply, i.e. level sizes and MANIFEST are updated as in the normal flow. +func (s *levelsController) dropPrefixes(prefixes [][]byte) error { + opt := s.kv.opt + // Iterate levels in the reverse order because if we were to iterate from + // lower level (say level 0) to a higher level (say level 3) we could have + // a state in which level 0 is compacted and an older version of a key exists in lower level. + // At this point, if someone creates an iterator, they would see an old + // value for a key from lower levels. Iterating in reverse order ensures we + // drop the oldest data first so that lookups never return stale data. + for i := len(s.levels) - 1; i >= 0; i-- { + l := s.levels[i] + + l.RLock() + if l.level == 0 { + size := len(l.tables) + l.RUnlock() + + if size > 0 { + cp := compactionPriority{ + level: 0, + score: 1.74, + // A unique number greater than 1.0 does two things. Helps identify this + // function in logs, and forces a compaction. + dropPrefixes: prefixes, + } + if err := s.doCompact(174, cp); err != nil { + opt.Warningf("While compacting level 0: %v", err) + return nil + } + } + continue + } + + // Build a list of compaction tableGroups affecting all the prefixes we + // need to drop. We need to build tableGroups that satisfy the invariant that + // bottom tables are consecutive. + // tableGroup contains groups of consecutive tables. + var tableGroups [][]*table.Table + var tableGroup []*table.Table + + finishGroup := func() { + if len(tableGroup) > 0 { + tableGroups = append(tableGroups, tableGroup) + tableGroup = nil + } + } + + for _, table := range l.tables { + if containsAnyPrefixes(table, prefixes) { + tableGroup = append(tableGroup, table) + } else { + finishGroup() + } + } + finishGroup() + + l.RUnlock() + + if len(tableGroups) == 0 { + continue + } + _, span := otrace.StartSpan(context.Background(), "Badger.Compaction") + span.Annotatef(nil, "Compaction level: %v", l.level) + span.Annotatef(nil, "Drop Prefixes: %v", prefixes) + defer span.End() + opt.Infof("Dropping prefix at level %d (%d tableGroups)", l.level, len(tableGroups)) + for _, operation := range tableGroups { + cd := compactDef{ + span: span, + thisLevel: l, + nextLevel: l, + top: nil, + bot: operation, + dropPrefixes: prefixes, + t: s.levelTargets(), + } + cd.t.baseLevel = l.level + if err := s.runCompactDef(-1, l.level, cd); err != nil { + opt.Warningf("While running compact def: %+v. Error: %v", cd, err) + return err + } + } + } + return nil +} + +func (s *levelsController) startCompact(lc *z.Closer) { + n := s.kv.opt.NumCompactors + lc.AddRunning(n - 1) + for i := 0; i < n; i++ { + go s.runCompactor(i, lc) + } +} + +type targets struct { + baseLevel int + targetSz []int64 + fileSz []int64 +} + +// levelTargets calculates the targets for levels in the LSM tree. The idea comes from Dynamic Level +// Sizes ( https://rocksdb.org/blog/2015/07/23/dynamic-level.html ) in RocksDB. The sizes of levels +// are calculated based on the size of the lowest level, typically L6. So, if L6 size is 1GB, then +// L5 target size is 100MB, L4 target size is 10MB and so on. +// +// L0 files don't automatically go to L1. Instead, they get compacted to Lbase, where Lbase is +// chosen based on the first level which is non-empty from top (check L1 through L6). For an empty +// DB, that would be L6. So, L0 compactions go to L6, then L5, L4 and so on. +// +// Lbase is advanced to the upper levels when its target size exceeds BaseLevelSize. For +// example, when L6 reaches 1.1GB, then L4 target sizes becomes 11MB, thus exceeding the +// BaseLevelSize of 10MB. L3 would then become the new Lbase, with a target size of 1MB < +// BaseLevelSize. +func (s *levelsController) levelTargets() targets { + adjust := func(sz int64) int64 { + if sz < s.kv.opt.BaseLevelSize { + return s.kv.opt.BaseLevelSize + } + return sz + } + + t := targets{ + targetSz: make([]int64, len(s.levels)), + fileSz: make([]int64, len(s.levels)), + } + // DB size is the size of the last level. + dbSize := s.lastLevel().getTotalSize() + for i := len(s.levels) - 1; i > 0; i-- { + ltarget := adjust(dbSize) + t.targetSz[i] = ltarget + if t.baseLevel == 0 && ltarget <= s.kv.opt.BaseLevelSize { + t.baseLevel = i + } + dbSize /= int64(s.kv.opt.LevelSizeMultiplier) + } + + tsz := s.kv.opt.BaseTableSize + for i := 0; i < len(s.levels); i++ { + if i == 0 { + // Use MemTableSize for Level 0. Because at Level 0, we stop compactions based on the + // number of tables, not the size of the level. So, having a 1:1 size ratio between + // memtable size and the size of L0 files is better than churning out 32 files per + // memtable (assuming 64MB MemTableSize and 2MB BaseTableSize). + t.fileSz[i] = s.kv.opt.MemTableSize + } else if i <= t.baseLevel { + t.fileSz[i] = tsz + } else { + tsz *= int64(s.kv.opt.TableSizeMultiplier) + t.fileSz[i] = tsz + } + } + + // Bring the base level down to the last empty level. + for i := t.baseLevel + 1; i < len(s.levels)-1; i++ { + if s.levels[i].getTotalSize() > 0 { + break + } + t.baseLevel = i + } + + // If the base level is empty and the next level size is less than the + // target size, pick the next level as the base level. + b := t.baseLevel + lvl := s.levels + if b < len(lvl)-1 && lvl[b].getTotalSize() == 0 && lvl[b+1].getTotalSize() < t.targetSz[b+1] { + t.baseLevel++ + } + return t +} + +func (s *levelsController) runCompactor(id int, lc *z.Closer) { + defer lc.Done() + + randomDelay := time.NewTimer(time.Duration(rand.Int31n(1000)) * time.Millisecond) + select { + case <-randomDelay.C: + case <-lc.HasBeenClosed(): + randomDelay.Stop() + return + } + + moveL0toFront := func(prios []compactionPriority) []compactionPriority { + idx := -1 + for i, p := range prios { + if p.level == 0 { + idx = i + break + } + } + // If idx == -1, we didn't find L0. + // If idx == 0, then we don't need to do anything. L0 is already at the front. + if idx > 0 { + out := append([]compactionPriority{}, prios[idx]) + out = append(out, prios[:idx]...) + out = append(out, prios[idx+1:]...) + return out + } + return prios + } + + run := func(p compactionPriority) bool { + err := s.doCompact(id, p) + switch err { + case nil: + return true + case errFillTables: + // pass + default: + s.kv.opt.Warningf("While running doCompact: %v\n", err) + } + return false + } + runOnce := func() bool { + prios := s.pickCompactLevels() + if id == 0 { + // Worker ID zero prefers to compact L0 always. + prios = moveL0toFront(prios) + } + for _, p := range prios { + if id == 0 && p.level == 0 { + // Allow worker zero to run level 0, irrespective of its adjusted score. + } else if p.adjusted < 1.0 { + break + } + if run(p) { + return true + } + } + + return false + } + + tryLmaxToLmaxCompaction := func() { + p := compactionPriority{ + level: s.lastLevel().level, + t: s.levelTargets(), + } + run(p) + + } + count := 0 + ticker := time.NewTicker(50 * time.Millisecond) + defer ticker.Stop() + for { + select { + // Can add a done channel or other stuff. + case <-ticker.C: + count++ + // Each ticker is 50ms so 50*200=10seconds. + if s.kv.opt.LmaxCompaction && id == 2 && count >= 200 { + tryLmaxToLmaxCompaction() + count = 0 + } else { + runOnce() + } + case <-lc.HasBeenClosed(): + return + } + } +} + +type compactionPriority struct { + level int + score float64 + adjusted float64 + dropPrefixes [][]byte + t targets +} + +func (s *levelsController) lastLevel() *levelHandler { + return s.levels[len(s.levels)-1] +} + +// pickCompactLevel determines which level to compact. +// Based on: https://github.com/facebook/rocksdb/wiki/Leveled-Compaction +func (s *levelsController) pickCompactLevels() (prios []compactionPriority) { + t := s.levelTargets() + addPriority := func(level int, score float64) { + pri := compactionPriority{ + level: level, + score: score, + adjusted: score, + t: t, + } + prios = append(prios, pri) + } + + // Add L0 priority based on the number of tables. + addPriority(0, float64(s.levels[0].numTables())/float64(s.kv.opt.NumLevelZeroTables)) + + // All other levels use size to calculate priority. + for i := 1; i < len(s.levels); i++ { + // Don't consider those tables that are already being compacted right now. + delSize := s.cstatus.delSize(i) + + l := s.levels[i] + sz := l.getTotalSize() - delSize + addPriority(i, float64(sz)/float64(t.targetSz[i])) + } + y.AssertTrue(len(prios) == len(s.levels)) + + // The following code is borrowed from PebbleDB and results in healthier LSM tree structure. + // If Li-1 has score > 1.0, then we'll divide Li-1 score by Li. If Li score is >= 1.0, then Li-1 + // score is reduced, which means we'll prioritize the compaction of lower levels (L5, L4 and so + // on) over the higher levels (L0, L1 and so on). On the other hand, if Li score is < 1.0, then + // we'll increase the priority of Li-1. + // Overall what this means is, if the bottom level is already overflowing, then de-prioritize + // compaction of the above level. If the bottom level is not full, then increase the priority of + // above level. + var prevLevel int + for level := t.baseLevel; level < len(s.levels); level++ { + if prios[prevLevel].adjusted >= 1 { + // Avoid absurdly large scores by placing a floor on the score that we'll + // adjust a level by. The value of 0.01 was chosen somewhat arbitrarily + const minScore = 0.01 + if prios[level].score >= minScore { + prios[prevLevel].adjusted /= prios[level].adjusted + } else { + prios[prevLevel].adjusted /= minScore + } + } + prevLevel = level + } + + // Pick all the levels whose original score is >= 1.0, irrespective of their adjusted score. + // We'll still sort them by their adjusted score below. Having both these scores allows us to + // make better decisions about compacting L0. If we see a score >= 1.0, we can do L0->L0 + // compactions. If the adjusted score >= 1.0, then we can do L0->Lbase compactions. + out := prios[:0] + for _, p := range prios[:len(prios)-1] { + if p.score >= 1.0 { + out = append(out, p) + } + } + prios = out + + // Sort by the adjusted score. + sort.Slice(prios, func(i, j int) bool { + return prios[i].adjusted > prios[j].adjusted + }) + return prios +} + +// checkOverlap checks if the given tables overlap with any level from the given "lev" onwards. +func (s *levelsController) checkOverlap(tables []*table.Table, lev int) bool { + kr := getKeyRange(tables...) + for i, lh := range s.levels { + if i < lev { // Skip upper levels. + continue + } + lh.RLock() + left, right := lh.overlappingTables(levelHandlerRLocked{}, kr) + lh.RUnlock() + if right-left > 0 { + return true + } + } + return false +} + +// subcompact runs a single sub-compaction, iterating over the specified key-range only. +// +// We use splits to do a single compaction concurrently. If we have >= 3 tables +// involved in the bottom level during compaction, we choose key ranges to +// split the main compaction up into sub-compactions. Each sub-compaction runs +// concurrently, only iterating over the provided key range, generating tables. +// This speeds up the compaction significantly. +func (s *levelsController) subcompact(it y.Iterator, kr keyRange, cd compactDef, + inflightBuilders *y.Throttle, res chan<- *table.Table) { + + // Check overlap of the top level with the levels which are not being + // compacted in this compaction. + hasOverlap := s.checkOverlap(cd.allTables(), cd.nextLevel.level+1) + + // Pick a discard ts, so we can discard versions below this ts. We should + // never discard any versions starting from above this timestamp, because + // that would affect the snapshot view guarantee provided by transactions. + discardTs := s.kv.orc.discardAtOrBelow() + + // Try to collect stats so that we can inform value log about GC. That would help us find which + // value log file should be GCed. + discardStats := make(map[uint32]int64) + updateStats := func(vs y.ValueStruct) { + // We don't need to store/update discard stats when badger is running in Disk-less mode. + if s.kv.opt.InMemory { + return + } + if vs.Meta&bitValuePointer > 0 { + var vp valuePointer + vp.Decode(vs.Value) + discardStats[vp.Fid] += int64(vp.Len) + } + } + + // exceedsAllowedOverlap returns true if the given key range would overlap with more than 10 + // tables from level below nextLevel (nextLevel+1). This helps avoid generating tables at Li + // with huge overlaps with Li+1. + exceedsAllowedOverlap := func(kr keyRange) bool { + n2n := cd.nextLevel.level + 1 + if n2n <= 1 || n2n >= len(s.levels) { + return false + } + n2nl := s.levels[n2n] + n2nl.RLock() + defer n2nl.RUnlock() + + l, r := n2nl.overlappingTables(levelHandlerRLocked{}, kr) + return r-l >= 10 + } + + var ( + lastKey, skipKey []byte + numBuilds, numVersions int + // Denotes if the first key is a series of duplicate keys had + // "DiscardEarlierVersions" set + firstKeyHasDiscardSet bool + ) + + addKeys := func(builder *table.Builder) { + timeStart := time.Now() + var numKeys, numSkips uint64 + var rangeCheck int + var tableKr keyRange + for ; it.Valid(); it.Next() { + // See if we need to skip the prefix. + if len(cd.dropPrefixes) > 0 && hasAnyPrefixes(it.Key(), cd.dropPrefixes) { + numSkips++ + updateStats(it.Value()) + continue + } + + // See if we need to skip this key. + if len(skipKey) > 0 { + if y.SameKey(it.Key(), skipKey) { + numSkips++ + updateStats(it.Value()) + continue + } else { + skipKey = skipKey[:0] + } + } + + if !y.SameKey(it.Key(), lastKey) { + firstKeyHasDiscardSet = false + if len(kr.right) > 0 && y.CompareKeys(it.Key(), kr.right) >= 0 { + break + } + if builder.ReachedCapacity() { + // Only break if we are on a different key, and have reached capacity. We want + // to ensure that all versions of the key are stored in the same sstable, and + // not divided across multiple tables at the same level. + break + } + lastKey = y.SafeCopy(lastKey, it.Key()) + numVersions = 0 + firstKeyHasDiscardSet = it.Value().Meta&bitDiscardEarlierVersions > 0 + + if len(tableKr.left) == 0 { + tableKr.left = y.SafeCopy(tableKr.left, it.Key()) + } + tableKr.right = lastKey + + rangeCheck++ + if rangeCheck%5000 == 0 { + // This table's range exceeds the allowed range overlap with the level after + // next. So, we stop writing to this table. If we don't do this, then we end up + // doing very expensive compactions involving too many tables. To amortize the + // cost of this check, we do it only every N keys. + if exceedsAllowedOverlap(tableKr) { + // s.kv.opt.Debugf("L%d -> L%d Breaking due to exceedsAllowedOverlap with + // kr: %s\n", cd.thisLevel.level, cd.nextLevel.level, tableKr) + break + } + } + } + + vs := it.Value() + version := y.ParseTs(it.Key()) + + isExpired := isDeletedOrExpired(vs.Meta, vs.ExpiresAt) + + // Do not discard entries inserted by merge operator. These entries will be + // discarded once they're merged + if version <= discardTs && vs.Meta&bitMergeEntry == 0 { + // Keep track of the number of versions encountered for this key. Only consider the + // versions which are below the minReadTs, otherwise, we might end up discarding the + // only valid version for a running transaction. + numVersions++ + // Keep the current version and discard all the next versions if + // - The `discardEarlierVersions` bit is set OR + // - We've already processed `NumVersionsToKeep` number of versions + // (including the current item being processed) + lastValidVersion := vs.Meta&bitDiscardEarlierVersions > 0 || + numVersions == s.kv.opt.NumVersionsToKeep + + if isExpired || lastValidVersion { + // If this version of the key is deleted or expired, skip all the rest of the + // versions. Ensure that we're only removing versions below readTs. + skipKey = y.SafeCopy(skipKey, it.Key()) + + switch { + // Add the key to the table only if it has not expired. + // We don't want to add the deleted/expired keys. + case !isExpired && lastValidVersion: + // Add this key. We have set skipKey, so the following key versions + // would be skipped. + case hasOverlap: + // If this key range has overlap with lower levels, then keep the deletion + // marker with the latest version, discarding the rest. We have set skipKey, + // so the following key versions would be skipped. + default: + // If no overlap, we can skip all the versions, by continuing here. + numSkips++ + updateStats(vs) + continue // Skip adding this key. + } + } + } + numKeys++ + var vp valuePointer + if vs.Meta&bitValuePointer > 0 { + vp.Decode(vs.Value) + } + switch { + case firstKeyHasDiscardSet: + // This key is same as the last key which had "DiscardEarlierVersions" set. The + // the next compactions will drop this key if its ts > + // discardTs (of the next compaction). + builder.AddStaleKey(it.Key(), vs, vp.Len) + case isExpired: + // If the key is expired, the next compaction will drop it if + // its ts > discardTs (of the next compaction). + builder.AddStaleKey(it.Key(), vs, vp.Len) + default: + builder.Add(it.Key(), vs, vp.Len) + } + } + s.kv.opt.Debugf("[%d] LOG Compact. Added %d keys. Skipped %d keys. Iteration took: %v", + cd.compactorId, numKeys, numSkips, time.Since(timeStart).Round(time.Millisecond)) + } // End of function: addKeys + + if len(kr.left) > 0 { + it.Seek(kr.left) + } else { + it.Rewind() + } + for it.Valid() { + if len(kr.right) > 0 && y.CompareKeys(it.Key(), kr.right) >= 0 { + break + } + + bopts := buildTableOptions(s.kv) + // Set TableSize to the target file size for that level. + bopts.TableSize = uint64(cd.t.fileSz[cd.nextLevel.level]) + builder := table.NewTableBuilder(bopts) + + // This would do the iteration and add keys to builder. + addKeys(builder) + + // It was true that it.Valid() at least once in the loop above, which means we + // called Add() at least once, and builder is not Empty(). + if builder.Empty() { + // Cleanup builder resources: + builder.Finish() + builder.Close() + continue + } + numBuilds++ + if err := inflightBuilders.Do(); err != nil { + // Can't return from here, until I decrRef all the tables that I built so far. + break + } + go func(builder *table.Builder, fileID uint64) { + var err error + defer inflightBuilders.Done(err) + defer builder.Close() + + var tbl *table.Table + if s.kv.opt.InMemory { + tbl, err = table.OpenInMemoryTable(builder.Finish(), fileID, &bopts) + } else { + fname := table.NewFilename(fileID, s.kv.opt.Dir) + tbl, err = table.CreateTable(fname, builder) + } + + // If we couldn't build the table, return fast. + if err != nil { + return + } + res <- tbl + }(builder, s.reserveFileID()) + } + s.kv.vlog.updateDiscardStats(discardStats) + s.kv.opt.Debugf("Discard stats: %v", discardStats) +} + +// compactBuildTables merges topTables and botTables to form a list of new tables. +func (s *levelsController) compactBuildTables( + lev int, cd compactDef) ([]*table.Table, func() error, error) { + + topTables := cd.top + botTables := cd.bot + + numTables := int64(len(topTables) + len(botTables)) + y.NumCompactionTablesAdd(s.kv.opt.MetricsEnabled, numTables) + defer y.NumCompactionTablesAdd(s.kv.opt.MetricsEnabled, -numTables) + + cd.span.Annotatef(nil, "Top tables count: %v Bottom tables count: %v", + len(topTables), len(botTables)) + + keepTable := func(t *table.Table) bool { + for _, prefix := range cd.dropPrefixes { + if bytes.HasPrefix(t.Smallest(), prefix) && + bytes.HasPrefix(t.Biggest(), prefix) { + // All the keys in this table have the dropPrefix. So, this + // table does not need to be in the iterator and can be + // dropped immediately. + return false + } + } + return true + } + var valid []*table.Table + for _, table := range botTables { + if keepTable(table) { + valid = append(valid, table) + } + } + + newIterator := func() []y.Iterator { + // Create iterators across all the tables involved first. + var iters []y.Iterator + switch { + case lev == 0: + iters = appendIteratorsReversed(iters, topTables, table.NOCACHE) + case len(topTables) > 0: + y.AssertTrue(len(topTables) == 1) + iters = []y.Iterator{topTables[0].NewIterator(table.NOCACHE)} + } + // Next level has level>=1 and we can use ConcatIterator as key ranges do not overlap. + return append(iters, table.NewConcatIterator(valid, table.NOCACHE)) + } + + res := make(chan *table.Table, 3) + inflightBuilders := y.NewThrottle(8 + len(cd.splits)) + for _, kr := range cd.splits { + // Initiate Do here so we can register the goroutines for buildTables too. + if err := inflightBuilders.Do(); err != nil { + s.kv.opt.Errorf("cannot start subcompaction: %+v", err) + return nil, nil, err + } + go func(kr keyRange) { + defer inflightBuilders.Done(nil) + it := table.NewMergeIterator(newIterator(), false) + defer it.Close() + s.subcompact(it, kr, cd, inflightBuilders, res) + }(kr) + } + + var newTables []*table.Table + var wg sync.WaitGroup + wg.Add(1) + go func() { + defer wg.Done() + for t := range res { + newTables = append(newTables, t) + } + }() + + // Wait for all table builders to finish and also for newTables accumulator to finish. + err := inflightBuilders.Finish() + close(res) + wg.Wait() // Wait for all tables to be picked up. + + if err == nil { + // Ensure created files' directory entries are visible. We don't mind the extra latency + // from not doing this ASAP after all file creation has finished because this is a + // background operation. + err = s.kv.syncDir(s.kv.opt.Dir) + } + + if err != nil { + // An error happened. Delete all the newly created table files (by calling DecrRef + // -- we're the only holders of a ref). + _ = decrRefs(newTables) + return nil, nil, y.Wrapf(err, "while running compactions for: %+v", cd) + } + + sort.Slice(newTables, func(i, j int) bool { + return y.CompareKeys(newTables[i].Biggest(), newTables[j].Biggest()) < 0 + }) + return newTables, func() error { return decrRefs(newTables) }, nil +} + +func buildChangeSet(cd *compactDef, newTables []*table.Table) pb.ManifestChangeSet { + changes := []*pb.ManifestChange{} + for _, table := range newTables { + changes = append(changes, + newCreateChange(table.ID(), cd.nextLevel.level, table.KeyID(), table.CompressionType())) + } + for _, table := range cd.top { + // Add a delete change only if the table is not in memory. + if !table.IsInmemory { + changes = append(changes, newDeleteChange(table.ID())) + } + } + for _, table := range cd.bot { + changes = append(changes, newDeleteChange(table.ID())) + } + return pb.ManifestChangeSet{Changes: changes} +} + +func hasAnyPrefixes(s []byte, listOfPrefixes [][]byte) bool { + for _, prefix := range listOfPrefixes { + if bytes.HasPrefix(s, prefix) { + return true + } + } + + return false +} + +func containsPrefix(table *table.Table, prefix []byte) bool { + smallValue := table.Smallest() + largeValue := table.Biggest() + if bytes.HasPrefix(smallValue, prefix) { + return true + } + if bytes.HasPrefix(largeValue, prefix) { + return true + } + isPresent := func() bool { + ti := table.NewIterator(0) + defer ti.Close() + // In table iterator's Seek, we assume that key has version in last 8 bytes. We set + // version=0 (ts=math.MaxUint64), so that we don't skip the key prefixed with prefix. + ti.Seek(y.KeyWithTs(prefix, math.MaxUint64)) + return bytes.HasPrefix(ti.Key(), prefix) + } + + if bytes.Compare(prefix, smallValue) > 0 && + bytes.Compare(prefix, largeValue) < 0 { + // There may be a case when table contains [0x0000,...., 0xffff]. If we are searching for + // k=0x0011, we should not directly infer that k is present. It may not be present. + return isPresent() + } + + return false +} + +func containsAnyPrefixes(table *table.Table, listOfPrefixes [][]byte) bool { + for _, prefix := range listOfPrefixes { + if containsPrefix(table, prefix) { + return true + } + } + + return false +} + +type compactDef struct { + span *otrace.Span + + compactorId int + t targets + p compactionPriority + thisLevel *levelHandler + nextLevel *levelHandler + + top []*table.Table + bot []*table.Table + + thisRange keyRange + nextRange keyRange + splits []keyRange + + thisSize int64 + + dropPrefixes [][]byte +} + +// addSplits can allow us to run multiple sub-compactions in parallel across the split key ranges. +func (s *levelsController) addSplits(cd *compactDef) { + cd.splits = cd.splits[:0] + + // Let's say we have 10 tables in cd.bot and min width = 3. Then, we'll pick + // 0, 1, 2 (pick), 3, 4, 5 (pick), 6, 7, 8 (pick), 9 (pick, because last table). + // This gives us 4 picks for 10 tables. + // In an edge case, 142 tables in bottom led to 48 splits. That's too many splits, because it + // then uses up a lot of memory for table builder. + // We should keep it so we have at max 5 splits. + width := int(math.Ceil(float64(len(cd.bot)) / 5.0)) + if width < 3 { + width = 3 + } + skr := cd.thisRange + skr.extend(cd.nextRange) + + addRange := func(right []byte) { + skr.right = y.Copy(right) + cd.splits = append(cd.splits, skr) + + skr.left = skr.right + } + + for i, t := range cd.bot { + // last entry in bottom table. + if i == len(cd.bot)-1 { + addRange([]byte{}) + return + } + if i%width == width-1 { + // Right is assigned ts=0. The encoding ts bytes take MaxUint64-ts, + // so, those with smaller TS will be considered larger for the same key. + // Consider the following. + // Top table is [A1...C3(deleted)] + // bot table is [B1....C2] + // It will generate a split [A1 ... C0], including any records of Key C. + right := y.KeyWithTs(y.ParseKey(t.Biggest()), 0) + addRange(right) + } + } +} + +func (cd *compactDef) lockLevels() { + cd.thisLevel.RLock() + cd.nextLevel.RLock() +} + +func (cd *compactDef) unlockLevels() { + cd.nextLevel.RUnlock() + cd.thisLevel.RUnlock() +} + +func (cd *compactDef) allTables() []*table.Table { + ret := make([]*table.Table, 0, len(cd.top)+len(cd.bot)) + ret = append(ret, cd.top...) + ret = append(ret, cd.bot...) + return ret +} + +func (s *levelsController) fillTablesL0ToL0(cd *compactDef) bool { + if cd.compactorId != 0 { + // Only compactor zero can work on this. + return false + } + + cd.nextLevel = s.levels[0] + cd.nextRange = keyRange{} + cd.bot = nil + + // Because this level and next level are both level 0, we should NOT acquire + // the read lock twice, because it can result in a deadlock. So, we don't + // call compactDef.lockLevels, instead locking the level only once and + // directly here. + // + // As per godocs on RWMutex: + // If a goroutine holds a RWMutex for reading and another goroutine might + // call Lock, no goroutine should expect to be able to acquire a read lock + // until the initial read lock is released. In particular, this prohibits + // recursive read locking. This is to ensure that the lock eventually + // becomes available; a blocked Lock call excludes new readers from + // acquiring the lock. + y.AssertTrue(cd.thisLevel.level == 0) + y.AssertTrue(cd.nextLevel.level == 0) + s.levels[0].RLock() + defer s.levels[0].RUnlock() + + s.cstatus.Lock() + defer s.cstatus.Unlock() + + top := cd.thisLevel.tables + var out []*table.Table + now := time.Now() + for _, t := range top { + if t.Size() >= 2*cd.t.fileSz[0] { + // This file is already big, don't include it. + continue + } + if now.Sub(t.CreatedAt) < 10*time.Second { + // Just created it 10s ago. Don't pick for compaction. + continue + } + if _, beingCompacted := s.cstatus.tables[t.ID()]; beingCompacted { + continue + } + out = append(out, t) + } + + if len(out) < 4 { + // If we don't have enough tables to merge in L0, don't do it. + return false + } + cd.thisRange = infRange + cd.top = out + + // Avoid any other L0 -> Lbase from happening, while this is going on. + thisLevel := s.cstatus.levels[cd.thisLevel.level] + thisLevel.ranges = append(thisLevel.ranges, infRange) + for _, t := range out { + s.cstatus.tables[t.ID()] = struct{}{} + } + + // For L0->L0 compaction, we set the target file size to max, so the output is always one file. + // This significantly decreases the L0 table stalls and improves the performance. + cd.t.fileSz[0] = math.MaxUint32 + return true +} + +func (s *levelsController) fillTablesL0ToLbase(cd *compactDef) bool { + if cd.nextLevel.level == 0 { + panic("Base level can't be zero.") + } + // We keep cd.p.adjusted > 0.0 here to allow functions in db.go to artificially trigger + // L0->Lbase compactions. Those functions wouldn't be setting the adjusted score. + if cd.p.adjusted > 0.0 && cd.p.adjusted < 1.0 { + // Do not compact to Lbase if adjusted score is less than 1.0. + return false + } + cd.lockLevels() + defer cd.unlockLevels() + + top := cd.thisLevel.tables + if len(top) == 0 { + return false + } + + var out []*table.Table + if len(cd.dropPrefixes) > 0 { + // Use all tables if drop prefix is set. We don't want to compact only a + // sub-range. We want to compact all the tables. + out = top + + } else { + var kr keyRange + // cd.top[0] is the oldest file. So we start from the oldest file first. + for _, t := range top { + dkr := getKeyRange(t) + if kr.overlapsWith(dkr) { + out = append(out, t) + kr.extend(dkr) + } else { + break + } + } + } + cd.thisRange = getKeyRange(out...) + cd.top = out + + left, right := cd.nextLevel.overlappingTables(levelHandlerRLocked{}, cd.thisRange) + cd.bot = make([]*table.Table, right-left) + copy(cd.bot, cd.nextLevel.tables[left:right]) + + if len(cd.bot) == 0 { + cd.nextRange = cd.thisRange + } else { + cd.nextRange = getKeyRange(cd.bot...) + } + return s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) +} + +// fillTablesL0 would try to fill tables from L0 to be compacted with Lbase. If +// it can not do that, it would try to compact tables from L0 -> L0. +// +// Say L0 has 10 tables. +// fillTablesL0ToLbase picks up 5 tables to compact from L0 -> L5. +// Next call to fillTablesL0 would run L0ToLbase again, which fails this time. +// So, instead, we run fillTablesL0ToL0, which picks up rest of the 5 tables to +// be compacted within L0. Additionally, it would set the compaction range in +// cstatus to inf, so no other L0 -> Lbase compactions can happen. +// Thus, L0 -> L0 must finish for the next L0 -> Lbase to begin. +func (s *levelsController) fillTablesL0(cd *compactDef) bool { + if ok := s.fillTablesL0ToLbase(cd); ok { + return true + } + return s.fillTablesL0ToL0(cd) +} + +// sortByStaleData sorts tables based on the amount of stale data they have. +// This is useful in removing tombstones. +func (s *levelsController) sortByStaleDataSize(tables []*table.Table, cd *compactDef) { + if len(tables) == 0 || cd.nextLevel == nil { + return + } + + sort.Slice(tables, func(i, j int) bool { + return tables[i].StaleDataSize() > tables[j].StaleDataSize() + }) +} + +// sortByHeuristic sorts tables in increasing order of MaxVersion, so we +// compact older tables first. +func (s *levelsController) sortByHeuristic(tables []*table.Table, cd *compactDef) { + if len(tables) == 0 || cd.nextLevel == nil { + return + } + + // Sort tables by max version. This is what RocksDB does. + sort.Slice(tables, func(i, j int) bool { + return tables[i].MaxVersion() < tables[j].MaxVersion() + }) +} + +// This function should be called with lock on levels. +func (s *levelsController) fillMaxLevelTables(tables []*table.Table, cd *compactDef) bool { + sortedTables := make([]*table.Table, len(tables)) + copy(sortedTables, tables) + s.sortByStaleDataSize(sortedTables, cd) + + if len(sortedTables) > 0 && sortedTables[0].StaleDataSize() == 0 { + // This is a maxLevel to maxLevel compaction and we don't have any stale data. + return false + } + cd.bot = []*table.Table{} + collectBotTables := func(t *table.Table, needSz int64) { + totalSize := t.Size() + + j := sort.Search(len(tables), func(i int) bool { + return y.CompareKeys(tables[i].Smallest(), t.Smallest()) >= 0 + }) + y.AssertTrue(tables[j].ID() == t.ID()) + j++ + // Collect tables until we reach the the required size. + for j < len(tables) { + newT := tables[j] + totalSize += newT.Size() + + if totalSize >= needSz { + break + } + cd.bot = append(cd.bot, newT) + cd.nextRange.extend(getKeyRange(newT)) + j++ + } + } + now := time.Now() + for _, t := range sortedTables { + // If the maxVersion is above the discardTs, we won't clean anything in + // the compaction. So skip this table. + if t.MaxVersion() > s.kv.orc.discardAtOrBelow() { + continue + } + if now.Sub(t.CreatedAt) < time.Hour { + // Just created it an hour ago. Don't pick for compaction. + continue + } + // If the stale data size is less than 10 MB, it might not be worth + // rewriting the table. Skip it. + if t.StaleDataSize() < 10<<20 { + continue + } + + cd.thisSize = t.Size() + cd.thisRange = getKeyRange(t) + // Set the next range as the same as the current range. If we don't do + // this, we won't be able to run more than one max level compactions. + cd.nextRange = cd.thisRange + // If we're already compacting this range, don't do anything. + if s.cstatus.overlapsWith(cd.thisLevel.level, cd.thisRange) { + continue + } + + // Found a valid table! + cd.top = []*table.Table{t} + + needFileSz := cd.t.fileSz[cd.thisLevel.level] + // The table size is what we want so no need to collect more tables. + if t.Size() >= needFileSz { + break + } + // TableSize is less than what we want. Collect more tables for compaction. + // If the level has multiple small tables, we collect all of them + // together to form a bigger table. + collectBotTables(t, needFileSz) + if !s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) { + cd.bot = cd.bot[:0] + cd.nextRange = keyRange{} + continue + } + return true + } + if len(cd.top) == 0 { + return false + } + + return s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) +} + +func (s *levelsController) fillTables(cd *compactDef) bool { + cd.lockLevels() + defer cd.unlockLevels() + + tables := make([]*table.Table, len(cd.thisLevel.tables)) + copy(tables, cd.thisLevel.tables) + if len(tables) == 0 { + return false + } + // We're doing a maxLevel to maxLevel compaction. Pick tables based on the stale data size. + if cd.thisLevel.isLastLevel() { + return s.fillMaxLevelTables(tables, cd) + } + // We pick tables, so we compact older tables first. This is similar to + // kOldestLargestSeqFirst in RocksDB. + s.sortByHeuristic(tables, cd) + + for _, t := range tables { + cd.thisSize = t.Size() + cd.thisRange = getKeyRange(t) + // If we're already compacting this range, don't do anything. + if s.cstatus.overlapsWith(cd.thisLevel.level, cd.thisRange) { + continue + } + cd.top = []*table.Table{t} + left, right := cd.nextLevel.overlappingTables(levelHandlerRLocked{}, cd.thisRange) + + cd.bot = make([]*table.Table, right-left) + copy(cd.bot, cd.nextLevel.tables[left:right]) + + if len(cd.bot) == 0 { + cd.bot = []*table.Table{} + cd.nextRange = cd.thisRange + if !s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) { + continue + } + return true + } + cd.nextRange = getKeyRange(cd.bot...) + + if s.cstatus.overlapsWith(cd.nextLevel.level, cd.nextRange) { + continue + } + if !s.cstatus.compareAndAdd(thisAndNextLevelRLocked{}, *cd) { + continue + } + return true + } + return false +} + +func (s *levelsController) runCompactDef(id, l int, cd compactDef) (err error) { + if len(cd.t.fileSz) == 0 { + return errors.New("Filesizes cannot be zero. Targets are not set") + } + timeStart := time.Now() + + thisLevel := cd.thisLevel + nextLevel := cd.nextLevel + + y.AssertTrue(len(cd.splits) == 0) + if thisLevel.level == nextLevel.level { + // don't do anything for L0 -> L0 and Lmax -> Lmax. + } else { + s.addSplits(&cd) + } + if len(cd.splits) == 0 { + cd.splits = append(cd.splits, keyRange{}) + } + + // Table should never be moved directly between levels, + // always be rewritten to allow discarding invalid versions. + + newTables, decr, err := s.compactBuildTables(l, cd) + if err != nil { + return err + } + defer func() { + // Only assign to err, if it's not already nil. + if decErr := decr(); err == nil { + err = decErr + } + }() + changeSet := buildChangeSet(&cd, newTables) + + // We write to the manifest _before_ we delete files (and after we created files) + if err := s.kv.manifest.addChanges(changeSet.Changes); err != nil { + return err + } + + getSizes := func(tables []*table.Table) int64 { + size := int64(0) + for _, i := range tables { + size += i.Size() + } + return size + } + + sizeNewTables := int64(0) + sizeOldTables := int64(0) + if s.kv.opt.MetricsEnabled { + sizeNewTables = getSizes(newTables) + sizeOldTables = getSizes(cd.bot) + getSizes(cd.top) + y.NumBytesCompactionWrittenAdd(s.kv.opt.MetricsEnabled, nextLevel.strLevel, sizeNewTables) + } + + // See comment earlier in this function about the ordering of these ops, and the order in which + // we access levels when reading. + if err := nextLevel.replaceTables(cd.bot, newTables); err != nil { + return err + } + if err := thisLevel.deleteTables(cd.top); err != nil { + return err + } + + // Note: For level 0, while doCompact is running, it is possible that new tables are added. + // However, the tables are added only to the end, so it is ok to just delete the first table. + + from := append(tablesToString(cd.top), tablesToString(cd.bot)...) + to := tablesToString(newTables) + if dur := time.Since(timeStart); dur > 2*time.Second { + var expensive string + if dur > time.Second { + expensive = " [E]" + } + s.kv.opt.Infof("[%d]%s LOG Compact %d->%d (%d, %d -> %d tables with %d splits)."+ + " [%s] -> [%s], took %v\n, deleted %d bytes", + id, expensive, thisLevel.level, nextLevel.level, len(cd.top), len(cd.bot), + len(newTables), len(cd.splits), strings.Join(from, " "), strings.Join(to, " "), + dur.Round(time.Millisecond), sizeOldTables-sizeNewTables) + } + + if cd.thisLevel.level != 0 && len(newTables) > 2*s.kv.opt.LevelSizeMultiplier { + s.kv.opt.Infof("This Range (numTables: %d)\nLeft:\n%s\nRight:\n%s\n", + len(cd.top), hex.Dump(cd.thisRange.left), hex.Dump(cd.thisRange.right)) + s.kv.opt.Infof("Next Range (numTables: %d)\nLeft:\n%s\nRight:\n%s\n", + len(cd.bot), hex.Dump(cd.nextRange.left), hex.Dump(cd.nextRange.right)) + } + return nil +} + +func tablesToString(tables []*table.Table) []string { + var res []string + for _, t := range tables { + res = append(res, fmt.Sprintf("%05d", t.ID())) + } + res = append(res, ".") + return res +} + +var errFillTables = errors.New("Unable to fill tables") + +// doCompact picks some table on level l and compacts it away to the next level. +func (s *levelsController) doCompact(id int, p compactionPriority) error { + l := p.level + y.AssertTrue(l < s.kv.opt.MaxLevels) // Sanity check. + if p.t.baseLevel == 0 { + p.t = s.levelTargets() + } + + _, span := otrace.StartSpan(context.Background(), "Badger.Compaction") + defer span.End() + + cd := compactDef{ + compactorId: id, + span: span, + p: p, + t: p.t, + thisLevel: s.levels[l], + dropPrefixes: p.dropPrefixes, + } + + // While picking tables to be compacted, both levels' tables are expected to + // remain unchanged. + if l == 0 { + cd.nextLevel = s.levels[p.t.baseLevel] + if !s.fillTablesL0(&cd) { + return errFillTables + } + } else { + cd.nextLevel = cd.thisLevel + // We're not compacting the last level so pick the next level. + if !cd.thisLevel.isLastLevel() { + cd.nextLevel = s.levels[l+1] + } + if !s.fillTables(&cd) { + return errFillTables + } + } + defer s.cstatus.delete(cd) // Remove the ranges from compaction status. + + span.Annotatef(nil, "Compaction: %+v", cd) + if err := s.runCompactDef(id, l, cd); err != nil { + // This compaction couldn't be done successfully. + s.kv.opt.Warningf("[Compactor: %d] LOG Compact FAILED with error: %+v: %+v", id, err, cd) + return err + } + + s.kv.opt.Debugf("[Compactor: %d] Compaction for level: %d DONE", id, cd.thisLevel.level) + return nil +} + +func (s *levelsController) addLevel0Table(t *table.Table) error { + // Add table to manifest file only if it is not opened in memory. We don't want to add a table + // to the manifest file if it exists only in memory. + if !t.IsInmemory { + // We update the manifest _before_ the table becomes part of a levelHandler, because at that + // point it could get used in some compaction. This ensures the manifest file gets updated in + // the proper order. (That means this update happens before that of some compaction which + // deletes the table.) + err := s.kv.manifest.addChanges([]*pb.ManifestChange{ + newCreateChange(t.ID(), 0, t.KeyID(), t.CompressionType()), + }) + if err != nil { + return err + } + } + + for !s.levels[0].tryAddLevel0Table(t) { + // Before we unstall, we need to make sure that level 0 is healthy. + timeStart := time.Now() + for s.levels[0].numTables() >= s.kv.opt.NumLevelZeroTablesStall { + time.Sleep(10 * time.Millisecond) + } + dur := time.Since(timeStart) + if dur > time.Second { + s.kv.opt.Infof("L0 was stalled for %s\n", dur.Round(time.Millisecond)) + } + s.l0stallsMs.Add(int64(dur.Round(time.Millisecond))) + } + + return nil +} + +func (s *levelsController) close() error { + err := s.cleanupLevels() + return y.Wrap(err, "levelsController.Close") +} + +// get searches for a given key in all the levels of the LSM tree. It returns +// key version <= the expected version (version in key). If not found, +// it returns an empty y.ValueStruct. +func (s *levelsController) get(key []byte, maxVs y.ValueStruct, startLevel int) ( + y.ValueStruct, error) { + if s.kv.IsClosed() { + return y.ValueStruct{}, ErrDBClosed + } + // It's important that we iterate the levels from 0 on upward. The reason is, if we iterated + // in opposite order, or in parallel (naively calling all the h.RLock() in some order) we could + // read level L's tables post-compaction and level L+1's tables pre-compaction. (If we do + // parallelize this, we will need to call the h.RLock() function by increasing order of level + // number.) + version := y.ParseTs(key) + for _, h := range s.levels { + // Ignore all levels below startLevel. This is useful for GC when L0 is kept in memory. + if h.level < startLevel { + continue + } + vs, err := h.get(key) // Calls h.RLock() and h.RUnlock(). + if err != nil { + return y.ValueStruct{}, y.Wrapf(err, "get key: %q", key) + } + if vs.Value == nil && vs.Meta == 0 { + continue + } + y.NumBytesReadsLSMAdd(s.kv.opt.MetricsEnabled, int64(len(vs.Value))) + if vs.Version == version { + return vs, nil + } + if maxVs.Version < vs.Version { + maxVs = vs + } + } + if len(maxVs.Value) > 0 { + y.NumGetsWithResultsAdd(s.kv.opt.MetricsEnabled, 1) + } + return maxVs, nil +} + +func appendIteratorsReversed(out []y.Iterator, th []*table.Table, opt int) []y.Iterator { + for i := len(th) - 1; i >= 0; i-- { + // This will increment the reference of the table handler. + out = append(out, th[i].NewIterator(opt)) + } + return out +} + +// appendIterators appends iterators to an array of iterators, for merging. +// Note: This obtains references for the table handlers. Remember to close these iterators. +func (s *levelsController) appendIterators( + iters []y.Iterator, opt *IteratorOptions) []y.Iterator { + // Just like with get, it's important we iterate the levels from 0 on upward, to avoid missing + // data when there's a compaction. + for _, level := range s.levels { + iters = level.appendIterators(iters, opt) + } + return iters +} + +// TableInfo represents the information about a table. +type TableInfo struct { + ID uint64 + Level int + Left []byte + Right []byte + KeyCount uint32 // Number of keys in the table + OnDiskSize uint32 + StaleDataSize uint32 + UncompressedSize uint32 + MaxVersion uint64 + IndexSz int + BloomFilterSize int +} + +func (s *levelsController) getTableInfo() (result []TableInfo) { + for _, l := range s.levels { + l.RLock() + for _, t := range l.tables { + info := TableInfo{ + ID: t.ID(), + Level: l.level, + Left: t.Smallest(), + Right: t.Biggest(), + KeyCount: t.KeyCount(), + OnDiskSize: t.OnDiskSize(), + StaleDataSize: t.StaleDataSize(), + IndexSz: t.IndexSize(), + BloomFilterSize: t.BloomFilterSize(), + UncompressedSize: t.UncompressedSize(), + MaxVersion: t.MaxVersion(), + } + result = append(result, info) + } + l.RUnlock() + } + sort.Slice(result, func(i, j int) bool { + if result[i].Level != result[j].Level { + return result[i].Level < result[j].Level + } + return result[i].ID < result[j].ID + }) + return +} + +type LevelInfo struct { + Level int + NumTables int + Size int64 + TargetSize int64 + TargetFileSize int64 + IsBaseLevel bool + Score float64 + Adjusted float64 + StaleDatSize int64 +} + +func (s *levelsController) getLevelInfo() []LevelInfo { + t := s.levelTargets() + prios := s.pickCompactLevels() + result := make([]LevelInfo, len(s.levels)) + for i, l := range s.levels { + l.RLock() + result[i].Level = i + result[i].Size = l.totalSize + result[i].NumTables = len(l.tables) + result[i].StaleDatSize = l.totalStaleSize + + l.RUnlock() + + result[i].TargetSize = t.targetSz[i] + result[i].TargetFileSize = t.fileSz[i] + result[i].IsBaseLevel = t.baseLevel == i + } + for _, p := range prios { + result[p.level].Score = p.score + result[p.level].Adjusted = p.adjusted + } + return result +} + +// verifyChecksum verifies checksum for all tables on all levels. +func (s *levelsController) verifyChecksum() error { + var tables []*table.Table + for _, l := range s.levels { + l.RLock() + tables = tables[:0] + for _, t := range l.tables { + tables = append(tables, t) + t.IncrRef() + } + l.RUnlock() + + for _, t := range tables { + errChkVerify := t.VerifyChecksum() + if err := t.DecrRef(); err != nil { + s.kv.opt.Errorf("unable to decrease reference of table: %s while "+ + "verifying checksum with error: %s", t.Filename(), err) + } + + if errChkVerify != nil { + return errChkVerify + } + } + } + + return nil +} + +// Returns the sorted list of splits for all the levels and tables based +// on the block offsets. +func (s *levelsController) keySplits(numPerTable int, prefix []byte) []string { + splits := make([]string, 0) + for _, l := range s.levels { + l.RLock() + for _, t := range l.tables { + tableSplits := t.KeySplits(numPerTable, prefix) + splits = append(splits, tableSplits...) + } + l.RUnlock() + } + sort.Strings(splits) + return splits +} diff --git a/vendor/github.com/dgraph-io/badger/v4/logger.go b/vendor/github.com/dgraph-io/badger/v4/logger.go new file mode 100644 index 00000000000..c7b4cd6c1f4 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/logger.go @@ -0,0 +1,105 @@ +/* + * Copyright 2018 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "log" + "os" +) + +// Logger is implemented by any logging system that is used for standard logs. +type Logger interface { + Errorf(string, ...interface{}) + Warningf(string, ...interface{}) + Infof(string, ...interface{}) + Debugf(string, ...interface{}) +} + +// Errorf logs an ERROR log message to the logger specified in opts or to the +// global logger if no logger is specified in opts. +func (opt *Options) Errorf(format string, v ...interface{}) { + if opt.Logger == nil { + return + } + opt.Logger.Errorf(format, v...) +} + +// Infof logs an INFO message to the logger specified in opts. +func (opt *Options) Infof(format string, v ...interface{}) { + if opt.Logger == nil { + return + } + opt.Logger.Infof(format, v...) +} + +// Warningf logs a WARNING message to the logger specified in opts. +func (opt *Options) Warningf(format string, v ...interface{}) { + if opt.Logger == nil { + return + } + opt.Logger.Warningf(format, v...) +} + +// Debugf logs a DEBUG message to the logger specified in opts. +func (opt *Options) Debugf(format string, v ...interface{}) { + if opt.Logger == nil { + return + } + opt.Logger.Debugf(format, v...) +} + +type loggingLevel int + +const ( + DEBUG loggingLevel = iota + INFO + WARNING + ERROR +) + +type defaultLog struct { + *log.Logger + level loggingLevel +} + +func defaultLogger(level loggingLevel) *defaultLog { + return &defaultLog{Logger: log.New(os.Stderr, "badger ", log.LstdFlags), level: level} +} + +func (l *defaultLog) Errorf(f string, v ...interface{}) { + if l.level <= ERROR { + l.Printf("ERROR: "+f, v...) + } +} + +func (l *defaultLog) Warningf(f string, v ...interface{}) { + if l.level <= WARNING { + l.Printf("WARNING: "+f, v...) + } +} + +func (l *defaultLog) Infof(f string, v ...interface{}) { + if l.level <= INFO { + l.Printf("INFO: "+f, v...) + } +} + +func (l *defaultLog) Debugf(f string, v ...interface{}) { + if l.level <= DEBUG { + l.Printf("DEBUG: "+f, v...) + } +} diff --git a/vendor/github.com/dgraph-io/badger/v4/managed_db.go b/vendor/github.com/dgraph-io/badger/v4/managed_db.go new file mode 100644 index 00000000000..23c79884578 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/managed_db.go @@ -0,0 +1,89 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +// OpenManaged returns a new DB, which allows more control over setting +// transaction timestamps, aka managed mode. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +func OpenManaged(opts Options) (*DB, error) { + opts.managedTxns = true + return Open(opts) +} + +// NewTransactionAt follows the same logic as DB.NewTransaction(), but uses the +// provided read timestamp. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +func (db *DB) NewTransactionAt(readTs uint64, update bool) *Txn { + if !db.opt.managedTxns { + panic("Cannot use NewTransactionAt with managedDB=false. Use NewTransaction instead.") + } + txn := db.newTransaction(update, true) + txn.readTs = readTs + return txn +} + +// NewWriteBatchAt is similar to NewWriteBatch but it allows user to set the commit timestamp. +// NewWriteBatchAt is supposed to be used only in the managed mode. +func (db *DB) NewWriteBatchAt(commitTs uint64) *WriteBatch { + if !db.opt.managedTxns { + panic("cannot use NewWriteBatchAt with managedDB=false. Use NewWriteBatch instead") + } + + wb := db.newWriteBatch(true) + wb.commitTs = commitTs + wb.txn.commitTs = commitTs + return wb +} +func (db *DB) NewManagedWriteBatch() *WriteBatch { + if !db.opt.managedTxns { + panic("cannot use NewManagedWriteBatch with managedDB=false. Use NewWriteBatch instead") + } + + wb := db.newWriteBatch(true) + return wb +} + +// CommitAt commits the transaction, following the same logic as Commit(), but +// at the given commit timestamp. This will panic if not used with managed transactions. +// +// This is only useful for databases built on top of Badger (like Dgraph), and +// can be ignored by most users. +func (txn *Txn) CommitAt(commitTs uint64, callback func(error)) error { + if !txn.db.opt.managedTxns { + panic("Cannot use CommitAt with managedDB=false. Use Commit instead.") + } + txn.commitTs = commitTs + if callback == nil { + return txn.Commit() + } + txn.CommitWith(callback) + return nil +} + +// SetDiscardTs sets a timestamp at or below which, any invalid or deleted +// versions can be discarded from the LSM tree, and thence from the value log to +// reclaim disk space. Can only be used with managed transactions. +func (db *DB) SetDiscardTs(ts uint64) { + if !db.opt.managedTxns { + panic("Cannot use SetDiscardTs with managedDB=false.") + } + db.orc.setDiscardTs(ts) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/manifest.go b/vendor/github.com/dgraph-io/badger/v4/manifest.go new file mode 100644 index 00000000000..e681ae0a010 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/manifest.go @@ -0,0 +1,499 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bufio" + "bytes" + "encoding/binary" + "fmt" + "hash/crc32" + "io" + "math" + "os" + "path/filepath" + "sync" + + "github.com/golang/protobuf/proto" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/options" + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" +) + +// Manifest represents the contents of the MANIFEST file in a Badger store. +// +// The MANIFEST file describes the startup state of the db -- all LSM files and what level they're +// at. +// +// It consists of a sequence of ManifestChangeSet objects. Each of these is treated atomically, +// and contains a sequence of ManifestChange's (file creations/deletions) which we use to +// reconstruct the manifest at startup. +type Manifest struct { + Levels []levelManifest + Tables map[uint64]TableManifest + + // Contains total number of creation and deletion changes in the manifest -- used to compute + // whether it'd be useful to rewrite the manifest. + Creations int + Deletions int +} + +func createManifest() Manifest { + levels := make([]levelManifest, 0) + return Manifest{ + Levels: levels, + Tables: make(map[uint64]TableManifest), + } +} + +// levelManifest contains information about LSM tree levels +// in the MANIFEST file. +type levelManifest struct { + Tables map[uint64]struct{} // Set of table id's +} + +// TableManifest contains information about a specific table +// in the LSM tree. +type TableManifest struct { + Level uint8 + KeyID uint64 + Compression options.CompressionType +} + +// manifestFile holds the file pointer (and other info) about the manifest file, which is a log +// file we append to. +type manifestFile struct { + fp *os.File + directory string + + // The external magic number used by the application running badger. + externalMagic uint16 + + // We make this configurable so that unit tests can hit rewrite() code quickly + deletionsRewriteThreshold int + + // Guards appends, which includes access to the manifest field. + appendLock sync.Mutex + + // Used to track the current state of the manifest, used when rewriting. + manifest Manifest + + // Used to indicate if badger was opened in InMemory mode. + inMemory bool +} + +const ( + // ManifestFilename is the filename for the manifest file. + ManifestFilename = "MANIFEST" + manifestRewriteFilename = "MANIFEST-REWRITE" + manifestDeletionsRewriteThreshold = 10000 + manifestDeletionsRatio = 10 +) + +// asChanges returns a sequence of changes that could be used to recreate the Manifest in its +// present state. +func (m *Manifest) asChanges() []*pb.ManifestChange { + changes := make([]*pb.ManifestChange, 0, len(m.Tables)) + for id, tm := range m.Tables { + changes = append(changes, newCreateChange(id, int(tm.Level), tm.KeyID, tm.Compression)) + } + return changes +} + +func (m *Manifest) clone() Manifest { + changeSet := pb.ManifestChangeSet{Changes: m.asChanges()} + ret := createManifest() + y.Check(applyChangeSet(&ret, &changeSet)) + return ret +} + +// openOrCreateManifestFile opens a Badger manifest file if it exists, or creates one if +// doesn’t exists. +func openOrCreateManifestFile(opt Options) ( + ret *manifestFile, result Manifest, err error) { + if opt.InMemory { + return &manifestFile{inMemory: true}, Manifest{}, nil + } + return helpOpenOrCreateManifestFile(opt.Dir, opt.ReadOnly, opt.ExternalMagicVersion, + manifestDeletionsRewriteThreshold) +} + +func helpOpenOrCreateManifestFile(dir string, readOnly bool, extMagic uint16, + deletionsThreshold int) (*manifestFile, Manifest, error) { + + path := filepath.Join(dir, ManifestFilename) + var flags y.Flags + if readOnly { + flags |= y.ReadOnly + } + fp, err := y.OpenExistingFile(path, flags) // We explicitly sync in addChanges, outside the lock. + if err != nil { + if !os.IsNotExist(err) { + return nil, Manifest{}, err + } + if readOnly { + return nil, Manifest{}, fmt.Errorf("no manifest found, required for read-only db") + } + m := createManifest() + fp, netCreations, err := helpRewrite(dir, &m, extMagic) + if err != nil { + return nil, Manifest{}, err + } + y.AssertTrue(netCreations == 0) + mf := &manifestFile{ + fp: fp, + directory: dir, + externalMagic: extMagic, + manifest: m.clone(), + deletionsRewriteThreshold: deletionsThreshold, + } + return mf, m, nil + } + + manifest, truncOffset, err := ReplayManifestFile(fp, extMagic) + if err != nil { + _ = fp.Close() + return nil, Manifest{}, err + } + + if !readOnly { + // Truncate file so we don't have a half-written entry at the end. + if err := fp.Truncate(truncOffset); err != nil { + _ = fp.Close() + return nil, Manifest{}, err + } + } + if _, err = fp.Seek(0, io.SeekEnd); err != nil { + _ = fp.Close() + return nil, Manifest{}, err + } + + mf := &manifestFile{ + fp: fp, + directory: dir, + externalMagic: extMagic, + manifest: manifest.clone(), + deletionsRewriteThreshold: deletionsThreshold, + } + return mf, manifest, nil +} + +func (mf *manifestFile) close() error { + if mf.inMemory { + return nil + } + return mf.fp.Close() +} + +// addChanges writes a batch of changes, atomically, to the file. By "atomically" that means when +// we replay the MANIFEST file, we'll either replay all the changes or none of them. (The truth of +// this depends on the filesystem -- some might append garbage data if a system crash happens at +// the wrong time.) +func (mf *manifestFile) addChanges(changesParam []*pb.ManifestChange) error { + if mf.inMemory { + return nil + } + changes := pb.ManifestChangeSet{Changes: changesParam} + buf, err := proto.Marshal(&changes) + if err != nil { + return err + } + + // Maybe we could use O_APPEND instead (on certain file systems) + mf.appendLock.Lock() + defer mf.appendLock.Unlock() + if err := applyChangeSet(&mf.manifest, &changes); err != nil { + return err + } + // Rewrite manifest if it'd shrink by 1/10 and it's big enough to care + if mf.manifest.Deletions > mf.deletionsRewriteThreshold && + mf.manifest.Deletions > manifestDeletionsRatio*(mf.manifest.Creations-mf.manifest.Deletions) { + if err := mf.rewrite(); err != nil { + return err + } + } else { + var lenCrcBuf [8]byte + binary.BigEndian.PutUint32(lenCrcBuf[0:4], uint32(len(buf))) + binary.BigEndian.PutUint32(lenCrcBuf[4:8], crc32.Checksum(buf, y.CastagnoliCrcTable)) + buf = append(lenCrcBuf[:], buf...) + if _, err := mf.fp.Write(buf); err != nil { + return err + } + } + + return syncFunc(mf.fp) +} + +// this function is saved here to allow injection of fake filesystem latency at test time. +var syncFunc = func(f *os.File) error { return f.Sync() } + +// Has to be 4 bytes. The value can never change, ever, anyway. +var magicText = [4]byte{'B', 'd', 'g', 'r'} + +// The magic version number. It is allocated 2 bytes, so it's value must be <= math.MaxUint16 +const badgerMagicVersion = 8 + +func helpRewrite(dir string, m *Manifest, extMagic uint16) (*os.File, int, error) { + rewritePath := filepath.Join(dir, manifestRewriteFilename) + // We explicitly sync. + fp, err := y.OpenTruncFile(rewritePath, false) + if err != nil { + return nil, 0, err + } + + // magic bytes are structured as + // +---------------------+-------------------------+-----------------------+ + // | magicText (4 bytes) | externalMagic (2 bytes) | badgerMagic (2 bytes) | + // +---------------------+-------------------------+-----------------------+ + + y.AssertTrue(badgerMagicVersion <= math.MaxUint16) + buf := make([]byte, 8) + copy(buf[0:4], magicText[:]) + binary.BigEndian.PutUint16(buf[4:6], extMagic) + binary.BigEndian.PutUint16(buf[6:8], badgerMagicVersion) + + netCreations := len(m.Tables) + changes := m.asChanges() + set := pb.ManifestChangeSet{Changes: changes} + + changeBuf, err := proto.Marshal(&set) + if err != nil { + fp.Close() + return nil, 0, err + } + var lenCrcBuf [8]byte + binary.BigEndian.PutUint32(lenCrcBuf[0:4], uint32(len(changeBuf))) + binary.BigEndian.PutUint32(lenCrcBuf[4:8], crc32.Checksum(changeBuf, y.CastagnoliCrcTable)) + buf = append(buf, lenCrcBuf[:]...) + buf = append(buf, changeBuf...) + if _, err := fp.Write(buf); err != nil { + fp.Close() + return nil, 0, err + } + if err := fp.Sync(); err != nil { + fp.Close() + return nil, 0, err + } + + // In Windows the files should be closed before doing a Rename. + if err = fp.Close(); err != nil { + return nil, 0, err + } + manifestPath := filepath.Join(dir, ManifestFilename) + if err := os.Rename(rewritePath, manifestPath); err != nil { + return nil, 0, err + } + fp, err = y.OpenExistingFile(manifestPath, 0) + if err != nil { + return nil, 0, err + } + if _, err := fp.Seek(0, io.SeekEnd); err != nil { + fp.Close() + return nil, 0, err + } + if err := syncDir(dir); err != nil { + fp.Close() + return nil, 0, err + } + + return fp, netCreations, nil +} + +// Must be called while appendLock is held. +func (mf *manifestFile) rewrite() error { + // In Windows the files should be closed before doing a Rename. + if err := mf.fp.Close(); err != nil { + return err + } + fp, netCreations, err := helpRewrite(mf.directory, &mf.manifest, mf.externalMagic) + if err != nil { + return err + } + mf.fp = fp + mf.manifest.Creations = netCreations + mf.manifest.Deletions = 0 + + return nil +} + +type countingReader struct { + wrapped *bufio.Reader + count int64 +} + +func (r *countingReader) Read(p []byte) (n int, err error) { + n, err = r.wrapped.Read(p) + r.count += int64(n) + return +} + +func (r *countingReader) ReadByte() (b byte, err error) { + b, err = r.wrapped.ReadByte() + if err == nil { + r.count++ + } + return +} + +var ( + errBadMagic = errors.New("manifest has bad magic") + errBadChecksum = errors.New("manifest has checksum mismatch") +) + +// ReplayManifestFile reads the manifest file and constructs two manifest objects. (We need one +// immutable copy and one mutable copy of the manifest. Easiest way is to construct two of them.) +// Also, returns the last offset after a completely read manifest entry -- the file must be +// truncated at that point before further appends are made (if there is a partial entry after +// that). In normal conditions, truncOffset is the file size. +func ReplayManifestFile(fp *os.File, extMagic uint16) (Manifest, int64, error) { + r := countingReader{wrapped: bufio.NewReader(fp)} + + var magicBuf [8]byte + if _, err := io.ReadFull(&r, magicBuf[:]); err != nil { + return Manifest{}, 0, errBadMagic + } + if !bytes.Equal(magicBuf[0:4], magicText[:]) { + return Manifest{}, 0, errBadMagic + } + + extVersion := y.BytesToU16(magicBuf[4:6]) + version := y.BytesToU16(magicBuf[6:8]) + + if version != badgerMagicVersion { + return Manifest{}, 0, + //nolint:lll + fmt.Errorf("manifest has unsupported version: %d (we support %d).\n"+ + "Please see https://dgraph.io/docs/badger/faq/#i-see-manifest-has-unsupported-version-x-we-support-y-error"+ + " on how to fix this.", + version, badgerMagicVersion) + } + if extVersion != extMagic { + return Manifest{}, 0, + fmt.Errorf("Cannot open DB because the external magic number doesn't match. "+ + "Expected: %d, version present in manifest: %d\n", extMagic, extVersion) + } + + stat, err := fp.Stat() + if err != nil { + return Manifest{}, 0, err + } + + build := createManifest() + var offset int64 + for { + offset = r.count + var lenCrcBuf [8]byte + _, err := io.ReadFull(&r, lenCrcBuf[:]) + if err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + break + } + return Manifest{}, 0, err + } + length := y.BytesToU32(lenCrcBuf[0:4]) + // Sanity check to ensure we don't over-allocate memory. + if length > uint32(stat.Size()) { + return Manifest{}, 0, errors.Errorf( + "Buffer length: %d greater than file size: %d. Manifest file might be corrupted", + length, stat.Size()) + } + var buf = make([]byte, length) + if _, err := io.ReadFull(&r, buf); err != nil { + if err == io.EOF || err == io.ErrUnexpectedEOF { + break + } + return Manifest{}, 0, err + } + if crc32.Checksum(buf, y.CastagnoliCrcTable) != y.BytesToU32(lenCrcBuf[4:8]) { + return Manifest{}, 0, errBadChecksum + } + + var changeSet pb.ManifestChangeSet + if err := proto.Unmarshal(buf, &changeSet); err != nil { + return Manifest{}, 0, err + } + + if err := applyChangeSet(&build, &changeSet); err != nil { + return Manifest{}, 0, err + } + } + + return build, offset, nil +} + +func applyManifestChange(build *Manifest, tc *pb.ManifestChange) error { + switch tc.Op { + case pb.ManifestChange_CREATE: + if _, ok := build.Tables[tc.Id]; ok { + return fmt.Errorf("MANIFEST invalid, table %d exists", tc.Id) + } + build.Tables[tc.Id] = TableManifest{ + Level: uint8(tc.Level), + KeyID: tc.KeyId, + Compression: options.CompressionType(tc.Compression), + } + for len(build.Levels) <= int(tc.Level) { + build.Levels = append(build.Levels, levelManifest{make(map[uint64]struct{})}) + } + build.Levels[tc.Level].Tables[tc.Id] = struct{}{} + build.Creations++ + case pb.ManifestChange_DELETE: + tm, ok := build.Tables[tc.Id] + if !ok { + return fmt.Errorf("MANIFEST removes non-existing table %d", tc.Id) + } + delete(build.Levels[tm.Level].Tables, tc.Id) + delete(build.Tables, tc.Id) + build.Deletions++ + default: + return fmt.Errorf("MANIFEST file has invalid manifestChange op") + } + return nil +} + +// This is not a "recoverable" error -- opening the KV store fails because the MANIFEST file is +// just plain broken. +func applyChangeSet(build *Manifest, changeSet *pb.ManifestChangeSet) error { + for _, change := range changeSet.Changes { + if err := applyManifestChange(build, change); err != nil { + return err + } + } + return nil +} + +func newCreateChange( + id uint64, level int, keyID uint64, c options.CompressionType) *pb.ManifestChange { + return &pb.ManifestChange{ + Id: id, + Op: pb.ManifestChange_CREATE, + Level: uint32(level), + KeyId: keyID, + // Hard coding it, since we're supporting only AES for now. + EncryptionAlgo: pb.EncryptionAlgo_aes, + Compression: uint32(c), + } +} + +func newDeleteChange(id uint64) *pb.ManifestChange { + return &pb.ManifestChange{ + Id: id, + Op: pb.ManifestChange_DELETE, + } +} diff --git a/vendor/github.com/dgraph-io/badger/v4/memtable.go b/vendor/github.com/dgraph-io/badger/v4/memtable.go new file mode 100644 index 00000000000..b78895e4812 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/memtable.go @@ -0,0 +1,628 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bufio" + "bytes" + "crypto/aes" + cryptorand "crypto/rand" + "encoding/binary" + "fmt" + "hash/crc32" + "io" + "os" + "path/filepath" + "sort" + "strconv" + "strings" + "sync" + "sync/atomic" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/skl" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// memTable structure stores a skiplist and a corresponding WAL. Writes to memTable are written +// both to the WAL and the skiplist. On a crash, the WAL is replayed to bring the skiplist back to +// its pre-crash form. +type memTable struct { + // TODO: Give skiplist z.Calloc'd []byte. + sl *skl.Skiplist + wal *logFile + maxVersion uint64 + opt Options + buf *bytes.Buffer +} + +func (db *DB) openMemTables(opt Options) error { + // We don't need to open any tables in in-memory mode. + if db.opt.InMemory { + return nil + } + files, err := os.ReadDir(db.opt.Dir) + if err != nil { + return errFile(err, db.opt.Dir, "Unable to open mem dir.") + } + + var fids []int + for _, file := range files { + if !strings.HasSuffix(file.Name(), memFileExt) { + continue + } + fsz := len(file.Name()) + fid, err := strconv.ParseInt(file.Name()[:fsz-len(memFileExt)], 10, 64) + if err != nil { + return errFile(err, file.Name(), "Unable to parse log id.") + } + fids = append(fids, int(fid)) + } + + // Sort in ascending order. + sort.Slice(fids, func(i, j int) bool { + return fids[i] < fids[j] + }) + for _, fid := range fids { + flags := os.O_RDWR + if db.opt.ReadOnly { + flags = os.O_RDONLY + } + mt, err := db.openMemTable(fid, flags) + if err != nil { + return y.Wrapf(err, "while opening fid: %d", fid) + } + // If this memtable is empty we don't need to add it. This is a + // memtable that was completely truncated. + if mt.sl.Empty() { + mt.DecrRef() + continue + } + // These should no longer be written to. So, make them part of the imm. + db.imm = append(db.imm, mt) + } + if len(fids) != 0 { + db.nextMemFid = fids[len(fids)-1] + } + db.nextMemFid++ + return nil +} + +const memFileExt string = ".mem" + +func (db *DB) openMemTable(fid, flags int) (*memTable, error) { + filepath := db.mtFilePath(fid) + s := skl.NewSkiplist(arenaSize(db.opt)) + mt := &memTable{ + sl: s, + opt: db.opt, + buf: &bytes.Buffer{}, + } + // We don't need to create the wal for the skiplist in in-memory mode so return the mt. + if db.opt.InMemory { + return mt, z.NewFile + } + + mt.wal = &logFile{ + fid: uint32(fid), + path: filepath, + registry: db.registry, + writeAt: vlogHeaderSize, + opt: db.opt, + } + lerr := mt.wal.open(filepath, flags, 2*db.opt.MemTableSize) + if lerr != z.NewFile && lerr != nil { + return nil, y.Wrapf(lerr, "While opening memtable: %s", filepath) + } + + // Have a callback set to delete WAL when skiplist reference count goes down to zero. That is, + // when it gets flushed to L0. + s.OnClose = func() { + if err := mt.wal.Delete(); err != nil { + db.opt.Errorf("while deleting file: %s, err: %v", filepath, err) + } + } + + if lerr == z.NewFile { + return mt, lerr + } + err := mt.UpdateSkipList() + return mt, y.Wrapf(err, "while updating skiplist") +} + +func (db *DB) newMemTable() (*memTable, error) { + mt, err := db.openMemTable(db.nextMemFid, os.O_CREATE|os.O_RDWR) + if err == z.NewFile { + db.nextMemFid++ + return mt, nil + } + + if err != nil { + db.opt.Errorf("Got error: %v for id: %d\n", err, db.nextMemFid) + return nil, y.Wrapf(err, "newMemTable") + } + return nil, errors.Errorf("File %s already exists", mt.wal.Fd.Name()) +} + +func (db *DB) mtFilePath(fid int) string { + return filepath.Join(db.opt.Dir, fmt.Sprintf("%05d%s", fid, memFileExt)) +} + +func (mt *memTable) SyncWAL() error { + return mt.wal.Sync() +} + +func (mt *memTable) isFull() bool { + if mt.sl.MemSize() >= mt.opt.MemTableSize { + return true + } + if mt.opt.InMemory { + // InMemory mode doesn't have any WAL. + return false + } + return int64(mt.wal.writeAt) >= mt.opt.MemTableSize +} + +func (mt *memTable) Put(key []byte, value y.ValueStruct) error { + entry := &Entry{ + Key: key, + Value: value.Value, + UserMeta: value.UserMeta, + meta: value.Meta, + ExpiresAt: value.ExpiresAt, + } + + // wal is nil only when badger in running in in-memory mode and we don't need the wal. + if mt.wal != nil { + // If WAL exceeds opt.ValueLogFileSize, we'll force flush the memTable. See logic in + // ensureRoomForWrite. + if err := mt.wal.writeEntry(mt.buf, entry, mt.opt); err != nil { + return y.Wrapf(err, "cannot write entry to WAL file") + } + } + // We insert the finish marker in the WAL but not in the memtable. + if entry.meta&bitFinTxn > 0 { + return nil + } + + // Write to skiplist and update maxVersion encountered. + mt.sl.Put(key, value) + if ts := y.ParseTs(entry.Key); ts > mt.maxVersion { + mt.maxVersion = ts + } + y.NumBytesWrittenToL0Add(mt.opt.MetricsEnabled, entry.estimateSizeAndSetThreshold(mt.opt.ValueThreshold)) + return nil +} + +func (mt *memTable) UpdateSkipList() error { + if mt.wal == nil || mt.sl == nil { + return nil + } + endOff, err := mt.wal.iterate(true, 0, mt.replayFunction(mt.opt)) + if err != nil { + return y.Wrapf(err, "while iterating wal: %s", mt.wal.Fd.Name()) + } + if endOff < mt.wal.size.Load() && mt.opt.ReadOnly { + return y.Wrapf(ErrTruncateNeeded, "end offset: %d < size: %d", endOff, mt.wal.size.Load()) + } + return mt.wal.Truncate(int64(endOff)) +} + +// IncrRef increases the refcount +func (mt *memTable) IncrRef() { + mt.sl.IncrRef() +} + +// DecrRef decrements the refcount, deallocating the Skiplist when done using it +func (mt *memTable) DecrRef() { + mt.sl.DecrRef() +} + +func (mt *memTable) replayFunction(opt Options) func(Entry, valuePointer) error { + first := true + return func(e Entry, _ valuePointer) error { // Function for replaying. + if first { + opt.Debugf("First key=%q\n", e.Key) + } + first = false + if ts := y.ParseTs(e.Key); ts > mt.maxVersion { + mt.maxVersion = ts + } + v := y.ValueStruct{ + Value: e.Value, + Meta: e.meta, + UserMeta: e.UserMeta, + ExpiresAt: e.ExpiresAt, + } + // This is already encoded correctly. Value would be either a vptr, or a full value + // depending upon how big the original value was. Skiplist makes a copy of the key and + // value. + mt.sl.Put(e.Key, v) + return nil + } +} + +type logFile struct { + *z.MmapFile + path string + // This is a lock on the log file. It guards the fd’s value, the file’s + // existence and the file’s memory map. + // + // Use shared ownership when reading/writing the file or memory map, use + // exclusive ownership to open/close the descriptor, unmap or remove the file. + lock sync.RWMutex + fid uint32 + size atomic.Uint32 + dataKey *pb.DataKey + baseIV []byte + registry *KeyRegistry + writeAt uint32 + opt Options +} + +func (lf *logFile) Truncate(end int64) error { + if fi, err := lf.Fd.Stat(); err != nil { + return fmt.Errorf("while file.stat on file: %s, error: %v\n", lf.Fd.Name(), err) + } else if fi.Size() == end { + return nil + } + y.AssertTrue(!lf.opt.ReadOnly) + lf.size.Store(uint32(end)) + return lf.MmapFile.Truncate(end) +} + +// encodeEntry will encode entry to the buf +// layout of entry +// +--------+-----+-------+-------+ +// | header | key | value | crc32 | +// +--------+-----+-------+-------+ +func (lf *logFile) encodeEntry(buf *bytes.Buffer, e *Entry, offset uint32) (int, error) { + h := header{ + klen: uint32(len(e.Key)), + vlen: uint32(len(e.Value)), + expiresAt: e.ExpiresAt, + meta: e.meta, + userMeta: e.UserMeta, + } + + hash := crc32.New(y.CastagnoliCrcTable) + writer := io.MultiWriter(buf, hash) + + // encode header. + var headerEnc [maxHeaderSize]byte + sz := h.Encode(headerEnc[:]) + y.Check2(writer.Write(headerEnc[:sz])) + // we'll encrypt only key and value. + if lf.encryptionEnabled() { + // TODO: no need to allocate the bytes. we can calculate the encrypted buf one by one + // since we're using ctr mode of AES encryption. Ordering won't changed. Need some + // refactoring in XORBlock which will work like stream cipher. + eBuf := make([]byte, 0, len(e.Key)+len(e.Value)) + eBuf = append(eBuf, e.Key...) + eBuf = append(eBuf, e.Value...) + if err := y.XORBlockStream( + writer, eBuf, lf.dataKey.Data, lf.generateIV(offset)); err != nil { + return 0, y.Wrapf(err, "Error while encoding entry for vlog.") + } + } else { + // Encryption is disabled so writing directly to the buffer. + y.Check2(writer.Write(e.Key)) + y.Check2(writer.Write(e.Value)) + } + // write crc32 hash. + var crcBuf [crc32.Size]byte + binary.BigEndian.PutUint32(crcBuf[:], hash.Sum32()) + y.Check2(buf.Write(crcBuf[:])) + // return encoded length. + return len(headerEnc[:sz]) + len(e.Key) + len(e.Value) + len(crcBuf), nil +} + +func (lf *logFile) writeEntry(buf *bytes.Buffer, e *Entry, opt Options) error { + buf.Reset() + plen, err := lf.encodeEntry(buf, e, lf.writeAt) + if err != nil { + return err + } + y.AssertTrue(plen == copy(lf.Data[lf.writeAt:], buf.Bytes())) + lf.writeAt += uint32(plen) + + lf.zeroNextEntry() + return nil +} + +func (lf *logFile) decodeEntry(buf []byte, offset uint32) (*Entry, error) { + var h header + hlen := h.Decode(buf) + kv := buf[hlen:] + if lf.encryptionEnabled() { + var err error + // No need to worry about mmap. because, XORBlock allocates a byte array to do the + // xor. So, the given slice is not being mutated. + if kv, err = lf.decryptKV(kv, offset); err != nil { + return nil, err + } + } + e := &Entry{ + meta: h.meta, + UserMeta: h.userMeta, + ExpiresAt: h.expiresAt, + offset: offset, + Key: kv[:h.klen], + Value: kv[h.klen : h.klen+h.vlen], + } + return e, nil +} + +func (lf *logFile) decryptKV(buf []byte, offset uint32) ([]byte, error) { + return y.XORBlockAllocate(buf, lf.dataKey.Data, lf.generateIV(offset)) +} + +// KeyID returns datakey's ID. +func (lf *logFile) keyID() uint64 { + if lf.dataKey == nil { + // If there is no datakey, then we'll return 0. Which means no encryption. + return 0 + } + return lf.dataKey.KeyId +} + +func (lf *logFile) encryptionEnabled() bool { + return lf.dataKey != nil +} + +// Acquire lock on mmap/file if you are calling this +func (lf *logFile) read(p valuePointer) (buf []byte, err error) { + offset := p.Offset + // Do not convert size to uint32, because the lf.Data can be of size + // 4GB, which overflows the uint32 during conversion to make the size 0, + // causing the read to fail with ErrEOF. See issue #585. + size := int64(len(lf.Data)) + valsz := p.Len + lfsz := lf.size.Load() + if int64(offset) >= size || int64(offset+valsz) > size || + // Ensure that the read is within the file's actual size. It might be possible that + // the offset+valsz length is beyond the file's actual size. This could happen when + // dropAll and iterations are running simultaneously. + int64(offset+valsz) > int64(lfsz) { + err = y.ErrEOF + } else { + buf = lf.Data[offset : offset+valsz] + } + return buf, err +} + +// generateIV will generate IV by appending given offset with the base IV. +func (lf *logFile) generateIV(offset uint32) []byte { + iv := make([]byte, aes.BlockSize) + // baseIV is of 12 bytes. + y.AssertTrue(12 == copy(iv[:12], lf.baseIV)) + // remaining 4 bytes is obtained from offset. + binary.BigEndian.PutUint32(iv[12:], offset) + return iv +} + +func (lf *logFile) doneWriting(offset uint32) error { + if lf.opt.SyncWrites { + if err := lf.Sync(); err != nil { + return y.Wrapf(err, "Unable to sync value log: %q", lf.path) + } + } + + // Before we were acquiring a lock here on lf.lock, because we were invalidating the file + // descriptor due to reopening it as read-only. Now, we don't invalidate the fd, but unmap it, + // truncate it and remap it. That creates a window where we have segfaults because the mmap is + // no longer valid, while someone might be reading it. Therefore, we need a lock here again. + lf.lock.Lock() + defer lf.lock.Unlock() + + if err := lf.Truncate(int64(offset)); err != nil { + return y.Wrapf(err, "Unable to truncate file: %q", lf.path) + } + + // Previously we used to close the file after it was written and reopen it in read-only mode. + // We no longer open files in read-only mode. We keep all vlog files open in read-write mode. + return nil +} + +// iterate iterates over log file. It doesn't not allocate new memory for every kv pair. +// Therefore, the kv pair is only valid for the duration of fn call. +func (lf *logFile) iterate(readOnly bool, offset uint32, fn logEntry) (uint32, error) { + if offset == 0 { + // If offset is set to zero, let's advance past the encryption key header. + offset = vlogHeaderSize + } + + // For now, read directly from file, because it allows + reader := bufio.NewReader(lf.NewReader(int(offset))) + read := &safeRead{ + k: make([]byte, 10), + v: make([]byte, 10), + recordOffset: offset, + lf: lf, + } + + var lastCommit uint64 + var validEndOffset uint32 = offset + + var entries []*Entry + var vptrs []valuePointer + +loop: + for { + e, err := read.Entry(reader) + switch { + // We have not reached the end of the file but the entry we read is + // zero. This happens because we have truncated the file and + // zero'ed it out. + case err == io.EOF: + break loop + case err == io.ErrUnexpectedEOF || err == errTruncate: + break loop + case err != nil: + return 0, err + case e == nil: + continue + case e.isZero(): + break loop + } + + var vp valuePointer + vp.Len = uint32(e.hlen + len(e.Key) + len(e.Value) + crc32.Size) + read.recordOffset += vp.Len + + vp.Offset = e.offset + vp.Fid = lf.fid + + switch { + case e.meta&bitTxn > 0: + txnTs := y.ParseTs(e.Key) + if lastCommit == 0 { + lastCommit = txnTs + } + if lastCommit != txnTs { + break loop + } + entries = append(entries, e) + vptrs = append(vptrs, vp) + + case e.meta&bitFinTxn > 0: + txnTs, err := strconv.ParseUint(string(e.Value), 10, 64) + if err != nil || lastCommit != txnTs { + break loop + } + // Got the end of txn. Now we can store them. + lastCommit = 0 + validEndOffset = read.recordOffset + + for i, e := range entries { + vp := vptrs[i] + if err := fn(*e, vp); err != nil { + if err == errStop { + break + } + return 0, errFile(err, lf.path, "Iteration function") + } + } + entries = entries[:0] + vptrs = vptrs[:0] + + default: + if lastCommit != 0 { + // This is most likely an entry which was moved as part of GC. + // We shouldn't get this entry in the middle of a transaction. + break loop + } + validEndOffset = read.recordOffset + + if err := fn(*e, vp); err != nil { + if err == errStop { + break + } + return 0, errFile(err, lf.path, "Iteration function") + } + } + } + return validEndOffset, nil +} + +// Zero out the next entry to deal with any crashes. +func (lf *logFile) zeroNextEntry() { + z.ZeroOut(lf.Data, int(lf.writeAt), int(lf.writeAt+maxHeaderSize)) +} + +func (lf *logFile) open(path string, flags int, fsize int64) error { + mf, ferr := z.OpenMmapFile(path, flags, int(fsize)) + lf.MmapFile = mf + + if ferr == z.NewFile { + if err := lf.bootstrap(); err != nil { + os.Remove(path) + return err + } + lf.size.Store(vlogHeaderSize) + + } else if ferr != nil { + return y.Wrapf(ferr, "while opening file: %s", path) + } + lf.size.Store(uint32(len(lf.Data))) + + if lf.size.Load() < vlogHeaderSize { + // Every vlog file should have at least vlogHeaderSize. If it is less than vlogHeaderSize + // then it must have been corrupted. But no need to handle here. log replayer will truncate + // and bootstrap the logfile. So ignoring here. + return nil + } + + // Copy over the encryption registry data. + buf := make([]byte, vlogHeaderSize) + + y.AssertTruef(vlogHeaderSize == copy(buf, lf.Data), + "Unable to copy from %s, size %d", path, lf.size.Load()) + keyID := binary.BigEndian.Uint64(buf[:8]) + // retrieve datakey. + if dk, err := lf.registry.DataKey(keyID); err != nil { + return y.Wrapf(err, "While opening vlog file %d", lf.fid) + } else { + lf.dataKey = dk + } + lf.baseIV = buf[8:] + y.AssertTrue(len(lf.baseIV) == 12) + + // Preserved ferr so we can return if this was a new file. + return ferr +} + +// bootstrap will initialize the log file with key id and baseIV. +// The below figure shows the layout of log file. +// +----------------+------------------+------------------+ +// | keyID(8 bytes) | baseIV(12 bytes)| entry... | +// +----------------+------------------+------------------+ +func (lf *logFile) bootstrap() error { + var err error + + // generate data key for the log file. + var dk *pb.DataKey + if dk, err = lf.registry.LatestDataKey(); err != nil { + return y.Wrapf(err, "Error while retrieving datakey in logFile.bootstarp") + } + lf.dataKey = dk + + // We'll always preserve vlogHeaderSize for key id and baseIV. + buf := make([]byte, vlogHeaderSize) + + // write key id to the buf. + // key id will be zero if the logfile is in plain text. + binary.BigEndian.PutUint64(buf[:8], lf.keyID()) + // generate base IV. It'll be used with offset of the vptr to encrypt the entry. + if _, err := cryptorand.Read(buf[8:]); err != nil { + return y.Wrapf(err, "Error while creating base IV, while creating logfile") + } + + // Initialize base IV. + lf.baseIV = buf[8:] + y.AssertTrue(len(lf.baseIV) == 12) + + // Copy over to the logFile. + y.AssertTrue(vlogHeaderSize == copy(lf.Data[0:], buf)) + + // Zero out the next entry. + lf.zeroNextEntry() + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/merge.go b/vendor/github.com/dgraph-io/badger/v4/merge.go new file mode 100644 index 00000000000..3ec28e90ed5 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/merge.go @@ -0,0 +1,182 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "sync" + "time" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// MergeOperator represents a Badger merge operator. +type MergeOperator struct { + sync.RWMutex + f MergeFunc + db *DB + key []byte + closer *z.Closer +} + +// MergeFunc accepts two byte slices, one representing an existing value, and +// another representing a new value that needs to be ‘merged’ into it. MergeFunc +// contains the logic to perform the ‘merge’ and return an updated value. +// MergeFunc could perform operations like integer addition, list appends etc. +// Note that the ordering of the operands is maintained. +type MergeFunc func(existingVal, newVal []byte) []byte + +// GetMergeOperator creates a new MergeOperator for a given key and returns a +// pointer to it. It also fires off a goroutine that performs a compaction using +// the merge function that runs periodically, as specified by dur. +func (db *DB) GetMergeOperator(key []byte, + f MergeFunc, dur time.Duration) *MergeOperator { + op := &MergeOperator{ + f: f, + db: db, + key: key, + closer: z.NewCloser(1), + } + + go op.runCompactions(dur) + return op +} + +var errNoMerge = errors.New("No need for merge") + +func (op *MergeOperator) iterateAndMerge() (newVal []byte, latest uint64, err error) { + txn := op.db.NewTransaction(false) + defer txn.Discard() + opt := DefaultIteratorOptions + opt.AllVersions = true + it := txn.NewKeyIterator(op.key, opt) + defer it.Close() + + var numVersions int + for it.Rewind(); it.Valid(); it.Next() { + item := it.Item() + if item.IsDeletedOrExpired() { + break + } + numVersions++ + if numVersions == 1 { + // This should be the newVal, considering this is the latest version. + newVal, err = item.ValueCopy(newVal) + if err != nil { + return nil, 0, err + } + latest = item.Version() + } else { + if err := item.Value(func(oldVal []byte) error { + // The merge should always be on the newVal considering it has the merge result of + // the latest version. The value read should be the oldVal. + newVal = op.f(oldVal, newVal) + return nil + }); err != nil { + return nil, 0, err + } + } + if item.DiscardEarlierVersions() { + break + } + } + if numVersions == 0 { + return nil, latest, ErrKeyNotFound + } else if numVersions == 1 { + return newVal, latest, errNoMerge + } + return newVal, latest, nil +} + +func (op *MergeOperator) compact() error { + op.Lock() + defer op.Unlock() + val, version, err := op.iterateAndMerge() + if err == ErrKeyNotFound || err == errNoMerge { + return nil + } else if err != nil { + return err + } + entries := []*Entry{ + { + Key: y.KeyWithTs(op.key, version), + Value: val, + meta: bitDiscardEarlierVersions, + }, + } + // Write value back to the DB. It is important that we do not set the bitMergeEntry bit + // here. When compaction happens, all the older merged entries will be removed. + return op.db.batchSetAsync(entries, func(err error) { + if err != nil { + op.db.opt.Errorf("failed to insert the result of merge compaction: %s", err) + } + }) +} + +func (op *MergeOperator) runCompactions(dur time.Duration) { + ticker := time.NewTicker(dur) + defer op.closer.Done() + var stop bool + for { + select { + case <-op.closer.HasBeenClosed(): + stop = true + case <-ticker.C: // wait for tick + } + if err := op.compact(); err != nil { + op.db.opt.Errorf("failure while running merge operation: %s", err) + } + if stop { + ticker.Stop() + break + } + } +} + +// Add records a value in Badger which will eventually be merged by a background +// routine into the values that were recorded by previous invocations to Add(). +func (op *MergeOperator) Add(val []byte) error { + return op.db.Update(func(txn *Txn) error { + return txn.SetEntry(NewEntry(op.key, val).withMergeBit()) + }) +} + +// Get returns the latest value for the merge operator, which is derived by +// applying the merge function to all the values added so far. +// +// If Add has not been called even once, Get will return ErrKeyNotFound. +func (op *MergeOperator) Get() ([]byte, error) { + op.RLock() + defer op.RUnlock() + var existing []byte + err := op.db.View(func(txn *Txn) (err error) { + existing, _, err = op.iterateAndMerge() + return err + }) + if err == errNoMerge { + return existing, nil + } + return existing, err +} + +// Stop waits for any pending merge to complete and then stops the background +// goroutine. +func (op *MergeOperator) Stop() { + op.closer.SignalAndWait() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/options.go b/vendor/github.com/dgraph-io/badger/v4/options.go new file mode 100644 index 00000000000..ac046bc1db6 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/options.go @@ -0,0 +1,806 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "fmt" + "os" + "reflect" + "strconv" + "strings" + "time" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/options" + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// Note: If you add a new option X make sure you also add a WithX method on Options. + +// Options are params for creating DB object. +// +// This package provides DefaultOptions which contains options that should +// work for most applications. Consider using that as a starting point before +// customizing it for your own needs. +// +// Each option X is documented on the WithX method. +type Options struct { + testOnlyOptions + + // Required options. + + Dir string + ValueDir string + + // Usually modified options. + + SyncWrites bool + NumVersionsToKeep int + ReadOnly bool + Logger Logger + Compression options.CompressionType + InMemory bool + MetricsEnabled bool + // Sets the Stream.numGo field + NumGoroutines int + + // Fine tuning options. + + MemTableSize int64 + BaseTableSize int64 + BaseLevelSize int64 + LevelSizeMultiplier int + TableSizeMultiplier int + MaxLevels int + + VLogPercentile float64 + ValueThreshold int64 + NumMemtables int + // Changing BlockSize across DB runs will not break badger. The block size is + // read from the block index stored at the end of the table. + BlockSize int + BloomFalsePositive float64 + BlockCacheSize int64 + IndexCacheSize int64 + + NumLevelZeroTables int + NumLevelZeroTablesStall int + + ValueLogFileSize int64 + ValueLogMaxEntries uint32 + + NumCompactors int + CompactL0OnClose bool + LmaxCompaction bool + ZSTDCompressionLevel int + + // When set, checksum will be validated for each entry read from the value log file. + VerifyValueChecksum bool + + // Encryption related options. + EncryptionKey []byte // encryption key + EncryptionKeyRotationDuration time.Duration // key rotation duration + + // BypassLockGuard will bypass the lock guard on badger. Bypassing lock + // guard can cause data corruption if multiple badger instances are using + // the same directory. Use this options with caution. + BypassLockGuard bool + + // ChecksumVerificationMode decides when db should verify checksums for SSTable blocks. + ChecksumVerificationMode options.ChecksumVerificationMode + + // DetectConflicts determines whether the transactions would be checked for + // conflicts. The transactions can be processed at a higher rate when + // conflict detection is disabled. + DetectConflicts bool + + // NamespaceOffset specifies the offset from where the next 8 bytes contains the namespace. + NamespaceOffset int + + // Magic version used by the application using badger to ensure that it doesn't open the DB + // with incompatible data format. + ExternalMagicVersion uint16 + + // Transaction start and commit timestamps are managed by end-user. + // This is only useful for databases built on top of Badger (like Dgraph). + // Not recommended for most users. + managedTxns bool + + // 4. Flags for testing purposes + // ------------------------------ + maxBatchCount int64 // max entries in batch + maxBatchSize int64 // max batch size in bytes + + maxValueThreshold float64 +} + +// DefaultOptions sets a list of recommended options for good performance. +// Feel free to modify these to suit your needs with the WithX methods. +func DefaultOptions(path string) Options { + return Options{ + Dir: path, + ValueDir: path, + + MemTableSize: 64 << 20, + BaseTableSize: 2 << 20, + BaseLevelSize: 10 << 20, + TableSizeMultiplier: 2, + LevelSizeMultiplier: 10, + MaxLevels: 7, + NumGoroutines: 8, + MetricsEnabled: true, + + NumCompactors: 4, // Run at least 2 compactors. Zero-th compactor prioritizes L0. + NumLevelZeroTables: 5, + NumLevelZeroTablesStall: 15, + NumMemtables: 5, + BloomFalsePositive: 0.01, + BlockSize: 4 * 1024, + SyncWrites: false, + NumVersionsToKeep: 1, + CompactL0OnClose: false, + VerifyValueChecksum: false, + Compression: options.Snappy, + BlockCacheSize: 256 << 20, + IndexCacheSize: 0, + + // The following benchmarks were done on a 4 KB block size (default block size). The + // compression is ratio supposed to increase with increasing compression level but since the + // input for compression algorithm is small (4 KB), we don't get significant benefit at + // level 3. + // NOTE: The benchmarks are with DataDog ZSTD that requires CGO. Hence, no longer valid. + // no_compression-16 10 502848865 ns/op 165.46 MB/s - + // zstd_compression/level_1-16 7 739037966 ns/op 112.58 MB/s 2.93 + // zstd_compression/level_3-16 7 756950250 ns/op 109.91 MB/s 2.72 + // zstd_compression/level_15-16 1 11135686219 ns/op 7.47 MB/s 4.38 + // Benchmark code can be found in table/builder_test.go file + ZSTDCompressionLevel: 1, + + // Nothing to read/write value log using standard File I/O + // MemoryMap to mmap() the value log files + // (2^30 - 1)*2 when mmapping < 2^31 - 1, max int32. + // -1 so 2*ValueLogFileSize won't overflow on 32-bit systems. + ValueLogFileSize: 1<<30 - 1, + + ValueLogMaxEntries: 1000000, + + VLogPercentile: 0.0, + ValueThreshold: maxValueThreshold, + + Logger: defaultLogger(INFO), + EncryptionKey: []byte{}, + EncryptionKeyRotationDuration: 10 * 24 * time.Hour, // Default 10 days. + DetectConflicts: true, + NamespaceOffset: -1, + } +} + +func buildTableOptions(db *DB) table.Options { + opt := db.opt + dk, err := db.registry.LatestDataKey() + y.Check(err) + return table.Options{ + ReadOnly: opt.ReadOnly, + MetricsEnabled: db.opt.MetricsEnabled, + TableSize: uint64(opt.BaseTableSize), + BlockSize: opt.BlockSize, + BloomFalsePositive: opt.BloomFalsePositive, + ChkMode: opt.ChecksumVerificationMode, + Compression: opt.Compression, + ZSTDCompressionLevel: opt.ZSTDCompressionLevel, + BlockCache: db.blockCache, + IndexCache: db.indexCache, + AllocPool: db.allocPool, + DataKey: dk, + } +} + +const ( + maxValueThreshold = (1 << 20) // 1 MB +) + +// LSMOnlyOptions follows from DefaultOptions, but sets a higher ValueThreshold +// so values would be collocated with the LSM tree, with value log largely acting +// as a write-ahead log only. These options would reduce the disk usage of value +// log, and make Badger act more like a typical LSM tree. +func LSMOnlyOptions(path string) Options { + // Let's not set any other options, because they can cause issues with the + // size of key-value a user can pass to Badger. For e.g., if we set + // ValueLogFileSize to 64MB, a user can't pass a value more than that. + // Setting it to ValueLogMaxEntries to 1000, can generate too many files. + // These options are better configured on a usage basis, than broadly here. + // The ValueThreshold is the most important setting a user needs to do to + // achieve a heavier usage of LSM tree. + // NOTE: If a user does not want to set 64KB as the ValueThreshold because + // of performance reasons, 1KB would be a good option too, allowing + // values smaller than 1KB to be collocated with the keys in the LSM tree. + return DefaultOptions(path).WithValueThreshold(maxValueThreshold /* 1 MB */) +} + +// parseCompression returns badger.compressionType and compression level given compression string +// of format compression-type:compression-level +func parseCompression(cStr string) (options.CompressionType, int, error) { + cStrSplit := strings.Split(cStr, ":") + cType := cStrSplit[0] + level := 3 + + var err error + if len(cStrSplit) == 2 { + level, err = strconv.Atoi(cStrSplit[1]) + y.Check(err) + if level <= 0 { + return 0, 0, + errors.Errorf("ERROR: compression level(%v) must be greater than zero", level) + } + } else if len(cStrSplit) > 2 { + return 0, 0, errors.Errorf("ERROR: Invalid badger.compression argument") + } + switch cType { + case "zstd": + return options.ZSTD, level, nil + case "snappy": + return options.Snappy, 0, nil + case "none": + return options.None, 0, nil + } + return 0, 0, errors.Errorf("ERROR: compression type (%s) invalid", cType) +} + +// generateSuperFlag generates an identical SuperFlag string from the provided Options. +func generateSuperFlag(options Options) string { + superflag := "" + v := reflect.ValueOf(&options).Elem() + optionsStruct := v.Type() + for i := 0; i < v.NumField(); i++ { + if field := v.Field(i); field.CanInterface() { + name := strings.ToLower(optionsStruct.Field(i).Name) + kind := v.Field(i).Kind() + switch kind { + case reflect.Bool: + superflag += name + "=" + superflag += fmt.Sprintf("%v; ", field.Bool()) + case reflect.Int, reflect.Int64: + superflag += name + "=" + superflag += fmt.Sprintf("%v; ", field.Int()) + case reflect.Uint32, reflect.Uint64: + superflag += name + "=" + superflag += fmt.Sprintf("%v; ", field.Uint()) + case reflect.Float64: + superflag += name + "=" + superflag += fmt.Sprintf("%v; ", field.Float()) + case reflect.String: + superflag += name + "=" + superflag += fmt.Sprintf("%v; ", field.String()) + default: + continue + } + } + } + return superflag +} + +// FromSuperFlag fills Options fields for each flag within the superflag. For +// example, replacing the default Options.NumGoroutines: +// +// options := FromSuperFlag("numgoroutines=4", DefaultOptions("")) +// +// It's important to note that if you pass an empty Options struct, FromSuperFlag +// will not fill it with default values. FromSuperFlag only writes to the fields +// present within the superflag string (case insensitive). +// +// It specially handles compression subflag. +// Valid options are {none,snappy,zstd:} +// Example: compression=zstd:3; +// Unsupported: Options.Logger, Options.EncryptionKey +func (opt Options) FromSuperFlag(superflag string) Options { + // currentOptions act as a default value for the options superflag. + currentOptions := generateSuperFlag(opt) + currentOptions += "compression=;" + + flags := z.NewSuperFlag(superflag).MergeAndCheckDefault(currentOptions) + v := reflect.ValueOf(&opt).Elem() + optionsStruct := v.Type() + for i := 0; i < v.NumField(); i++ { + // only iterate over exported fields + if field := v.Field(i); field.CanInterface() { + // z.SuperFlag stores keys as lowercase, keep everything case + // insensitive + name := strings.ToLower(optionsStruct.Field(i).Name) + if name == "compression" { + // We will specially handle this later. Skip it here. + continue + } + kind := v.Field(i).Kind() + switch kind { + case reflect.Bool: + field.SetBool(flags.GetBool(name)) + case reflect.Int, reflect.Int64: + field.SetInt(flags.GetInt64(name)) + case reflect.Uint32, reflect.Uint64: + field.SetUint(flags.GetUint64(name)) + case reflect.Float64: + field.SetFloat(flags.GetFloat64(name)) + case reflect.String: + field.SetString(flags.GetString(name)) + } + } + } + + // Only update the options for special flags that were present in the input superflag. + inputFlag := z.NewSuperFlag(superflag) + if inputFlag.Has("compression") { + ctype, clevel, err := parseCompression(flags.GetString("compression")) + switch err { + case nil: + opt.Compression = ctype + opt.ZSTDCompressionLevel = clevel + default: + ctype = options.CompressionType(flags.GetUint32("compression")) + y.AssertTruef(ctype <= 2, "ERROR: Invalid format or compression type. Got: %s", + flags.GetString("compression")) + opt.Compression = ctype + } + } + + return opt +} + +// WithDir returns a new Options value with Dir set to the given value. +// +// Dir is the path of the directory where key data will be stored in. +// If it doesn't exist, Badger will try to create it for you. +// This is set automatically to be the path given to `DefaultOptions`. +func (opt Options) WithDir(val string) Options { + opt.Dir = val + return opt +} + +// WithValueDir returns a new Options value with ValueDir set to the given value. +// +// ValueDir is the path of the directory where value data will be stored in. +// If it doesn't exist, Badger will try to create it for you. +// This is set automatically to be the path given to `DefaultOptions`. +func (opt Options) WithValueDir(val string) Options { + opt.ValueDir = val + return opt +} + +// WithSyncWrites returns a new Options value with SyncWrites set to the given value. +// +// Badger does all writes via mmap. So, all writes can survive process crashes or k8s environments +// with SyncWrites set to false. +// +// When set to true, Badger would call an additional msync after writes to flush mmap buffer over to +// disk to survive hard reboots. Most users of Badger should not need to do this. +// +// The default value of SyncWrites is false. +func (opt Options) WithSyncWrites(val bool) Options { + opt.SyncWrites = val + return opt +} + +// WithNumVersionsToKeep returns a new Options value with NumVersionsToKeep set to the given value. +// +// NumVersionsToKeep sets how many versions to keep per key at most. +// +// The default value of NumVersionsToKeep is 1. +func (opt Options) WithNumVersionsToKeep(val int) Options { + opt.NumVersionsToKeep = val + return opt +} + +// WithNumGoroutines sets the number of goroutines to be used in Stream. +// +// The default value of NumGoroutines is 8. +func (opt Options) WithNumGoroutines(val int) Options { + opt.NumGoroutines = val + return opt +} + +// WithReadOnly returns a new Options value with ReadOnly set to the given value. +// +// When ReadOnly is true the DB will be opened on read-only mode. +// Multiple processes can open the same Badger DB. +// Note: if the DB being opened had crashed before and has vlog data to be replayed, +// ReadOnly will cause Open to fail with an appropriate message. +// +// The default value of ReadOnly is false. +func (opt Options) WithReadOnly(val bool) Options { + opt.ReadOnly = val + return opt +} + +// WithMetricsEnabled returns a new Options value with MetricsEnabled set to the given value. +// +// When MetricsEnabled is set to false, then the DB will be opened and no badger metrics +// will be logged. Metrics are defined in metric.go file. +// +// This flag is useful for use cases like in Dgraph where we open temporary badger instances to +// index data. In those cases we don't want badger metrics to be polluted with the noise from +// those temporary instances. +// +// Default value is set to true +func (opt Options) WithMetricsEnabled(val bool) Options { + opt.MetricsEnabled = val + return opt +} + +// WithLogger returns a new Options value with Logger set to the given value. +// +// Logger provides a way to configure what logger each value of badger.DB uses. +// +// The default value of Logger writes to stderr using the log package from the Go standard library. +func (opt Options) WithLogger(val Logger) Options { + opt.Logger = val + return opt +} + +// WithLoggingLevel returns a new Options value with logging level of the +// default logger set to the given value. +// LoggingLevel sets the level of logging. It should be one of DEBUG, INFO, +// WARNING or ERROR levels. +// +// The default value of LoggingLevel is INFO. +func (opt Options) WithLoggingLevel(val loggingLevel) Options { + opt.Logger = defaultLogger(val) + return opt +} + +// WithBaseTableSize returns a new Options value with MaxTableSize set to the given value. +// +// BaseTableSize sets the maximum size in bytes for LSM table or file in the base level. +// +// The default value of BaseTableSize is 2MB. +func (opt Options) WithBaseTableSize(val int64) Options { + opt.BaseTableSize = val + return opt +} + +// WithLevelSizeMultiplier returns a new Options value with LevelSizeMultiplier set to the given +// value. +// +// LevelSizeMultiplier sets the ratio between the maximum sizes of contiguous levels in the LSM. +// Once a level grows to be larger than this ratio allowed, the compaction process will be +// triggered. +// +// The default value of LevelSizeMultiplier is 10. +func (opt Options) WithLevelSizeMultiplier(val int) Options { + opt.LevelSizeMultiplier = val + return opt +} + +// WithMaxLevels returns a new Options value with MaxLevels set to the given value. +// +// Maximum number of levels of compaction allowed in the LSM. +// +// The default value of MaxLevels is 7. +func (opt Options) WithMaxLevels(val int) Options { + opt.MaxLevels = val + return opt +} + +// WithValueThreshold returns a new Options value with ValueThreshold set to the given value. +// +// ValueThreshold sets the threshold used to decide whether a value is stored directly in the LSM +// tree or separately in the log value files. +// +// The default value of ValueThreshold is 1 MB, and LSMOnlyOptions sets it to maxValueThreshold +// which is set to 1 MB too. +func (opt Options) WithValueThreshold(val int64) Options { + opt.ValueThreshold = val + return opt +} + +// WithVLogPercentile returns a new Options value with ValLogPercentile set to given value. +// +// VLogPercentile with 0.0 means no dynamic thresholding is enabled. +// MinThreshold value will always act as the value threshold. +// +// VLogPercentile with value 0.99 means 99 percentile of value will be put in LSM tree +// and only 1 percent in vlog. The value threshold will be dynamically updated within the range of +// [ValueThreshold, Options.maxValueThreshold] +// +// # Say VLogPercentile with 1.0 means threshold will eventually set to Options.maxValueThreshold +// +// The default value of VLogPercentile is 0.0. +func (opt Options) WithVLogPercentile(t float64) Options { + opt.VLogPercentile = t + return opt +} + +// WithNumMemtables returns a new Options value with NumMemtables set to the given value. +// +// NumMemtables sets the maximum number of tables to keep in memory before stalling. +// +// The default value of NumMemtables is 5. +func (opt Options) WithNumMemtables(val int) Options { + opt.NumMemtables = val + return opt +} + +// WithMemTableSize returns a new Options value with MemTableSize set to the given value. +// +// MemTableSize sets the maximum size in bytes for memtable table. +// +// The default value of MemTableSize is 64MB. +func (opt Options) WithMemTableSize(val int64) Options { + opt.MemTableSize = val + return opt +} + +// WithBloomFalsePositive returns a new Options value with BloomFalsePositive set +// to the given value. +// +// BloomFalsePositive sets the false positive probability of the bloom filter in any SSTable. +// Before reading a key from table, the bloom filter is checked for key existence. +// BloomFalsePositive might impact read performance of DB. Lower BloomFalsePositive value might +// consume more memory. +// +// The default value of BloomFalsePositive is 0.01. +// +// Setting this to 0 disables the bloom filter completely. +func (opt Options) WithBloomFalsePositive(val float64) Options { + opt.BloomFalsePositive = val + return opt +} + +// WithBlockSize returns a new Options value with BlockSize set to the given value. +// +// BlockSize sets the size of any block in SSTable. SSTable is divided into multiple blocks +// internally. Each block is compressed using prefix diff encoding. +// +// The default value of BlockSize is 4KB. +func (opt Options) WithBlockSize(val int) Options { + opt.BlockSize = val + return opt +} + +// WithNumLevelZeroTables sets the maximum number of Level 0 tables before compaction starts. +// +// The default value of NumLevelZeroTables is 5. +func (opt Options) WithNumLevelZeroTables(val int) Options { + opt.NumLevelZeroTables = val + return opt +} + +// WithNumLevelZeroTablesStall sets the number of Level 0 tables that once reached causes the DB to +// stall until compaction succeeds. +// +// The default value of NumLevelZeroTablesStall is 15. +func (opt Options) WithNumLevelZeroTablesStall(val int) Options { + opt.NumLevelZeroTablesStall = val + return opt +} + +// WithBaseLevelSize sets the maximum size target for the base level. +// +// The default value is 10MB. +func (opt Options) WithBaseLevelSize(val int64) Options { + opt.BaseLevelSize = val + return opt +} + +// WithValueLogFileSize sets the maximum size of a single value log file. +// +// The default value of ValueLogFileSize is 1GB. +func (opt Options) WithValueLogFileSize(val int64) Options { + opt.ValueLogFileSize = val + return opt +} + +// WithValueLogMaxEntries sets the maximum number of entries a value log file +// can hold approximately. A actual size limit of a value log file is the +// minimum of ValueLogFileSize and ValueLogMaxEntries. +// +// The default value of ValueLogMaxEntries is one million (1000000). +func (opt Options) WithValueLogMaxEntries(val uint32) Options { + opt.ValueLogMaxEntries = val + return opt +} + +// WithNumCompactors sets the number of compaction workers to run concurrently. Setting this to +// zero stops compactions, which could eventually cause writes to block forever. +// +// The default value of NumCompactors is 4. One is dedicated just for L0 and L1. +func (opt Options) WithNumCompactors(val int) Options { + opt.NumCompactors = val + return opt +} + +// WithCompactL0OnClose determines whether Level 0 should be compacted before closing the DB. This +// ensures that both reads and writes are efficient when the DB is opened later. +// +// The default value of CompactL0OnClose is false. +func (opt Options) WithCompactL0OnClose(val bool) Options { + opt.CompactL0OnClose = val + return opt +} + +// WithEncryptionKey is used to encrypt the data with AES. Type of AES is used based on the key +// size. For example 16 bytes will use AES-128. 24 bytes will use AES-192. 32 bytes will +// use AES-256. +func (opt Options) WithEncryptionKey(key []byte) Options { + opt.EncryptionKey = key + return opt +} + +// WithEncryptionKeyRotationDuration returns new Options value with the duration set to +// the given value. +// +// Key Registry will use this duration to create new keys. If the previous generated +// key exceed the given duration. Then the key registry will create new key. + +// The default value is set to 10 days. +func (opt Options) WithEncryptionKeyRotationDuration(d time.Duration) Options { + opt.EncryptionKeyRotationDuration = d + return opt +} + +// WithCompression is used to enable or disable compression. When compression is enabled, every +// block will be compressed using the specified algorithm. This option doesn't affect existing +// tables. Only the newly created tables will be compressed. +// +// The default compression algorithm used is snappy. Compression is enabled by default. +func (opt Options) WithCompression(cType options.CompressionType) Options { + opt.Compression = cType + return opt +} + +// WithVerifyValueChecksum is used to set VerifyValueChecksum. When VerifyValueChecksum is set to +// true, checksum will be verified for every entry read from the value log. If the value is stored +// in SST (value size less than value threshold) then the checksum validation will not be done. +// +// The default value of VerifyValueChecksum is False. +func (opt Options) WithVerifyValueChecksum(val bool) Options { + opt.VerifyValueChecksum = val + return opt +} + +// WithChecksumVerificationMode returns a new Options value with ChecksumVerificationMode set to +// the given value. +// +// ChecksumVerificationMode indicates when the db should verify checksums for SSTable blocks. +// +// The default value of VerifyValueChecksum is options.NoVerification. +func (opt Options) WithChecksumVerificationMode(cvMode options.ChecksumVerificationMode) Options { + opt.ChecksumVerificationMode = cvMode + return opt +} + +// WithBlockCacheSize returns a new Options value with BlockCacheSize set to the given value. +// +// This value specifies how much data cache should hold in memory. A small size +// of cache means lower memory consumption and lookups/iterations would take +// longer. It is recommended to use a cache if you're using compression or encryption. +// If compression and encryption both are disabled, adding a cache will lead to +// unnecessary overhead which will affect the read performance. Setting size to +// zero disables the cache altogether. +// +// Default value of BlockCacheSize is 256 MB. +func (opt Options) WithBlockCacheSize(size int64) Options { + opt.BlockCacheSize = size + return opt +} + +// WithInMemory returns a new Options value with Inmemory mode set to the given value. +// +// When badger is running in InMemory mode, everything is stored in memory. No value/sst files are +// created. In case of a crash all data will be lost. +func (opt Options) WithInMemory(b bool) Options { + opt.InMemory = b + return opt +} + +// WithZSTDCompressionLevel returns a new Options value with ZSTDCompressionLevel set +// to the given value. +// +// The ZSTD compression algorithm supports 20 compression levels. The higher the compression +// level, the better is the compression ratio but lower is the performance. Lower levels +// have better performance and higher levels have better compression ratios. +// We recommend using level 1 ZSTD Compression Level. Any level higher than 1 seems to +// deteriorate badger's performance. +// The following benchmarks were done on a 4 KB block size (default block size). The compression is +// ratio supposed to increase with increasing compression level but since the input for compression +// algorithm is small (4 KB), we don't get significant benefit at level 3. It is advised to write +// your own benchmarks before choosing a compression algorithm or level. +// +// NOTE: The benchmarks are with DataDog ZSTD that requires CGO. Hence, no longer valid. +// no_compression-16 10 502848865 ns/op 165.46 MB/s - +// zstd_compression/level_1-16 7 739037966 ns/op 112.58 MB/s 2.93 +// zstd_compression/level_3-16 7 756950250 ns/op 109.91 MB/s 2.72 +// zstd_compression/level_15-16 1 11135686219 ns/op 7.47 MB/s 4.38 +// Benchmark code can be found in table/builder_test.go file +func (opt Options) WithZSTDCompressionLevel(cLevel int) Options { + opt.ZSTDCompressionLevel = cLevel + return opt +} + +// WithBypassLockGuard returns a new Options value with BypassLockGuard +// set to the given value. +// +// When BypassLockGuard option is set, badger will not acquire a lock on the +// directory. This could lead to data corruption if multiple badger instances +// write to the same data directory. Use this option with caution. +// +// The default value of BypassLockGuard is false. +func (opt Options) WithBypassLockGuard(b bool) Options { + opt.BypassLockGuard = b + return opt +} + +// WithIndexCacheSize returns a new Options value with IndexCacheSize set to +// the given value. +// +// This value specifies how much memory should be used by table indices. These +// indices include the block offsets and the bloomfilters. Badger uses bloom +// filters to speed up lookups. Each table has its own bloom +// filter and each bloom filter is approximately of 5 MB. +// +// Zero value for IndexCacheSize means all the indices will be kept in +// memory and the cache is disabled. +// +// The default value of IndexCacheSize is 0 which means all indices are kept in +// memory. +func (opt Options) WithIndexCacheSize(size int64) Options { + opt.IndexCacheSize = size + return opt +} + +// WithDetectConflicts returns a new Options value with DetectConflicts set to the given value. +// +// Detect conflicts options determines if the transactions would be checked for +// conflicts before committing them. When this option is set to false +// (detectConflicts=false) badger can process transactions at a higher rate. +// Setting this options to false might be useful when the user application +// deals with conflict detection and resolution. +// +// The default value of Detect conflicts is True. +func (opt Options) WithDetectConflicts(b bool) Options { + opt.DetectConflicts = b + return opt +} + +// WithNamespaceOffset returns a new Options value with NamespaceOffset set to the given value. DB +// will expect the namespace in each key at the 8 bytes starting from NamespaceOffset. A negative +// value means that namespace is not stored in the key. +// +// The default value for NamespaceOffset is -1. +func (opt Options) WithNamespaceOffset(offset int) Options { + opt.NamespaceOffset = offset + return opt +} + +// WithExternalMagic returns a new Options value with ExternalMagicVersion set to the given value. +// The DB would fail to start if either the internal or the external magic number fails validated. +func (opt Options) WithExternalMagic(magic uint16) Options { + opt.ExternalMagicVersion = magic + return opt +} + +func (opt Options) getFileFlags() int { + var flags int + // opt.SyncWrites would be using msync to sync. All writes go through mmap. + if opt.ReadOnly { + flags |= os.O_RDONLY + } else { + flags |= os.O_RDWR + } + return flags +} diff --git a/vendor/github.com/dgraph-io/badger/v4/options/options.go b/vendor/github.com/dgraph-io/badger/v4/options/options.go new file mode 100644 index 00000000000..e5ee932c0c8 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/options/options.go @@ -0,0 +1,44 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package options + +// ChecksumVerificationMode tells when should DB verify checksum for SSTable blocks. +type ChecksumVerificationMode int + +const ( + // NoVerification indicates DB should not verify checksum for SSTable blocks. + NoVerification ChecksumVerificationMode = iota + // OnTableRead indicates checksum should be verified while opening SSTtable. + OnTableRead + // OnBlockRead indicates checksum should be verified on every SSTable block read. + OnBlockRead + // OnTableAndBlockRead indicates checksum should be verified + // on SSTable opening and on every block read. + OnTableAndBlockRead +) + +// CompressionType specifies how a block should be compressed. +type CompressionType uint32 + +const ( + // None mode indicates that a block is not compressed. + None CompressionType = 0 + // Snappy mode indicates that a block is compressed using Snappy algorithm. + Snappy CompressionType = 1 + // ZSTD mode indicates that a block is compressed using ZSTD algorithm. + ZSTD CompressionType = 2 +) diff --git a/vendor/github.com/dgraph-io/badger/v4/pb/badgerpb4.pb.go b/vendor/github.com/dgraph-io/badger/v4/pb/badgerpb4.pb.go new file mode 100644 index 00000000000..521f992937c --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/pb/badgerpb4.pb.go @@ -0,0 +1,2164 @@ +// Code generated by protoc-gen-gogo. DO NOT EDIT. +// source: badgerpb4.proto + +package pb + +import ( + fmt "fmt" + proto "github.com/gogo/protobuf/proto" + io "io" + math "math" + math_bits "math/bits" +) + +// Reference imports to suppress errors if they are not otherwise used. +var _ = proto.Marshal +var _ = fmt.Errorf +var _ = math.Inf + +// This is a compile-time assertion to ensure that this generated file +// is compatible with the proto package it is being compiled against. +// A compilation error at this line likely means your copy of the +// proto package needs to be updated. +const _ = proto.GoGoProtoPackageIsVersion3 // please upgrade the proto package + +type EncryptionAlgo int32 + +const ( + EncryptionAlgo_aes EncryptionAlgo = 0 +) + +var EncryptionAlgo_name = map[int32]string{ + 0: "aes", +} + +var EncryptionAlgo_value = map[string]int32{ + "aes": 0, +} + +func (x EncryptionAlgo) String() string { + return proto.EnumName(EncryptionAlgo_name, int32(x)) +} + +func (EncryptionAlgo) EnumDescriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{0} +} + +type ManifestChange_Operation int32 + +const ( + ManifestChange_CREATE ManifestChange_Operation = 0 + ManifestChange_DELETE ManifestChange_Operation = 1 +) + +var ManifestChange_Operation_name = map[int32]string{ + 0: "CREATE", + 1: "DELETE", +} + +var ManifestChange_Operation_value = map[string]int32{ + "CREATE": 0, + "DELETE": 1, +} + +func (x ManifestChange_Operation) String() string { + return proto.EnumName(ManifestChange_Operation_name, int32(x)) +} + +func (ManifestChange_Operation) EnumDescriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{3, 0} +} + +type Checksum_Algorithm int32 + +const ( + Checksum_CRC32C Checksum_Algorithm = 0 + Checksum_XXHash64 Checksum_Algorithm = 1 +) + +var Checksum_Algorithm_name = map[int32]string{ + 0: "CRC32C", + 1: "XXHash64", +} + +var Checksum_Algorithm_value = map[string]int32{ + "CRC32C": 0, + "XXHash64": 1, +} + +func (x Checksum_Algorithm) String() string { + return proto.EnumName(Checksum_Algorithm_name, int32(x)) +} + +func (Checksum_Algorithm) EnumDescriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{4, 0} +} + +type KV struct { + Key []byte `protobuf:"bytes,1,opt,name=key,proto3" json:"key,omitempty"` + Value []byte `protobuf:"bytes,2,opt,name=value,proto3" json:"value,omitempty"` + UserMeta []byte `protobuf:"bytes,3,opt,name=user_meta,json=userMeta,proto3" json:"user_meta,omitempty"` + Version uint64 `protobuf:"varint,4,opt,name=version,proto3" json:"version,omitempty"` + ExpiresAt uint64 `protobuf:"varint,5,opt,name=expires_at,json=expiresAt,proto3" json:"expires_at,omitempty"` + Meta []byte `protobuf:"bytes,6,opt,name=meta,proto3" json:"meta,omitempty"` + // Stream id is used to identify which stream the KV came from. + StreamId uint32 `protobuf:"varint,10,opt,name=stream_id,json=streamId,proto3" json:"stream_id,omitempty"` + // Stream done is used to indicate end of stream. + StreamDone bool `protobuf:"varint,11,opt,name=stream_done,json=streamDone,proto3" json:"stream_done,omitempty"` +} + +func (m *KV) Reset() { *m = KV{} } +func (m *KV) String() string { return proto.CompactTextString(m) } +func (*KV) ProtoMessage() {} +func (*KV) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{0} +} +func (m *KV) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *KV) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_KV.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *KV) XXX_Merge(src proto.Message) { + xxx_messageInfo_KV.Merge(m, src) +} +func (m *KV) XXX_Size() int { + return m.Size() +} +func (m *KV) XXX_DiscardUnknown() { + xxx_messageInfo_KV.DiscardUnknown(m) +} + +var xxx_messageInfo_KV proto.InternalMessageInfo + +func (m *KV) GetKey() []byte { + if m != nil { + return m.Key + } + return nil +} + +func (m *KV) GetValue() []byte { + if m != nil { + return m.Value + } + return nil +} + +func (m *KV) GetUserMeta() []byte { + if m != nil { + return m.UserMeta + } + return nil +} + +func (m *KV) GetVersion() uint64 { + if m != nil { + return m.Version + } + return 0 +} + +func (m *KV) GetExpiresAt() uint64 { + if m != nil { + return m.ExpiresAt + } + return 0 +} + +func (m *KV) GetMeta() []byte { + if m != nil { + return m.Meta + } + return nil +} + +func (m *KV) GetStreamId() uint32 { + if m != nil { + return m.StreamId + } + return 0 +} + +func (m *KV) GetStreamDone() bool { + if m != nil { + return m.StreamDone + } + return false +} + +type KVList struct { + Kv []*KV `protobuf:"bytes,1,rep,name=kv,proto3" json:"kv,omitempty"` + // alloc_ref used internally for memory management. + AllocRef uint64 `protobuf:"varint,10,opt,name=alloc_ref,json=allocRef,proto3" json:"alloc_ref,omitempty"` +} + +func (m *KVList) Reset() { *m = KVList{} } +func (m *KVList) String() string { return proto.CompactTextString(m) } +func (*KVList) ProtoMessage() {} +func (*KVList) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{1} +} +func (m *KVList) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *KVList) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_KVList.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *KVList) XXX_Merge(src proto.Message) { + xxx_messageInfo_KVList.Merge(m, src) +} +func (m *KVList) XXX_Size() int { + return m.Size() +} +func (m *KVList) XXX_DiscardUnknown() { + xxx_messageInfo_KVList.DiscardUnknown(m) +} + +var xxx_messageInfo_KVList proto.InternalMessageInfo + +func (m *KVList) GetKv() []*KV { + if m != nil { + return m.Kv + } + return nil +} + +func (m *KVList) GetAllocRef() uint64 { + if m != nil { + return m.AllocRef + } + return 0 +} + +type ManifestChangeSet struct { + // A set of changes that are applied atomically. + Changes []*ManifestChange `protobuf:"bytes,1,rep,name=changes,proto3" json:"changes,omitempty"` +} + +func (m *ManifestChangeSet) Reset() { *m = ManifestChangeSet{} } +func (m *ManifestChangeSet) String() string { return proto.CompactTextString(m) } +func (*ManifestChangeSet) ProtoMessage() {} +func (*ManifestChangeSet) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{2} +} +func (m *ManifestChangeSet) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *ManifestChangeSet) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_ManifestChangeSet.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *ManifestChangeSet) XXX_Merge(src proto.Message) { + xxx_messageInfo_ManifestChangeSet.Merge(m, src) +} +func (m *ManifestChangeSet) XXX_Size() int { + return m.Size() +} +func (m *ManifestChangeSet) XXX_DiscardUnknown() { + xxx_messageInfo_ManifestChangeSet.DiscardUnknown(m) +} + +var xxx_messageInfo_ManifestChangeSet proto.InternalMessageInfo + +func (m *ManifestChangeSet) GetChanges() []*ManifestChange { + if m != nil { + return m.Changes + } + return nil +} + +type ManifestChange struct { + Id uint64 `protobuf:"varint,1,opt,name=Id,proto3" json:"Id,omitempty"` + Op ManifestChange_Operation `protobuf:"varint,2,opt,name=Op,proto3,enum=badgerpb4.ManifestChange_Operation" json:"Op,omitempty"` + Level uint32 `protobuf:"varint,3,opt,name=Level,proto3" json:"Level,omitempty"` + KeyId uint64 `protobuf:"varint,4,opt,name=key_id,json=keyId,proto3" json:"key_id,omitempty"` + EncryptionAlgo EncryptionAlgo `protobuf:"varint,5,opt,name=encryption_algo,json=encryptionAlgo,proto3,enum=badgerpb4.EncryptionAlgo" json:"encryption_algo,omitempty"` + Compression uint32 `protobuf:"varint,6,opt,name=compression,proto3" json:"compression,omitempty"` +} + +func (m *ManifestChange) Reset() { *m = ManifestChange{} } +func (m *ManifestChange) String() string { return proto.CompactTextString(m) } +func (*ManifestChange) ProtoMessage() {} +func (*ManifestChange) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{3} +} +func (m *ManifestChange) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *ManifestChange) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_ManifestChange.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *ManifestChange) XXX_Merge(src proto.Message) { + xxx_messageInfo_ManifestChange.Merge(m, src) +} +func (m *ManifestChange) XXX_Size() int { + return m.Size() +} +func (m *ManifestChange) XXX_DiscardUnknown() { + xxx_messageInfo_ManifestChange.DiscardUnknown(m) +} + +var xxx_messageInfo_ManifestChange proto.InternalMessageInfo + +func (m *ManifestChange) GetId() uint64 { + if m != nil { + return m.Id + } + return 0 +} + +func (m *ManifestChange) GetOp() ManifestChange_Operation { + if m != nil { + return m.Op + } + return ManifestChange_CREATE +} + +func (m *ManifestChange) GetLevel() uint32 { + if m != nil { + return m.Level + } + return 0 +} + +func (m *ManifestChange) GetKeyId() uint64 { + if m != nil { + return m.KeyId + } + return 0 +} + +func (m *ManifestChange) GetEncryptionAlgo() EncryptionAlgo { + if m != nil { + return m.EncryptionAlgo + } + return EncryptionAlgo_aes +} + +func (m *ManifestChange) GetCompression() uint32 { + if m != nil { + return m.Compression + } + return 0 +} + +type Checksum struct { + Algo Checksum_Algorithm `protobuf:"varint,1,opt,name=algo,proto3,enum=badgerpb4.Checksum_Algorithm" json:"algo,omitempty"` + Sum uint64 `protobuf:"varint,2,opt,name=sum,proto3" json:"sum,omitempty"` +} + +func (m *Checksum) Reset() { *m = Checksum{} } +func (m *Checksum) String() string { return proto.CompactTextString(m) } +func (*Checksum) ProtoMessage() {} +func (*Checksum) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{4} +} +func (m *Checksum) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *Checksum) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_Checksum.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *Checksum) XXX_Merge(src proto.Message) { + xxx_messageInfo_Checksum.Merge(m, src) +} +func (m *Checksum) XXX_Size() int { + return m.Size() +} +func (m *Checksum) XXX_DiscardUnknown() { + xxx_messageInfo_Checksum.DiscardUnknown(m) +} + +var xxx_messageInfo_Checksum proto.InternalMessageInfo + +func (m *Checksum) GetAlgo() Checksum_Algorithm { + if m != nil { + return m.Algo + } + return Checksum_CRC32C +} + +func (m *Checksum) GetSum() uint64 { + if m != nil { + return m.Sum + } + return 0 +} + +type DataKey struct { + KeyId uint64 `protobuf:"varint,1,opt,name=key_id,json=keyId,proto3" json:"key_id,omitempty"` + Data []byte `protobuf:"bytes,2,opt,name=data,proto3" json:"data,omitempty"` + Iv []byte `protobuf:"bytes,3,opt,name=iv,proto3" json:"iv,omitempty"` + CreatedAt int64 `protobuf:"varint,4,opt,name=created_at,json=createdAt,proto3" json:"created_at,omitempty"` +} + +func (m *DataKey) Reset() { *m = DataKey{} } +func (m *DataKey) String() string { return proto.CompactTextString(m) } +func (*DataKey) ProtoMessage() {} +func (*DataKey) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{5} +} +func (m *DataKey) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *DataKey) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_DataKey.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *DataKey) XXX_Merge(src proto.Message) { + xxx_messageInfo_DataKey.Merge(m, src) +} +func (m *DataKey) XXX_Size() int { + return m.Size() +} +func (m *DataKey) XXX_DiscardUnknown() { + xxx_messageInfo_DataKey.DiscardUnknown(m) +} + +var xxx_messageInfo_DataKey proto.InternalMessageInfo + +func (m *DataKey) GetKeyId() uint64 { + if m != nil { + return m.KeyId + } + return 0 +} + +func (m *DataKey) GetData() []byte { + if m != nil { + return m.Data + } + return nil +} + +func (m *DataKey) GetIv() []byte { + if m != nil { + return m.Iv + } + return nil +} + +func (m *DataKey) GetCreatedAt() int64 { + if m != nil { + return m.CreatedAt + } + return 0 +} + +type Match struct { + Prefix []byte `protobuf:"bytes,1,opt,name=prefix,proto3" json:"prefix,omitempty"` + IgnoreBytes string `protobuf:"bytes,2,opt,name=ignore_bytes,json=ignoreBytes,proto3" json:"ignore_bytes,omitempty"` +} + +func (m *Match) Reset() { *m = Match{} } +func (m *Match) String() string { return proto.CompactTextString(m) } +func (*Match) ProtoMessage() {} +func (*Match) Descriptor() ([]byte, []int) { + return fileDescriptor_452c1d780baa15ef, []int{6} +} +func (m *Match) XXX_Unmarshal(b []byte) error { + return m.Unmarshal(b) +} +func (m *Match) XXX_Marshal(b []byte, deterministic bool) ([]byte, error) { + if deterministic { + return xxx_messageInfo_Match.Marshal(b, m, deterministic) + } else { + b = b[:cap(b)] + n, err := m.MarshalToSizedBuffer(b) + if err != nil { + return nil, err + } + return b[:n], nil + } +} +func (m *Match) XXX_Merge(src proto.Message) { + xxx_messageInfo_Match.Merge(m, src) +} +func (m *Match) XXX_Size() int { + return m.Size() +} +func (m *Match) XXX_DiscardUnknown() { + xxx_messageInfo_Match.DiscardUnknown(m) +} + +var xxx_messageInfo_Match proto.InternalMessageInfo + +func (m *Match) GetPrefix() []byte { + if m != nil { + return m.Prefix + } + return nil +} + +func (m *Match) GetIgnoreBytes() string { + if m != nil { + return m.IgnoreBytes + } + return "" +} + +func init() { + proto.RegisterEnum("badgerpb4.EncryptionAlgo", EncryptionAlgo_name, EncryptionAlgo_value) + proto.RegisterEnum("badgerpb4.ManifestChange_Operation", ManifestChange_Operation_name, ManifestChange_Operation_value) + proto.RegisterEnum("badgerpb4.Checksum_Algorithm", Checksum_Algorithm_name, Checksum_Algorithm_value) + proto.RegisterType((*KV)(nil), "badgerpb4.KV") + proto.RegisterType((*KVList)(nil), "badgerpb4.KVList") + proto.RegisterType((*ManifestChangeSet)(nil), "badgerpb4.ManifestChangeSet") + proto.RegisterType((*ManifestChange)(nil), "badgerpb4.ManifestChange") + proto.RegisterType((*Checksum)(nil), "badgerpb4.Checksum") + proto.RegisterType((*DataKey)(nil), "badgerpb4.DataKey") + proto.RegisterType((*Match)(nil), "badgerpb4.Match") +} + +func init() { proto.RegisterFile("badgerpb4.proto", fileDescriptor_452c1d780baa15ef) } + +var fileDescriptor_452c1d780baa15ef = []byte{ + // 653 bytes of a gzipped FileDescriptorProto + 0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x02, 0xff, 0x74, 0x54, 0x4f, 0x6b, 0xdb, 0x4e, + 0x10, 0xf5, 0xca, 0xf2, 0xbf, 0x71, 0xe2, 0xf8, 0xb7, 0xfc, 0x5a, 0x14, 0x4a, 0x5c, 0x47, 0xa1, + 0x60, 0x0a, 0xb5, 0x69, 0x1c, 0x7a, 0xe9, 0xc9, 0xff, 0x20, 0xc6, 0x09, 0x81, 0x6d, 0x08, 0xa1, + 0x17, 0xb3, 0x96, 0xc6, 0xb6, 0xb0, 0x2d, 0x89, 0xd5, 0x5a, 0xc4, 0x1f, 0xa2, 0xd0, 0x8f, 0xd5, + 0x63, 0x0e, 0x3d, 0xf4, 0x58, 0x92, 0x2f, 0x52, 0x76, 0xa5, 0xb8, 0xf6, 0xa1, 0xb7, 0x99, 0x37, + 0xa3, 0x79, 0xa3, 0xf7, 0x46, 0x82, 0xa3, 0x09, 0x77, 0x67, 0x28, 0xc2, 0xc9, 0x45, 0x33, 0x14, + 0x81, 0x0c, 0x68, 0x69, 0x0b, 0xd8, 0x3f, 0x09, 0x18, 0xa3, 0x3b, 0x5a, 0x85, 0xec, 0x02, 0x37, + 0x16, 0xa9, 0x93, 0xc6, 0x01, 0x53, 0x21, 0xfd, 0x1f, 0x72, 0x31, 0x5f, 0xae, 0xd1, 0x32, 0x34, + 0x96, 0x24, 0xf4, 0x0d, 0x94, 0xd6, 0x11, 0x8a, 0xf1, 0x0a, 0x25, 0xb7, 0xb2, 0xba, 0x52, 0x54, + 0xc0, 0x35, 0x4a, 0x4e, 0x2d, 0x28, 0xc4, 0x28, 0x22, 0x2f, 0xf0, 0x2d, 0xb3, 0x4e, 0x1a, 0x26, + 0x7b, 0x49, 0xe9, 0x09, 0x00, 0x3e, 0x84, 0x9e, 0xc0, 0x68, 0xcc, 0xa5, 0x95, 0xd3, 0xc5, 0x52, + 0x8a, 0x74, 0x24, 0xa5, 0x60, 0xea, 0x81, 0x79, 0x3d, 0x50, 0xc7, 0x8a, 0x29, 0x92, 0x02, 0xf9, + 0x6a, 0xec, 0xb9, 0x16, 0xd4, 0x49, 0xe3, 0x90, 0x15, 0x13, 0x60, 0xe8, 0xd2, 0xb7, 0x50, 0x4e, + 0x8b, 0x6e, 0xe0, 0xa3, 0x55, 0xae, 0x93, 0x46, 0x91, 0x41, 0x02, 0xf5, 0x03, 0x1f, 0xed, 0x3e, + 0xe4, 0x47, 0x77, 0x57, 0x5e, 0x24, 0xe9, 0x09, 0x18, 0x8b, 0xd8, 0x22, 0xf5, 0x6c, 0xa3, 0x7c, + 0x7e, 0xd8, 0xfc, 0xab, 0xc4, 0xe8, 0x8e, 0x19, 0x8b, 0x58, 0xd1, 0xf0, 0xe5, 0x32, 0x70, 0xc6, + 0x02, 0xa7, 0x9a, 0xc6, 0x64, 0x45, 0x0d, 0x30, 0x9c, 0xda, 0x97, 0xf0, 0xdf, 0x35, 0xf7, 0xbd, + 0x29, 0x46, 0xb2, 0x37, 0xe7, 0xfe, 0x0c, 0xbf, 0xa0, 0xa4, 0x6d, 0x28, 0x38, 0x3a, 0x89, 0xd2, + 0xa9, 0xc7, 0x3b, 0x53, 0xf7, 0xdb, 0xd9, 0x4b, 0xa7, 0xfd, 0xcd, 0x80, 0xca, 0x7e, 0x8d, 0x56, + 0xc0, 0x18, 0xba, 0x5a, 0x71, 0x93, 0x19, 0x43, 0x97, 0xb6, 0xc1, 0xb8, 0x09, 0xb5, 0xda, 0x95, + 0xf3, 0xb3, 0x7f, 0x8e, 0x6c, 0xde, 0x84, 0x28, 0xb8, 0xf4, 0x02, 0x9f, 0x19, 0x37, 0xa1, 0x72, + 0xe9, 0x0a, 0x63, 0x5c, 0x6a, 0x2f, 0x0e, 0x59, 0x92, 0xd0, 0x57, 0x90, 0x5f, 0xe0, 0x46, 0x09, + 0x97, 0xf8, 0x90, 0x5b, 0xe0, 0x66, 0xe8, 0xd2, 0x2e, 0x1c, 0xa1, 0xef, 0x88, 0x4d, 0xa8, 0x1e, + 0x1f, 0xf3, 0xe5, 0x2c, 0xd0, 0x56, 0x54, 0xf6, 0xde, 0x60, 0xb0, 0xed, 0xe8, 0x2c, 0x67, 0x01, + 0xab, 0xe0, 0x5e, 0x4e, 0xeb, 0x50, 0x76, 0x82, 0x55, 0x28, 0x30, 0xd2, 0x3e, 0xe7, 0x35, 0xed, + 0x2e, 0x64, 0x9f, 0x41, 0x69, 0xbb, 0x23, 0x05, 0xc8, 0xf7, 0xd8, 0xa0, 0x73, 0x3b, 0xa8, 0x66, + 0x54, 0xdc, 0x1f, 0x5c, 0x0d, 0x6e, 0x07, 0x55, 0x62, 0xc7, 0x50, 0xec, 0xcd, 0xd1, 0x59, 0x44, + 0xeb, 0x15, 0xfd, 0x08, 0xa6, 0xde, 0x85, 0xe8, 0x5d, 0x4e, 0x76, 0x76, 0x79, 0x69, 0x69, 0x2a, + 0x6a, 0xe1, 0xc9, 0xf9, 0x8a, 0xe9, 0x56, 0x75, 0xae, 0xd1, 0x7a, 0xa5, 0xc5, 0x32, 0x99, 0x0a, + 0xed, 0x77, 0x50, 0xda, 0x36, 0x25, 0xac, 0xbd, 0xf6, 0x79, 0xaf, 0x9a, 0xa1, 0x07, 0x50, 0xbc, + 0xbf, 0xbf, 0xe4, 0xd1, 0xfc, 0xd3, 0x45, 0x95, 0xd8, 0x0e, 0x14, 0xfa, 0x5c, 0xf2, 0x11, 0x6e, + 0x76, 0x44, 0x22, 0xbb, 0x22, 0x51, 0x30, 0x5d, 0x2e, 0x79, 0x7a, 0xf6, 0x3a, 0x56, 0x56, 0x79, + 0x71, 0x7a, 0xee, 0x86, 0x17, 0xab, 0x73, 0x76, 0x04, 0x72, 0x89, 0xae, 0x3a, 0x67, 0xa5, 0x71, + 0x96, 0x95, 0x52, 0xa4, 0x23, 0xed, 0x2e, 0xe4, 0xae, 0xb9, 0x74, 0xe6, 0xf4, 0x35, 0xe4, 0x43, + 0x81, 0x53, 0xef, 0x21, 0xfd, 0xb0, 0xd2, 0x8c, 0x9e, 0xc2, 0x81, 0x37, 0xf3, 0x03, 0x81, 0xe3, + 0xc9, 0x46, 0x62, 0xa4, 0xb9, 0x4a, 0xac, 0x9c, 0x60, 0x5d, 0x05, 0xbd, 0x3f, 0x86, 0xca, 0xbe, + 0x13, 0xb4, 0x00, 0x59, 0x8e, 0x51, 0x35, 0xd3, 0xfd, 0xfc, 0xe3, 0xa9, 0x46, 0x1e, 0x9f, 0x6a, + 0xe4, 0xf7, 0x53, 0x8d, 0x7c, 0x7f, 0xae, 0x65, 0x1e, 0x9f, 0x6b, 0x99, 0x5f, 0xcf, 0xb5, 0xcc, + 0xd7, 0xd3, 0x99, 0x27, 0xe7, 0xeb, 0x49, 0xd3, 0x09, 0x56, 0x2d, 0x77, 0x26, 0x78, 0x38, 0xff, + 0xe0, 0x05, 0xad, 0x44, 0xcf, 0x56, 0x7c, 0xd1, 0x0a, 0x27, 0x93, 0xbc, 0xfe, 0x03, 0xb4, 0xff, + 0x04, 0x00, 0x00, 0xff, 0xff, 0xec, 0x26, 0x3b, 0x76, 0x14, 0x04, 0x00, 0x00, +} + +func (m *KV) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *KV) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *KV) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if m.StreamDone { + i-- + if m.StreamDone { + dAtA[i] = 1 + } else { + dAtA[i] = 0 + } + i-- + dAtA[i] = 0x58 + } + if m.StreamId != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.StreamId)) + i-- + dAtA[i] = 0x50 + } + if len(m.Meta) > 0 { + i -= len(m.Meta) + copy(dAtA[i:], m.Meta) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.Meta))) + i-- + dAtA[i] = 0x32 + } + if m.ExpiresAt != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.ExpiresAt)) + i-- + dAtA[i] = 0x28 + } + if m.Version != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Version)) + i-- + dAtA[i] = 0x20 + } + if len(m.UserMeta) > 0 { + i -= len(m.UserMeta) + copy(dAtA[i:], m.UserMeta) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.UserMeta))) + i-- + dAtA[i] = 0x1a + } + if len(m.Value) > 0 { + i -= len(m.Value) + copy(dAtA[i:], m.Value) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.Value))) + i-- + dAtA[i] = 0x12 + } + if len(m.Key) > 0 { + i -= len(m.Key) + copy(dAtA[i:], m.Key) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.Key))) + i-- + dAtA[i] = 0xa + } + return len(dAtA) - i, nil +} + +func (m *KVList) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *KVList) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *KVList) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if m.AllocRef != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.AllocRef)) + i-- + dAtA[i] = 0x50 + } + if len(m.Kv) > 0 { + for iNdEx := len(m.Kv) - 1; iNdEx >= 0; iNdEx-- { + { + size, err := m.Kv[iNdEx].MarshalToSizedBuffer(dAtA[:i]) + if err != nil { + return 0, err + } + i -= size + i = encodeVarintBadgerpb4(dAtA, i, uint64(size)) + } + i-- + dAtA[i] = 0xa + } + } + return len(dAtA) - i, nil +} + +func (m *ManifestChangeSet) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *ManifestChangeSet) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *ManifestChangeSet) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if len(m.Changes) > 0 { + for iNdEx := len(m.Changes) - 1; iNdEx >= 0; iNdEx-- { + { + size, err := m.Changes[iNdEx].MarshalToSizedBuffer(dAtA[:i]) + if err != nil { + return 0, err + } + i -= size + i = encodeVarintBadgerpb4(dAtA, i, uint64(size)) + } + i-- + dAtA[i] = 0xa + } + } + return len(dAtA) - i, nil +} + +func (m *ManifestChange) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *ManifestChange) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *ManifestChange) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if m.Compression != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Compression)) + i-- + dAtA[i] = 0x30 + } + if m.EncryptionAlgo != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.EncryptionAlgo)) + i-- + dAtA[i] = 0x28 + } + if m.KeyId != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.KeyId)) + i-- + dAtA[i] = 0x20 + } + if m.Level != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Level)) + i-- + dAtA[i] = 0x18 + } + if m.Op != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Op)) + i-- + dAtA[i] = 0x10 + } + if m.Id != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Id)) + i-- + dAtA[i] = 0x8 + } + return len(dAtA) - i, nil +} + +func (m *Checksum) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *Checksum) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *Checksum) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if m.Sum != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Sum)) + i-- + dAtA[i] = 0x10 + } + if m.Algo != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.Algo)) + i-- + dAtA[i] = 0x8 + } + return len(dAtA) - i, nil +} + +func (m *DataKey) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *DataKey) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *DataKey) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if m.CreatedAt != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.CreatedAt)) + i-- + dAtA[i] = 0x20 + } + if len(m.Iv) > 0 { + i -= len(m.Iv) + copy(dAtA[i:], m.Iv) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.Iv))) + i-- + dAtA[i] = 0x1a + } + if len(m.Data) > 0 { + i -= len(m.Data) + copy(dAtA[i:], m.Data) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.Data))) + i-- + dAtA[i] = 0x12 + } + if m.KeyId != 0 { + i = encodeVarintBadgerpb4(dAtA, i, uint64(m.KeyId)) + i-- + dAtA[i] = 0x8 + } + return len(dAtA) - i, nil +} + +func (m *Match) Marshal() (dAtA []byte, err error) { + size := m.Size() + dAtA = make([]byte, size) + n, err := m.MarshalToSizedBuffer(dAtA[:size]) + if err != nil { + return nil, err + } + return dAtA[:n], nil +} + +func (m *Match) MarshalTo(dAtA []byte) (int, error) { + size := m.Size() + return m.MarshalToSizedBuffer(dAtA[:size]) +} + +func (m *Match) MarshalToSizedBuffer(dAtA []byte) (int, error) { + i := len(dAtA) + _ = i + var l int + _ = l + if len(m.IgnoreBytes) > 0 { + i -= len(m.IgnoreBytes) + copy(dAtA[i:], m.IgnoreBytes) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.IgnoreBytes))) + i-- + dAtA[i] = 0x12 + } + if len(m.Prefix) > 0 { + i -= len(m.Prefix) + copy(dAtA[i:], m.Prefix) + i = encodeVarintBadgerpb4(dAtA, i, uint64(len(m.Prefix))) + i-- + dAtA[i] = 0xa + } + return len(dAtA) - i, nil +} + +func encodeVarintBadgerpb4(dAtA []byte, offset int, v uint64) int { + offset -= sovBadgerpb4(v) + base := offset + for v >= 1<<7 { + dAtA[offset] = uint8(v&0x7f | 0x80) + v >>= 7 + offset++ + } + dAtA[offset] = uint8(v) + return base +} +func (m *KV) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + l = len(m.Key) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + l = len(m.Value) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + l = len(m.UserMeta) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + if m.Version != 0 { + n += 1 + sovBadgerpb4(uint64(m.Version)) + } + if m.ExpiresAt != 0 { + n += 1 + sovBadgerpb4(uint64(m.ExpiresAt)) + } + l = len(m.Meta) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + if m.StreamId != 0 { + n += 1 + sovBadgerpb4(uint64(m.StreamId)) + } + if m.StreamDone { + n += 2 + } + return n +} + +func (m *KVList) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + if len(m.Kv) > 0 { + for _, e := range m.Kv { + l = e.Size() + n += 1 + l + sovBadgerpb4(uint64(l)) + } + } + if m.AllocRef != 0 { + n += 1 + sovBadgerpb4(uint64(m.AllocRef)) + } + return n +} + +func (m *ManifestChangeSet) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + if len(m.Changes) > 0 { + for _, e := range m.Changes { + l = e.Size() + n += 1 + l + sovBadgerpb4(uint64(l)) + } + } + return n +} + +func (m *ManifestChange) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + if m.Id != 0 { + n += 1 + sovBadgerpb4(uint64(m.Id)) + } + if m.Op != 0 { + n += 1 + sovBadgerpb4(uint64(m.Op)) + } + if m.Level != 0 { + n += 1 + sovBadgerpb4(uint64(m.Level)) + } + if m.KeyId != 0 { + n += 1 + sovBadgerpb4(uint64(m.KeyId)) + } + if m.EncryptionAlgo != 0 { + n += 1 + sovBadgerpb4(uint64(m.EncryptionAlgo)) + } + if m.Compression != 0 { + n += 1 + sovBadgerpb4(uint64(m.Compression)) + } + return n +} + +func (m *Checksum) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + if m.Algo != 0 { + n += 1 + sovBadgerpb4(uint64(m.Algo)) + } + if m.Sum != 0 { + n += 1 + sovBadgerpb4(uint64(m.Sum)) + } + return n +} + +func (m *DataKey) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + if m.KeyId != 0 { + n += 1 + sovBadgerpb4(uint64(m.KeyId)) + } + l = len(m.Data) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + l = len(m.Iv) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + if m.CreatedAt != 0 { + n += 1 + sovBadgerpb4(uint64(m.CreatedAt)) + } + return n +} + +func (m *Match) Size() (n int) { + if m == nil { + return 0 + } + var l int + _ = l + l = len(m.Prefix) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + l = len(m.IgnoreBytes) + if l > 0 { + n += 1 + l + sovBadgerpb4(uint64(l)) + } + return n +} + +func sovBadgerpb4(x uint64) (n int) { + return (math_bits.Len64(x|1) + 6) / 7 +} +func sozBadgerpb4(x uint64) (n int) { + return sovBadgerpb4(uint64((x << 1) ^ uint64((int64(x) >> 63)))) +} +func (m *KV) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: KV: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: KV: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Key", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Key = append(m.Key[:0], dAtA[iNdEx:postIndex]...) + if m.Key == nil { + m.Key = []byte{} + } + iNdEx = postIndex + case 2: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Value", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Value = append(m.Value[:0], dAtA[iNdEx:postIndex]...) + if m.Value == nil { + m.Value = []byte{} + } + iNdEx = postIndex + case 3: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field UserMeta", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.UserMeta = append(m.UserMeta[:0], dAtA[iNdEx:postIndex]...) + if m.UserMeta == nil { + m.UserMeta = []byte{} + } + iNdEx = postIndex + case 4: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Version", wireType) + } + m.Version = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Version |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 5: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field ExpiresAt", wireType) + } + m.ExpiresAt = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.ExpiresAt |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 6: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Meta", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Meta = append(m.Meta[:0], dAtA[iNdEx:postIndex]...) + if m.Meta == nil { + m.Meta = []byte{} + } + iNdEx = postIndex + case 10: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field StreamId", wireType) + } + m.StreamId = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.StreamId |= uint32(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 11: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field StreamDone", wireType) + } + var v int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + v |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + m.StreamDone = bool(v != 0) + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *KVList) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: KVList: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: KVList: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Kv", wireType) + } + var msglen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + msglen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if msglen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + msglen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Kv = append(m.Kv, &KV{}) + if err := m.Kv[len(m.Kv)-1].Unmarshal(dAtA[iNdEx:postIndex]); err != nil { + return err + } + iNdEx = postIndex + case 10: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field AllocRef", wireType) + } + m.AllocRef = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.AllocRef |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *ManifestChangeSet) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: ManifestChangeSet: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: ManifestChangeSet: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Changes", wireType) + } + var msglen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + msglen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if msglen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + msglen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Changes = append(m.Changes, &ManifestChange{}) + if err := m.Changes[len(m.Changes)-1].Unmarshal(dAtA[iNdEx:postIndex]); err != nil { + return err + } + iNdEx = postIndex + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *ManifestChange) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: ManifestChange: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: ManifestChange: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Id", wireType) + } + m.Id = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Id |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 2: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Op", wireType) + } + m.Op = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Op |= ManifestChange_Operation(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 3: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Level", wireType) + } + m.Level = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Level |= uint32(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 4: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field KeyId", wireType) + } + m.KeyId = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.KeyId |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 5: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field EncryptionAlgo", wireType) + } + m.EncryptionAlgo = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.EncryptionAlgo |= EncryptionAlgo(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 6: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Compression", wireType) + } + m.Compression = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Compression |= uint32(b&0x7F) << shift + if b < 0x80 { + break + } + } + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *Checksum) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: Checksum: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: Checksum: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Algo", wireType) + } + m.Algo = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Algo |= Checksum_Algorithm(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 2: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field Sum", wireType) + } + m.Sum = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.Sum |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *DataKey) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: DataKey: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: DataKey: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field KeyId", wireType) + } + m.KeyId = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.KeyId |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + case 2: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Data", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Data = append(m.Data[:0], dAtA[iNdEx:postIndex]...) + if m.Data == nil { + m.Data = []byte{} + } + iNdEx = postIndex + case 3: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Iv", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Iv = append(m.Iv[:0], dAtA[iNdEx:postIndex]...) + if m.Iv == nil { + m.Iv = []byte{} + } + iNdEx = postIndex + case 4: + if wireType != 0 { + return fmt.Errorf("proto: wrong wireType = %d for field CreatedAt", wireType) + } + m.CreatedAt = 0 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + m.CreatedAt |= int64(b&0x7F) << shift + if b < 0x80 { + break + } + } + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func (m *Match) Unmarshal(dAtA []byte) error { + l := len(dAtA) + iNdEx := 0 + for iNdEx < l { + preIndex := iNdEx + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + fieldNum := int32(wire >> 3) + wireType := int(wire & 0x7) + if wireType == 4 { + return fmt.Errorf("proto: Match: wiretype end group for non-group") + } + if fieldNum <= 0 { + return fmt.Errorf("proto: Match: illegal tag %d (wire type %d)", fieldNum, wire) + } + switch fieldNum { + case 1: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field Prefix", wireType) + } + var byteLen int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + byteLen |= int(b&0x7F) << shift + if b < 0x80 { + break + } + } + if byteLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + byteLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.Prefix = append(m.Prefix[:0], dAtA[iNdEx:postIndex]...) + if m.Prefix == nil { + m.Prefix = []byte{} + } + iNdEx = postIndex + case 2: + if wireType != 2 { + return fmt.Errorf("proto: wrong wireType = %d for field IgnoreBytes", wireType) + } + var stringLen uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + stringLen |= uint64(b&0x7F) << shift + if b < 0x80 { + break + } + } + intStringLen := int(stringLen) + if intStringLen < 0 { + return ErrInvalidLengthBadgerpb4 + } + postIndex := iNdEx + intStringLen + if postIndex < 0 { + return ErrInvalidLengthBadgerpb4 + } + if postIndex > l { + return io.ErrUnexpectedEOF + } + m.IgnoreBytes = string(dAtA[iNdEx:postIndex]) + iNdEx = postIndex + default: + iNdEx = preIndex + skippy, err := skipBadgerpb4(dAtA[iNdEx:]) + if err != nil { + return err + } + if (skippy < 0) || (iNdEx+skippy) < 0 { + return ErrInvalidLengthBadgerpb4 + } + if (iNdEx + skippy) > l { + return io.ErrUnexpectedEOF + } + iNdEx += skippy + } + } + + if iNdEx > l { + return io.ErrUnexpectedEOF + } + return nil +} +func skipBadgerpb4(dAtA []byte) (n int, err error) { + l := len(dAtA) + iNdEx := 0 + depth := 0 + for iNdEx < l { + var wire uint64 + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + wire |= (uint64(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + wireType := int(wire & 0x7) + switch wireType { + case 0: + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + iNdEx++ + if dAtA[iNdEx-1] < 0x80 { + break + } + } + case 1: + iNdEx += 8 + case 2: + var length int + for shift := uint(0); ; shift += 7 { + if shift >= 64 { + return 0, ErrIntOverflowBadgerpb4 + } + if iNdEx >= l { + return 0, io.ErrUnexpectedEOF + } + b := dAtA[iNdEx] + iNdEx++ + length |= (int(b) & 0x7F) << shift + if b < 0x80 { + break + } + } + if length < 0 { + return 0, ErrInvalidLengthBadgerpb4 + } + iNdEx += length + case 3: + depth++ + case 4: + if depth == 0 { + return 0, ErrUnexpectedEndOfGroupBadgerpb4 + } + depth-- + case 5: + iNdEx += 4 + default: + return 0, fmt.Errorf("proto: illegal wireType %d", wireType) + } + if iNdEx < 0 { + return 0, ErrInvalidLengthBadgerpb4 + } + if depth == 0 { + return iNdEx, nil + } + } + return 0, io.ErrUnexpectedEOF +} + +var ( + ErrInvalidLengthBadgerpb4 = fmt.Errorf("proto: negative length found during unmarshaling") + ErrIntOverflowBadgerpb4 = fmt.Errorf("proto: integer overflow") + ErrUnexpectedEndOfGroupBadgerpb4 = fmt.Errorf("proto: unexpected end of group") +) diff --git a/vendor/github.com/dgraph-io/badger/v4/pb/badgerpb4.proto b/vendor/github.com/dgraph-io/badger/v4/pb/badgerpb4.proto new file mode 100644 index 00000000000..079c1cfeeb3 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/pb/badgerpb4.proto @@ -0,0 +1,86 @@ +/* + * Copyright (C) 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +// Use protos/gen.sh to generate .pb.go files. +syntax = "proto3"; + +package badgerpb4; + +option go_package = "github.com/dgraph-io/badger/v4/pb"; + +message KV { + bytes key = 1; + bytes value = 2; + bytes user_meta = 3; + uint64 version = 4; + uint64 expires_at = 5; + bytes meta = 6; + + // Stream id is used to identify which stream the KV came from. + uint32 stream_id = 10; + // Stream done is used to indicate end of stream. + bool stream_done = 11; +} + +message KVList { + repeated KV kv = 1; + + // alloc_ref used internally for memory management. + uint64 alloc_ref = 10; +} + +message ManifestChangeSet { + // A set of changes that are applied atomically. + repeated ManifestChange changes = 1; +} + +enum EncryptionAlgo { + aes = 0; +} + +message ManifestChange { + uint64 Id = 1; // Table ID. + enum Operation { + CREATE = 0; + DELETE = 1; + } + Operation Op = 2; + uint32 Level = 3; // Only used for CREATE. + uint64 key_id = 4; + EncryptionAlgo encryption_algo = 5; + uint32 compression = 6; // Only used for CREATE Op. +} + +message Checksum { + enum Algorithm { + CRC32C = 0; + XXHash64 = 1; + } + Algorithm algo = 1; // For storing type of Checksum algorithm used + uint64 sum = 2; +} + +message DataKey { + uint64 key_id = 1; + bytes data = 2; + bytes iv = 3; + int64 created_at = 4; +} + +message Match { + bytes prefix = 1; + string ignore_bytes = 2; // Comma separated with dash to represent ranges "1, 2-3, 4-7, 9" +} diff --git a/vendor/github.com/dgraph-io/badger/v4/pb/gen.sh b/vendor/github.com/dgraph-io/badger/v4/pb/gen.sh new file mode 100644 index 00000000000..864845bcfd0 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/pb/gen.sh @@ -0,0 +1,7 @@ +#!/bin/bash + +# Run this script from its directory, so that badgerpb4.proto is where it's expected to +# be. + +go install github.com/gogo/protobuf/protoc-gen-gogofaster@latest +protoc --gogofaster_out=. --gogofaster_opt=paths=source_relative -I=. badgerpb4.proto diff --git a/vendor/github.com/dgraph-io/badger/v4/publisher.go b/vendor/github.com/dgraph-io/badger/v4/publisher.go new file mode 100644 index 00000000000..dce524f0907 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/publisher.go @@ -0,0 +1,180 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "sync" + "sync/atomic" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/trie" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +type subscriber struct { + id uint64 + matches []pb.Match + sendCh chan *pb.KVList + subCloser *z.Closer + // this will be atomic pointer which will be used to + // track whether the subscriber is active or not + active *atomic.Uint64 +} + +type publisher struct { + sync.Mutex + pubCh chan requests + subscribers map[uint64]subscriber + nextID uint64 + indexer *trie.Trie +} + +func newPublisher() *publisher { + return &publisher{ + pubCh: make(chan requests, 1000), + subscribers: make(map[uint64]subscriber), + nextID: 0, + indexer: trie.NewTrie(), + } +} + +func (p *publisher) listenForUpdates(c *z.Closer) { + defer func() { + p.cleanSubscribers() + c.Done() + }() + slurp := func(batch requests) { + for { + select { + case reqs := <-p.pubCh: + batch = append(batch, reqs...) + default: + p.publishUpdates(batch) + return + } + } + } + for { + select { + case <-c.HasBeenClosed(): + return + case reqs := <-p.pubCh: + slurp(reqs) + } + } +} + +func (p *publisher) publishUpdates(reqs requests) { + p.Lock() + defer func() { + p.Unlock() + // Release all the request. + reqs.DecrRef() + }() + batchedUpdates := make(map[uint64]*pb.KVList) + for _, req := range reqs { + for _, e := range req.Entries { + ids := p.indexer.Get(e.Key) + if len(ids) == 0 { + continue + } + k := y.SafeCopy(nil, e.Key) + kv := &pb.KV{ + Key: y.ParseKey(k), + Value: y.SafeCopy(nil, e.Value), + Meta: []byte{e.UserMeta}, + ExpiresAt: e.ExpiresAt, + Version: y.ParseTs(k), + } + for id := range ids { + if _, ok := batchedUpdates[id]; !ok { + batchedUpdates[id] = &pb.KVList{} + } + batchedUpdates[id].Kv = append(batchedUpdates[id].Kv, kv) + } + } + } + + for id, kvs := range batchedUpdates { + if p.subscribers[id].active.Load() == 1 { + p.subscribers[id].sendCh <- kvs + } + } +} + +func (p *publisher) newSubscriber(c *z.Closer, matches []pb.Match) (subscriber, error) { + p.Lock() + defer p.Unlock() + ch := make(chan *pb.KVList, 1000) + id := p.nextID + // Increment next ID. + p.nextID++ + s := subscriber{ + id: id, + matches: matches, + sendCh: ch, + subCloser: c, + active: new(atomic.Uint64), + } + s.active.Store(1) + + p.subscribers[id] = s + for _, m := range matches { + if err := p.indexer.AddMatch(m, id); err != nil { + return subscriber{}, err + } + } + return s, nil +} + +// cleanSubscribers stops all the subscribers. Ideally, It should be called while closing DB. +func (p *publisher) cleanSubscribers() { + p.Lock() + defer p.Unlock() + for id, s := range p.subscribers { + for _, m := range s.matches { + _ = p.indexer.DeleteMatch(m, id) + } + delete(p.subscribers, id) + s.subCloser.SignalAndWait() + } +} + +func (p *publisher) deleteSubscriber(id uint64) { + p.Lock() + defer p.Unlock() + if s, ok := p.subscribers[id]; ok { + for _, m := range s.matches { + _ = p.indexer.DeleteMatch(m, id) + } + } + delete(p.subscribers, id) +} + +func (p *publisher) sendUpdates(reqs requests) { + if p.noOfSubscribers() != 0 { + reqs.IncrRef() + p.pubCh <- reqs + } +} + +func (p *publisher) noOfSubscribers() int { + p.Lock() + defer p.Unlock() + return len(p.subscribers) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/skl/README.md b/vendor/github.com/dgraph-io/badger/v4/skl/README.md new file mode 100644 index 00000000000..e22e4590bbf --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/skl/README.md @@ -0,0 +1,113 @@ +This is much better than `skiplist` and `slist`. + +``` +BenchmarkReadWrite/frac_0-8 3000000 537 ns/op +BenchmarkReadWrite/frac_1-8 3000000 503 ns/op +BenchmarkReadWrite/frac_2-8 3000000 492 ns/op +BenchmarkReadWrite/frac_3-8 3000000 475 ns/op +BenchmarkReadWrite/frac_4-8 3000000 440 ns/op +BenchmarkReadWrite/frac_5-8 5000000 442 ns/op +BenchmarkReadWrite/frac_6-8 5000000 380 ns/op +BenchmarkReadWrite/frac_7-8 5000000 338 ns/op +BenchmarkReadWrite/frac_8-8 5000000 294 ns/op +BenchmarkReadWrite/frac_9-8 10000000 268 ns/op +BenchmarkReadWrite/frac_10-8 100000000 26.3 ns/op +``` + +And even better than a simple map with read-write lock: + +``` +BenchmarkReadWriteMap/frac_0-8 2000000 774 ns/op +BenchmarkReadWriteMap/frac_1-8 2000000 647 ns/op +BenchmarkReadWriteMap/frac_2-8 3000000 605 ns/op +BenchmarkReadWriteMap/frac_3-8 3000000 603 ns/op +BenchmarkReadWriteMap/frac_4-8 3000000 556 ns/op +BenchmarkReadWriteMap/frac_5-8 3000000 472 ns/op +BenchmarkReadWriteMap/frac_6-8 3000000 476 ns/op +BenchmarkReadWriteMap/frac_7-8 3000000 457 ns/op +BenchmarkReadWriteMap/frac_8-8 5000000 444 ns/op +BenchmarkReadWriteMap/frac_9-8 5000000 361 ns/op +BenchmarkReadWriteMap/frac_10-8 10000000 212 ns/op +``` + +# Node Pooling + +Command used + +``` +rm -Rf tmp && /usr/bin/time -l ./populate -keys_mil 10 +``` + +For pprof results, we run without using /usr/bin/time. There are four runs below. + +Results seem to vary quite a bit between runs. + +## Before node pooling + +``` +1311.53MB of 1338.69MB total (97.97%) +Dropped 30 nodes (cum <= 6.69MB) +Showing top 10 nodes out of 37 (cum >= 12.50MB) + flat flat% sum% cum cum% + 523.04MB 39.07% 39.07% 523.04MB 39.07% github.com/dgraph-io/badger/skl.(*Skiplist).Put + 184.51MB 13.78% 52.85% 184.51MB 13.78% runtime.stringtoslicebyte + 166.01MB 12.40% 65.25% 689.04MB 51.47% github.com/dgraph-io/badger/mem.(*Table).Put + 165MB 12.33% 77.58% 165MB 12.33% runtime.convT2E + 116.92MB 8.73% 86.31% 116.92MB 8.73% bytes.makeSlice + 62.50MB 4.67% 90.98% 62.50MB 4.67% main.newValue + 34.50MB 2.58% 93.56% 34.50MB 2.58% github.com/dgraph-io/badger/table.(*BlockIterator).parseKV + 25.50MB 1.90% 95.46% 100.06MB 7.47% github.com/dgraph-io/badger/y.(*MergeIterator).Next + 21.06MB 1.57% 97.04% 21.06MB 1.57% github.com/dgraph-io/badger/table.(*Table).read + 12.50MB 0.93% 97.97% 12.50MB 0.93% github.com/dgraph-io/badger/table.header.Encode + + 128.31 real 329.37 user 17.11 sys +3355660288 maximum resident set size + 0 average shared memory size + 0 average unshared data size + 0 average unshared stack size + 2203080 page reclaims + 764 page faults + 0 swaps + 275 block input operations + 76 block output operations + 0 messages sent + 0 messages received + 0 signals received + 49173 voluntary context switches + 599922 involuntary context switches +``` + +## After node pooling + +``` +1963.13MB of 2026.09MB total (96.89%) +Dropped 29 nodes (cum <= 10.13MB) +Showing top 10 nodes out of 41 (cum >= 185.62MB) + flat flat% sum% cum cum% + 658.05MB 32.48% 32.48% 658.05MB 32.48% github.com/dgraph-io/badger/skl.glob..func1 + 297.51MB 14.68% 47.16% 297.51MB 14.68% runtime.convT2E + 257.51MB 12.71% 59.87% 257.51MB 12.71% runtime.stringtoslicebyte + 249.01MB 12.29% 72.16% 1007.06MB 49.70% github.com/dgraph-io/badger/mem.(*Table).Put + 142.43MB 7.03% 79.19% 142.43MB 7.03% bytes.makeSlice + 100MB 4.94% 84.13% 758.05MB 37.41% github.com/dgraph-io/badger/skl.newNode + 99.50MB 4.91% 89.04% 99.50MB 4.91% main.newValue + 75MB 3.70% 92.74% 75MB 3.70% github.com/dgraph-io/badger/table.(*BlockIterator).parseKV + 44.62MB 2.20% 94.94% 44.62MB 2.20% github.com/dgraph-io/badger/table.(*Table).read + 39.50MB 1.95% 96.89% 185.62MB 9.16% github.com/dgraph-io/badger/y.(*MergeIterator).Next + + 135.58 real 374.29 user 17.65 sys +3740614656 maximum resident set size + 0 average shared memory size + 0 average unshared data size + 0 average unshared stack size + 2276566 page reclaims + 770 page faults + 0 swaps + 128 block input operations + 90 block output operations + 0 messages sent + 0 messages received + 0 signals received + 46434 voluntary context switches + 597049 involuntary context switches +``` diff --git a/vendor/github.com/dgraph-io/badger/v4/skl/arena.go b/vendor/github.com/dgraph-io/badger/v4/skl/arena.go new file mode 100644 index 00000000000..b471e8101cf --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/skl/arena.go @@ -0,0 +1,133 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package skl + +import ( + "sync/atomic" + "unsafe" + + "github.com/dgraph-io/badger/v4/y" +) + +const ( + offsetSize = int(unsafe.Sizeof(uint32(0))) + + // Always align nodes on 64-bit boundaries, even on 32-bit architectures, + // so that the node.value field is 64-bit aligned. This is necessary because + // node.getValueOffset uses atomic.LoadUint64, which expects its input + // pointer to be 64-bit aligned. + nodeAlign = int(unsafe.Sizeof(uint64(0))) - 1 +) + +// Arena should be lock-free. +type Arena struct { + n atomic.Uint32 + buf []byte +} + +// newArena returns a new arena. +func newArena(n int64) *Arena { + // Don't store data at position 0 in order to reserve offset=0 as a kind + // of nil pointer. + out := &Arena{buf: make([]byte, n)} + out.n.Store(1) + return out +} + +func (s *Arena) size() int64 { + return int64(s.n.Load()) +} + +// putNode allocates a node in the arena. The node is aligned on a pointer-sized +// boundary. The arena offset of the node is returned. +func (s *Arena) putNode(height int) uint32 { + // Compute the amount of the tower that will never be used, since the height + // is less than maxHeight. + unusedSize := (maxHeight - height) * offsetSize + + // Pad the allocation with enough bytes to ensure pointer alignment. + l := uint32(MaxNodeSize - unusedSize + nodeAlign) + n := s.n.Add(l) + y.AssertTruef(int(n) <= len(s.buf), + "Arena too small, toWrite:%d newTotal:%d limit:%d", + l, n, len(s.buf)) + + // Return the aligned offset. + m := (n - l + uint32(nodeAlign)) & ^uint32(nodeAlign) + return m +} + +// Put will *copy* val into arena. To make better use of this, reuse your input +// val buffer. Returns an offset into buf. User is responsible for remembering +// size of val. We could also store this size inside arena but the encoding and +// decoding will incur some overhead. +func (s *Arena) putVal(v y.ValueStruct) uint32 { + l := v.EncodedSize() + n := s.n.Add(l) + y.AssertTruef(int(n) <= len(s.buf), + "Arena too small, toWrite:%d newTotal:%d limit:%d", + l, n, len(s.buf)) + m := n - l + v.Encode(s.buf[m:]) + return m +} + +func (s *Arena) putKey(key []byte) uint32 { + l := uint32(len(key)) + n := s.n.Add(l) + y.AssertTruef(int(n) <= len(s.buf), + "Arena too small, toWrite:%d newTotal:%d limit:%d", + l, n, len(s.buf)) + // m is the offset where you should write. + // n = new len - key len give you the offset at which you should write. + m := n - l + // Copy to buffer from m:n + y.AssertTrue(len(key) == copy(s.buf[m:n], key)) + return m +} + +// getNode returns a pointer to the node located at offset. If the offset is +// zero, then the nil node pointer is returned. +func (s *Arena) getNode(offset uint32) *node { + if offset == 0 { + return nil + } + + return (*node)(unsafe.Pointer(&s.buf[offset])) +} + +// getKey returns byte slice at offset. +func (s *Arena) getKey(offset uint32, size uint16) []byte { + return s.buf[offset : offset+uint32(size)] +} + +// getVal returns byte slice at offset. The given size should be just the value +// size and should NOT include the meta bytes. +func (s *Arena) getVal(offset uint32, size uint32) (ret y.ValueStruct) { + ret.Decode(s.buf[offset : offset+size]) + return +} + +// getNodeOffset returns the offset of node in the arena. If the node pointer is +// nil, then the zero offset is returned. +func (s *Arena) getNodeOffset(nd *node) uint32 { + if nd == nil { + return 0 + } + + return uint32(uintptr(unsafe.Pointer(nd)) - uintptr(unsafe.Pointer(&s.buf[0]))) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/skl/skl.go b/vendor/github.com/dgraph-io/badger/v4/skl/skl.go new file mode 100644 index 00000000000..3fe8cc4a885 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/skl/skl.go @@ -0,0 +1,522 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +/* +Adapted from RocksDB inline skiplist. + +Key differences: +- No optimization for sequential inserts (no "prev"). +- No custom comparator. +- Support overwrites. This requires care when we see the same key when inserting. + For RocksDB or LevelDB, overwrites are implemented as a newer sequence number in the key, so + there is no need for values. We don't intend to support versioning. In-place updates of values + would be more efficient. +- We discard all non-concurrent code. +- We do not support Splices. This simplifies the code a lot. +- No AllocateNode or other pointer arithmetic. +- We combine the findLessThan, findGreaterOrEqual, etc into one function. +*/ + +package skl + +import ( + "math" + "sync/atomic" + "unsafe" + + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +const ( + maxHeight = 20 + heightIncrease = math.MaxUint32 / 3 +) + +// MaxNodeSize is the memory footprint of a node of maximum height. +const MaxNodeSize = int(unsafe.Sizeof(node{})) + +type node struct { + // Multiple parts of the value are encoded as a single uint64 so that it + // can be atomically loaded and stored: + // value offset: uint32 (bits 0-31) + // value size : uint16 (bits 32-63) + value atomic.Uint64 + + // A byte slice is 24 bytes. We are trying to save space here. + keyOffset uint32 // Immutable. No need to lock to access key. + keySize uint16 // Immutable. No need to lock to access key. + + // Height of the tower. + height uint16 + + // Most nodes do not need to use the full height of the tower, since the + // probability of each successive level decreases exponentially. Because + // these elements are never accessed, they do not need to be allocated. + // Therefore, when a node is allocated in the arena, its memory footprint + // is deliberately truncated to not include unneeded tower elements. + // + // All accesses to elements should use CAS operations, with no need to lock. + tower [maxHeight]atomic.Uint32 +} + +type Skiplist struct { + height atomic.Int32 // Current height. 1 <= height <= kMaxHeight. CAS. + head *node + ref atomic.Int32 + arena *Arena + OnClose func() +} + +// IncrRef increases the refcount +func (s *Skiplist) IncrRef() { + s.ref.Add(1) +} + +// DecrRef decrements the refcount, deallocating the Skiplist when done using it +func (s *Skiplist) DecrRef() { + newRef := s.ref.Add(-1) + if newRef > 0 { + return + } + if s.OnClose != nil { + s.OnClose() + } + + // Indicate we are closed. Good for testing. Also, lets GC reclaim memory. Race condition + // here would suggest we are accessing skiplist when we are supposed to have no reference! + s.arena = nil + // Since the head references the arena's buf, as long as the head is kept around + // GC can't release the buf. + s.head = nil +} + +func newNode(arena *Arena, key []byte, v y.ValueStruct, height int) *node { + // The base level is already allocated in the node struct. + offset := arena.putNode(height) + node := arena.getNode(offset) + node.keyOffset = arena.putKey(key) + node.keySize = uint16(len(key)) + node.height = uint16(height) + node.value.Store(encodeValue(arena.putVal(v), v.EncodedSize())) + return node +} + +func encodeValue(valOffset uint32, valSize uint32) uint64 { + return uint64(valSize)<<32 | uint64(valOffset) +} + +func decodeValue(value uint64) (valOffset uint32, valSize uint32) { + valOffset = uint32(value) + valSize = uint32(value >> 32) + return +} + +// NewSkiplist makes a new empty skiplist, with a given arena size +func NewSkiplist(arenaSize int64) *Skiplist { + arena := newArena(arenaSize) + head := newNode(arena, nil, y.ValueStruct{}, maxHeight) + s := &Skiplist{head: head, arena: arena} + s.height.Store(1) + s.ref.Store(1) + return s +} + +func (s *node) getValueOffset() (uint32, uint32) { + value := s.value.Load() + return decodeValue(value) +} + +func (s *node) key(arena *Arena) []byte { + return arena.getKey(s.keyOffset, s.keySize) +} + +func (s *node) setValue(arena *Arena, v y.ValueStruct) { + valOffset := arena.putVal(v) + value := encodeValue(valOffset, v.EncodedSize()) + s.value.Store(value) +} + +func (s *node) getNextOffset(h int) uint32 { + return s.tower[h].Load() +} + +func (s *node) casNextOffset(h int, old, val uint32) bool { + return s.tower[h].CompareAndSwap(old, val) +} + +// Returns true if key is strictly > n.key. +// If n is nil, this is an "end" marker and we return false. +//func (s *Skiplist) keyIsAfterNode(key []byte, n *node) bool { +// y.AssertTrue(n != s.head) +// return n != nil && y.CompareKeys(key, n.key) > 0 +//} + +func (s *Skiplist) randomHeight() int { + h := 1 + for h < maxHeight && z.FastRand() <= heightIncrease { + h++ + } + return h +} + +func (s *Skiplist) getNext(nd *node, height int) *node { + return s.arena.getNode(nd.getNextOffset(height)) +} + +// findNear finds the node near to key. +// If less=true, it finds rightmost node such that node.key < key (if allowEqual=false) or +// node.key <= key (if allowEqual=true). +// If less=false, it finds leftmost node such that node.key > key (if allowEqual=false) or +// node.key >= key (if allowEqual=true). +// Returns the node found. The bool returned is true if the node has key equal to given key. +func (s *Skiplist) findNear(key []byte, less bool, allowEqual bool) (*node, bool) { + x := s.head + level := int(s.getHeight() - 1) + for { + // Assume x.key < key. + next := s.getNext(x, level) + if next == nil { + // x.key < key < END OF LIST + if level > 0 { + // Can descend further to iterate closer to the end. + level-- + continue + } + // Level=0. Cannot descend further. Let's return something that makes sense. + if !less { + return nil, false + } + // Try to return x. Make sure it is not a head node. + if x == s.head { + return nil, false + } + return x, false + } + + nextKey := next.key(s.arena) + cmp := y.CompareKeys(key, nextKey) + if cmp > 0 { + // x.key < next.key < key. We can continue to move right. + x = next + continue + } + if cmp == 0 { + // x.key < key == next.key. + if allowEqual { + return next, true + } + if !less { + // We want >, so go to base level to grab the next bigger note. + return s.getNext(next, 0), false + } + // We want <. If not base level, we should go closer in the next level. + if level > 0 { + level-- + continue + } + // On base level. Return x. + if x == s.head { + return nil, false + } + return x, false + } + // cmp < 0. In other words, x.key < key < next. + if level > 0 { + level-- + continue + } + // At base level. Need to return something. + if !less { + return next, false + } + // Try to return x. Make sure it is not a head node. + if x == s.head { + return nil, false + } + return x, false + } +} + +// findSpliceForLevel returns (outBefore, outAfter) with outBefore.key <= key <= outAfter.key. +// The input "before" tells us where to start looking. +// If we found a node with the same key, then we return outBefore = outAfter. +// Otherwise, outBefore.key < key < outAfter.key. +func (s *Skiplist) findSpliceForLevel(key []byte, before *node, level int) (*node, *node) { + for { + // Assume before.key < key. + next := s.getNext(before, level) + if next == nil { + return before, next + } + nextKey := next.key(s.arena) + cmp := y.CompareKeys(key, nextKey) + if cmp == 0 { + // Equality case. + return next, next + } + if cmp < 0 { + // before.key < key < next.key. We are done for this level. + return before, next + } + before = next // Keep moving right on this level. + } +} + +func (s *Skiplist) getHeight() int32 { + return s.height.Load() +} + +// Put inserts the key-value pair. +func (s *Skiplist) Put(key []byte, v y.ValueStruct) { + // Since we allow overwrite, we may not need to create a new node. We might not even need to + // increase the height. Let's defer these actions. + + listHeight := s.getHeight() + var prev [maxHeight + 1]*node + var next [maxHeight + 1]*node + prev[listHeight] = s.head + next[listHeight] = nil + for i := int(listHeight) - 1; i >= 0; i-- { + // Use higher level to speed up for current level. + prev[i], next[i] = s.findSpliceForLevel(key, prev[i+1], i) + if prev[i] == next[i] { + prev[i].setValue(s.arena, v) + return + } + } + + // We do need to create a new node. + height := s.randomHeight() + x := newNode(s.arena, key, v, height) + + // Try to increase s.height via CAS. + listHeight = s.getHeight() + for height > int(listHeight) { + if s.height.CompareAndSwap(listHeight, int32(height)) { + // Successfully increased skiplist.height. + break + } + listHeight = s.getHeight() + } + + // We always insert from the base level and up. After you add a node in base level, we cannot + // create a node in the level above because it would have discovered the node in the base level. + for i := 0; i < height; i++ { + for { + if prev[i] == nil { + y.AssertTrue(i > 1) // This cannot happen in base level. + // We haven't computed prev, next for this level because height exceeds old listHeight. + // For these levels, we expect the lists to be sparse, so we can just search from head. + prev[i], next[i] = s.findSpliceForLevel(key, s.head, i) + // Someone adds the exact same key before we are able to do so. This can only happen on + // the base level. But we know we are not on the base level. + y.AssertTrue(prev[i] != next[i]) + } + nextOffset := s.arena.getNodeOffset(next[i]) + x.tower[i].Store(nextOffset) + if prev[i].casNextOffset(i, nextOffset, s.arena.getNodeOffset(x)) { + // Managed to insert x between prev[i] and next[i]. Go to the next level. + break + } + // CAS failed. We need to recompute prev and next. + // It is unlikely to be helpful to try to use a different level as we redo the search, + // because it is unlikely that lots of nodes are inserted between prev[i] and next[i]. + prev[i], next[i] = s.findSpliceForLevel(key, prev[i], i) + if prev[i] == next[i] { + y.AssertTruef(i == 0, "Equality can happen only on base level: %d", i) + prev[i].setValue(s.arena, v) + return + } + } + } +} + +// Empty returns if the Skiplist is empty. +func (s *Skiplist) Empty() bool { + return s.findLast() == nil +} + +// findLast returns the last element. If head (empty list), we return nil. All the find functions +// will NEVER return the head nodes. +func (s *Skiplist) findLast() *node { + n := s.head + level := int(s.getHeight()) - 1 + for { + next := s.getNext(n, level) + if next != nil { + n = next + continue + } + if level == 0 { + if n == s.head { + return nil + } + return n + } + level-- + } +} + +// Get gets the value associated with the key. It returns a valid value if it finds equal or earlier +// version of the same key. +func (s *Skiplist) Get(key []byte) y.ValueStruct { + n, _ := s.findNear(key, false, true) // findGreaterOrEqual. + if n == nil { + return y.ValueStruct{} + } + + nextKey := s.arena.getKey(n.keyOffset, n.keySize) + if !y.SameKey(key, nextKey) { + return y.ValueStruct{} + } + + valOffset, valSize := n.getValueOffset() + vs := s.arena.getVal(valOffset, valSize) + vs.Version = y.ParseTs(nextKey) + return vs +} + +// NewIterator returns a skiplist iterator. You have to Close() the iterator. +func (s *Skiplist) NewIterator() *Iterator { + s.IncrRef() + return &Iterator{list: s} +} + +// MemSize returns the size of the Skiplist in terms of how much memory is used within its internal +// arena. +func (s *Skiplist) MemSize() int64 { return s.arena.size() } + +// Iterator is an iterator over skiplist object. For new objects, you just +// need to initialize Iterator.list. +type Iterator struct { + list *Skiplist + n *node +} + +// Close frees the resources held by the iterator +func (s *Iterator) Close() error { + s.list.DecrRef() + return nil +} + +// Valid returns true iff the iterator is positioned at a valid node. +func (s *Iterator) Valid() bool { return s.n != nil } + +// Key returns the key at the current position. +func (s *Iterator) Key() []byte { + return s.list.arena.getKey(s.n.keyOffset, s.n.keySize) +} + +// Value returns value. +func (s *Iterator) Value() y.ValueStruct { + valOffset, valSize := s.n.getValueOffset() + return s.list.arena.getVal(valOffset, valSize) +} + +// ValueUint64 returns the uint64 value of the current node. +func (s *Iterator) ValueUint64() uint64 { + return s.n.value.Load() +} + +// Next advances to the next position. +func (s *Iterator) Next() { + y.AssertTrue(s.Valid()) + s.n = s.list.getNext(s.n, 0) +} + +// Prev advances to the previous position. +func (s *Iterator) Prev() { + y.AssertTrue(s.Valid()) + s.n, _ = s.list.findNear(s.Key(), true, false) // find <. No equality allowed. +} + +// Seek advances to the first entry with a key >= target. +func (s *Iterator) Seek(target []byte) { + s.n, _ = s.list.findNear(target, false, true) // find >=. +} + +// SeekForPrev finds an entry with key <= target. +func (s *Iterator) SeekForPrev(target []byte) { + s.n, _ = s.list.findNear(target, true, true) // find <=. +} + +// SeekToFirst seeks position at the first entry in list. +// Final state of iterator is Valid() iff list is not empty. +func (s *Iterator) SeekToFirst() { + s.n = s.list.getNext(s.list.head, 0) +} + +// SeekToLast seeks position at the last entry in list. +// Final state of iterator is Valid() iff list is not empty. +func (s *Iterator) SeekToLast() { + s.n = s.list.findLast() +} + +// UniIterator is a unidirectional memtable iterator. It is a thin wrapper around +// Iterator. We like to keep Iterator as before, because it is more powerful and +// we might support bidirectional iterators in the future. +type UniIterator struct { + iter *Iterator + reversed bool +} + +// NewUniIterator returns a UniIterator. +func (s *Skiplist) NewUniIterator(reversed bool) *UniIterator { + return &UniIterator{ + iter: s.NewIterator(), + reversed: reversed, + } +} + +// Next implements y.Interface +func (s *UniIterator) Next() { + if !s.reversed { + s.iter.Next() + } else { + s.iter.Prev() + } +} + +// Rewind implements y.Interface +func (s *UniIterator) Rewind() { + if !s.reversed { + s.iter.SeekToFirst() + } else { + s.iter.SeekToLast() + } +} + +// Seek implements y.Interface +func (s *UniIterator) Seek(key []byte) { + if !s.reversed { + s.iter.Seek(key) + } else { + s.iter.SeekForPrev(key) + } +} + +// Key implements y.Interface +func (s *UniIterator) Key() []byte { return s.iter.Key() } + +// Value implements y.Interface +func (s *UniIterator) Value() y.ValueStruct { return s.iter.Value() } + +// Valid implements y.Interface +func (s *UniIterator) Valid() bool { return s.iter.Valid() } + +// Close implements y.Interface (and frees up the iter's resources) +func (s *UniIterator) Close() error { return s.iter.Close() } diff --git a/vendor/github.com/dgraph-io/badger/v4/stream.go b/vendor/github.com/dgraph-io/badger/v4/stream.go new file mode 100644 index 00000000000..21c5e992604 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/stream.go @@ -0,0 +1,492 @@ +/* + * Copyright 2018 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "context" + "sort" + "sync" + "sync/atomic" + "time" + + humanize "github.com/dustin/go-humanize" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +const batchSize = 16 << 20 // 16 MB + +// maxStreamSize is the maximum allowed size of a stream batch. This is a soft limit +// as a single list that is still over the limit will have to be sent as is since it +// cannot be split further. This limit prevents the framework from creating batches +// so big that sending them causes issues (e.g running into the max size gRPC limit). +var maxStreamSize = uint64(100 << 20) // 100MB + +// Stream provides a framework to concurrently iterate over a snapshot of Badger, pick up +// key-values, batch them up and call Send. Stream does concurrent iteration over many smaller key +// ranges. It does NOT send keys in lexicographical sorted order. To get keys in sorted +// order, use Iterator. +type Stream struct { + // Prefix to only iterate over certain range of keys. If set to nil (default), Stream would + // iterate over the entire DB. + Prefix []byte + + // Number of goroutines to use for iterating over key ranges. Defaults to 8. + NumGo int + + // Badger would produce log entries in Infof to indicate the progress of Stream. LogPrefix can + // be used to help differentiate them from other activities. Default is "Badger.Stream". + LogPrefix string + + // ChooseKey is invoked each time a new key is encountered. Note that this is not called + // on every version of the value, only the first encountered version (i.e. the highest version + // of the value a key has). ChooseKey can be left nil to select all keys. + // + // Note: Calls to ChooseKey are concurrent. + ChooseKey func(item *Item) bool + + // KeyToList, similar to ChooseKey, is only invoked on the highest version of the value. It + // is upto the caller to iterate over the versions and generate zero, one or more KVs. It + // is expected that the user would advance the iterator to go through the versions of the + // values. However, the user MUST immediately return from this function on the first encounter + // with a mismatching key. See example usage in ToList function. Can be left nil to use ToList + // function by default. + // + // KeyToList has access to z.Allocator accessible via stream.Allocator(itr.ThreadId). This + // allocator can be used to allocate KVs, to decrease the memory pressure on Go GC. Stream + // framework takes care of releasing those resources after calling Send. AllocRef does + // NOT need to be set in the returned KVList, as Stream framework would ignore that field, + // instead using the allocator assigned to that thread id. + // + // Note: Calls to KeyToList are concurrent. + KeyToList func(key []byte, itr *Iterator) (*pb.KVList, error) + + // This is the method where Stream sends the final output. All calls to Send are done by a + // single goroutine, i.e. logic within Send method can expect single threaded execution. + Send func(buf *z.Buffer) error + + // Read data above the sinceTs. All keys with version =< sinceTs will be ignored. + SinceTs uint64 + readTs uint64 + db *DB + rangeCh chan keyRange + kvChan chan *z.Buffer + nextStreamId atomic.Uint32 + doneMarkers bool + scanned atomic.Uint64 // used to estimate the ETA for data scan. + numProducers atomic.Int32 +} + +// SendDoneMarkers when true would send out done markers on the stream. False by default. +func (st *Stream) SendDoneMarkers(done bool) { + st.doneMarkers = done +} + +// ToList is a default implementation of KeyToList. It picks up all valid versions of the key, +// skipping over deleted or expired keys. +func (st *Stream) ToList(key []byte, itr *Iterator) (*pb.KVList, error) { + a := itr.Alloc + ka := a.Copy(key) + + list := &pb.KVList{} + for ; itr.Valid(); itr.Next() { + item := itr.Item() + if item.IsDeletedOrExpired() { + break + } + if !bytes.Equal(key, item.Key()) { + // Break out on the first encounter with another key. + break + } + + kv := y.NewKV(a) + kv.Key = ka + + if err := item.Value(func(val []byte) error { + kv.Value = a.Copy(val) + return nil + + }); err != nil { + return nil, err + } + kv.Version = item.Version() + kv.ExpiresAt = item.ExpiresAt() + kv.UserMeta = a.Copy([]byte{item.UserMeta()}) + + list.Kv = append(list.Kv, kv) + if st.db.opt.NumVersionsToKeep == 1 { + break + } + + if item.DiscardEarlierVersions() { + break + } + } + return list, nil +} + +// keyRange is [start, end), including start, excluding end. Do ensure that the start, +// end byte slices are owned by keyRange struct. +func (st *Stream) produceRanges(ctx context.Context) { + ranges := st.db.Ranges(st.Prefix, 16) + y.AssertTrue(len(ranges) > 0) + y.AssertTrue(ranges[0].left == nil) + y.AssertTrue(ranges[len(ranges)-1].right == nil) + st.db.opt.Infof("Number of ranges found: %d\n", len(ranges)) + + // Sort in descending order of size. + sort.Slice(ranges, func(i, j int) bool { + return ranges[i].size > ranges[j].size + }) + for i, r := range ranges { + st.rangeCh <- *r + st.db.opt.Infof("Sent range %d for iteration: [%x, %x) of size: %s\n", + i, r.left, r.right, humanize.IBytes(uint64(r.size))) + } + close(st.rangeCh) +} + +// produceKVs picks up ranges from rangeCh, generates KV lists and sends them to kvChan. +func (st *Stream) produceKVs(ctx context.Context, threadId int) error { + st.numProducers.Add(1) + defer st.numProducers.Add(-1) + + var txn *Txn + if st.readTs > 0 { + txn = st.db.NewTransactionAt(st.readTs, false) + } else { + txn = st.db.NewTransaction(false) + } + defer txn.Discard() + + // produceKVs is running iterate serially. So, we can define the outList here. + outList := z.NewBuffer(2*batchSize, "Stream.ProduceKVs") + defer func() { + // The outList variable changes. So, we need to evaluate the variable in the defer. DO NOT + // call `defer outList.Release()`. + _ = outList.Release() + }() + + iterate := func(kr keyRange) error { + iterOpts := DefaultIteratorOptions + iterOpts.AllVersions = true + iterOpts.Prefix = st.Prefix + iterOpts.PrefetchValues = false + iterOpts.SinceTs = st.SinceTs + itr := txn.NewIterator(iterOpts) + itr.ThreadId = threadId + defer itr.Close() + + itr.Alloc = z.NewAllocator(1<<20, "Stream.Iterate") + defer itr.Alloc.Release() + + // This unique stream id is used to identify all the keys from this iteration. + streamId := st.nextStreamId.Add(1) + var scanned int + + sendIt := func() error { + select { + case st.kvChan <- outList: + outList = z.NewBuffer(2*batchSize, "Stream.ProduceKVs") + st.scanned.Add(uint64(itr.scanned - scanned)) + scanned = itr.scanned + case <-ctx.Done(): + return ctx.Err() + } + return nil + } + + var prevKey []byte + for itr.Seek(kr.left); itr.Valid(); { + // it.Valid would only return true for keys with the provided Prefix in iterOpts. + item := itr.Item() + if bytes.Equal(item.Key(), prevKey) { + itr.Next() + continue + } + prevKey = append(prevKey[:0], item.Key()...) + + // Check if we reached the end of the key range. + if len(kr.right) > 0 && bytes.Compare(item.Key(), kr.right) >= 0 { + break + } + + // Check if we should pick this key. + if st.ChooseKey != nil && !st.ChooseKey(item) { + continue + } + + // Now convert to key value. + itr.Alloc.Reset() + list, err := st.KeyToList(item.KeyCopy(nil), itr) + if err != nil { + st.db.opt.Warningf("While reading key: %x, got error: %v", item.Key(), err) + continue + } + if list == nil || len(list.Kv) == 0 { + continue + } + for _, kv := range list.Kv { + kv.StreamId = streamId + KVToBuffer(kv, outList) + if outList.LenNoPadding() < batchSize { + continue + } + if err := sendIt(); err != nil { + return err + } + } + } + // Mark the stream as done. + if st.doneMarkers { + kv := &pb.KV{ + StreamId: streamId, + StreamDone: true, + } + KVToBuffer(kv, outList) + } + return sendIt() + } + + for { + select { + case kr, ok := <-st.rangeCh: + if !ok { + // Done with the keys. + return nil + } + if err := iterate(kr); err != nil { + return err + } + case <-ctx.Done(): + return ctx.Err() + } + } +} + +func (st *Stream) streamKVs(ctx context.Context) error { + onDiskSize, uncompressedSize := st.db.EstimateSize(st.Prefix) + // Manish has seen uncompressed size to be in 20% error margin. + uncompressedSize = uint64(float64(uncompressedSize) * 1.2) + st.db.opt.Infof("%s Streaming about %s of uncompressed data (%s on disk)\n", + st.LogPrefix, humanize.IBytes(uncompressedSize), humanize.IBytes(onDiskSize)) + + tickerDur := 5 * time.Second + var bytesSent uint64 + t := time.NewTicker(tickerDur) + defer t.Stop() + now := time.Now() + + sendBatch := func(batch *z.Buffer) error { + defer func() { _ = batch.Release() }() + sz := uint64(batch.LenNoPadding()) + if sz == 0 { + return nil + } + bytesSent += sz + // st.db.opt.Infof("%s Sending batch of size: %s.\n", st.LogPrefix, humanize.IBytes(sz)) + if err := st.Send(batch); err != nil { + st.db.opt.Warningf("Error while sending: %v\n", err) + return err + } + return nil + } + + slurp := func(batch *z.Buffer) error { + loop: + for { + // Send the batch immediately if it already exceeds the maximum allowed size. + // If the size of the batch exceeds maxStreamSize, break from the loop to + // avoid creating a batch that is so big that certain limits are reached. + if batch.LenNoPadding() > int(maxStreamSize) { + break loop + } + select { + case kvs, ok := <-st.kvChan: + if !ok { + break loop + } + y.AssertTrue(kvs != nil) + y.Check2(batch.Write(kvs.Bytes())) + y.Check(kvs.Release()) + + default: + break loop + } + } + return sendBatch(batch) + } // end of slurp. + + writeRate := y.NewRateMonitor(20) + scanRate := y.NewRateMonitor(20) +outer: + for { + var batch *z.Buffer + select { + case <-ctx.Done(): + return ctx.Err() + + case <-t.C: + // Instead of calculating speed over the entire lifetime, we average the speed over + // ticker duration. + writeRate.Capture(bytesSent) + scanned := st.scanned.Load() + scanRate.Capture(scanned) + numProducers := st.numProducers.Load() + + st.db.opt.Infof("%s [%s] Scan (%d): ~%s/%s at %s/sec. Sent: %s at %s/sec."+ + " jemalloc: %s\n", + st.LogPrefix, y.FixedDuration(time.Since(now)), numProducers, + y.IBytesToString(scanned, 1), humanize.IBytes(uncompressedSize), + humanize.IBytes(scanRate.Rate()), + y.IBytesToString(bytesSent, 1), humanize.IBytes(writeRate.Rate()), + humanize.IBytes(uint64(z.NumAllocBytes()))) + + case kvs, ok := <-st.kvChan: + if !ok { + break outer + } + y.AssertTrue(kvs != nil) + batch = kvs + + // Otherwise, slurp more keys into this batch. + if err := slurp(batch); err != nil { + return err + } + } + } + + st.db.opt.Infof("%s Sent data of size %s\n", st.LogPrefix, humanize.IBytes(bytesSent)) + return nil +} + +// Orchestrate runs Stream. It picks up ranges from the SSTables, then runs NumGo number of +// goroutines to iterate over these ranges and batch up KVs in lists. It concurrently runs a single +// goroutine to pick these lists, batch them up further and send to Output.Send. Orchestrate also +// spits logs out to Infof, using provided LogPrefix. Note that all calls to Output.Send +// are serial. In case any of these steps encounter an error, Orchestrate would stop execution and +// return that error. Orchestrate can be called multiple times, but in serial order. +func (st *Stream) Orchestrate(ctx context.Context) error { + ctx, cancel := context.WithCancel(ctx) + defer cancel() + st.rangeCh = make(chan keyRange, 3) // Contains keys for posting lists. + + // kvChan should only have a small capacity to ensure that we don't buffer up too much data if + // sending is slow. Page size is set to 4MB, which is used to lazily cap the size of each + // KVList. To get 128MB buffer, we can set the channel size to 32. + st.kvChan = make(chan *z.Buffer, 32) + + if st.KeyToList == nil { + st.KeyToList = st.ToList + } + + // Picks up ranges from Badger, and sends them to rangeCh. + go st.produceRanges(ctx) + + errCh := make(chan error, st.NumGo) // Stores error by consumeKeys. + var wg sync.WaitGroup + for i := 0; i < st.NumGo; i++ { + wg.Add(1) + + go func(threadId int) { + defer wg.Done() + // Picks up ranges from rangeCh, generates KV lists, and sends them to kvChan. + if err := st.produceKVs(ctx, threadId); err != nil { + select { + case errCh <- err: + default: + } + } + }(i) + } + + // Pick up key-values from kvChan and send to stream. + kvErr := make(chan error, 1) + go func() { + // Picks up KV lists from kvChan, and sends them to Output. + err := st.streamKVs(ctx) + if err != nil { + cancel() // Stop all the go routines. + } + kvErr <- err + }() + wg.Wait() // Wait for produceKVs to be over. + close(st.kvChan) // Now we can close kvChan. + defer func() { + // If due to some error, we have buffers left in kvChan, we should release them. + for buf := range st.kvChan { + _ = buf.Release() + } + }() + + select { + case err := <-errCh: // Check error from produceKVs. + return err + default: + } + + // Wait for key streaming to be over. + err := <-kvErr + return err +} + +func (db *DB) newStream() *Stream { + return &Stream{ + db: db, + NumGo: db.opt.NumGoroutines, + LogPrefix: "Badger.Stream", + } +} + +// NewStream creates a new Stream. +func (db *DB) NewStream() *Stream { + if db.opt.managedTxns { + panic("This API can not be called in managed mode.") + } + return db.newStream() +} + +// NewStreamAt creates a new Stream at a particular timestamp. Should only be used with managed DB. +func (db *DB) NewStreamAt(readTs uint64) *Stream { + if !db.opt.managedTxns { + panic("This API can only be called in managed mode.") + } + stream := db.newStream() + stream.readTs = readTs + return stream +} + +func BufferToKVList(buf *z.Buffer) (*pb.KVList, error) { + var list pb.KVList + err := buf.SliceIterate(func(s []byte) error { + kv := new(pb.KV) + if err := kv.Unmarshal(s); err != nil { + return err + } + list.Kv = append(list.Kv, kv) + return nil + }) + return &list, err +} + +func KVToBuffer(kv *pb.KV, buf *z.Buffer) { + out := buf.SliceAllocate(kv.Size()) + y.Check2(kv.MarshalToSizedBuffer(out)) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/stream_writer.go b/vendor/github.com/dgraph-io/badger/v4/stream_writer.go new file mode 100644 index 00000000000..a160e55cb93 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/stream_writer.go @@ -0,0 +1,531 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "encoding/hex" + "fmt" + "sync" + + humanize "github.com/dustin/go-humanize" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// StreamWriter is used to write data coming from multiple streams. The streams must not have any +// overlapping key ranges. Within each stream, the keys must be sorted. Badger Stream framework is +// capable of generating such an output. So, this StreamWriter can be used at the other end to build +// BadgerDB at a much faster pace by writing SSTables (and value logs) directly to LSM tree levels +// without causing any compactions at all. This is way faster than using batched writer or using +// transactions, but only applicable in situations where the keys are pre-sorted and the DB is being +// bootstrapped. Existing data would get deleted when using this writer. So, this is only useful +// when restoring from backup or replicating DB across servers. +// +// StreamWriter should not be called on in-use DB instances. It is designed only to bootstrap new +// DBs. +type StreamWriter struct { + writeLock sync.Mutex + db *DB + done func() + throttle *y.Throttle + maxVersion uint64 + writers map[uint32]*sortedWriter + prevLevel int +} + +// NewStreamWriter creates a StreamWriter. Right after creating StreamWriter, Prepare must be +// called. The memory usage of a StreamWriter is directly proportional to the number of streams +// possible. So, efforts must be made to keep the number of streams low. Stream framework would +// typically use 16 goroutines and hence create 16 streams. +func (db *DB) NewStreamWriter() *StreamWriter { + return &StreamWriter{ + db: db, + // throttle shouldn't make much difference. Memory consumption is based on the number of + // concurrent streams being processed. + throttle: y.NewThrottle(16), + writers: make(map[uint32]*sortedWriter), + } +} + +// Prepare should be called before writing any entry to StreamWriter. It deletes all data present in +// existing DB, stops compactions and any writes being done by other means. Be very careful when +// calling Prepare, because it could result in permanent data loss. Not calling Prepare would result +// in a corrupt Badger instance. Use PrepareIncremental to do incremental stream write. +func (sw *StreamWriter) Prepare() error { + sw.writeLock.Lock() + defer sw.writeLock.Unlock() + + done, err := sw.db.dropAll() + // Ensure that done() is never called more than once. + var once sync.Once + sw.done = func() { once.Do(done) } + return err +} + +// PrepareIncremental should be called before writing any entry to StreamWriter incrementally. +// In incremental stream write, the tables are written at one level above the current base level. +func (sw *StreamWriter) PrepareIncremental() error { + sw.writeLock.Lock() + defer sw.writeLock.Unlock() + + // Ensure that done() is never called more than once. + var once sync.Once + + // prepareToDrop will stop all the incoming writes and process any pending flush tasks. + // Before we start writing, we'll stop the compactions because no one else should be writing to + // the same level as the stream writer is writing to. + f, err := sw.db.prepareToDrop() + if err != nil { + sw.done = func() { once.Do(f) } + return err + } + sw.db.stopCompactions() + done := func() { + sw.db.startCompactions() + f() + } + sw.done = func() { once.Do(done) } + + mts, decr := sw.db.getMemTables() + defer decr() + for _, m := range mts { + if !m.sl.Empty() { + return fmt.Errorf("Unable to do incremental writes because MemTable has data") + } + } + + isEmptyDB := true + for _, level := range sw.db.Levels() { + if level.NumTables > 0 { + sw.prevLevel = level.Level + isEmptyDB = false + break + } + } + if isEmptyDB { + // If DB is empty, we should allow doing incremental stream write. + return nil + } + if sw.prevLevel == 0 { + // It seems that data is present in all levels from Lmax to L0. If we call flatten + // on the tree, all the data will go to Lmax. All the levels above will be empty + // after flatten call. Now, we should be able to use incremental stream writer again. + if err := sw.db.Flatten(3); err != nil { + return errors.Wrapf(err, "error during flatten in StreamWriter") + } + sw.prevLevel = len(sw.db.Levels()) - 1 + } + return nil +} + +// Write writes KVList to DB. Each KV within the list contains the stream id which StreamWriter +// would use to demux the writes. Write is thread safe and can be called concurrently by multiple +// goroutines. +func (sw *StreamWriter) Write(buf *z.Buffer) error { + if buf.LenNoPadding() == 0 { + return nil + } + + // closedStreams keeps track of all streams which are going to be marked as done. We are + // keeping track of all streams so that we can close them at the end, after inserting all + // the valid kvs. + closedStreams := make(map[uint32]struct{}) + streamReqs := make(map[uint32]*request) + + err := buf.SliceIterate(func(s []byte) error { + var kv pb.KV + if err := kv.Unmarshal(s); err != nil { + return err + } + if kv.StreamDone { + closedStreams[kv.StreamId] = struct{}{} + return nil + } + + // Panic if some kv comes after stream has been marked as closed. + if _, ok := closedStreams[kv.StreamId]; ok { + panic(fmt.Sprintf("write performed on closed stream: %d", kv.StreamId)) + } + + sw.writeLock.Lock() + if sw.maxVersion < kv.Version { + sw.maxVersion = kv.Version + } + if sw.prevLevel == 0 { + // If prevLevel is 0, that means that we have not written anything yet. + // So, we can write to the maxLevel. newWriter writes to prevLevel - 1, + // so we can set prevLevel to len(levels). + sw.prevLevel = len(sw.db.lc.levels) + } + sw.writeLock.Unlock() + + var meta, userMeta byte + if len(kv.Meta) > 0 { + meta = kv.Meta[0] + } + if len(kv.UserMeta) > 0 { + userMeta = kv.UserMeta[0] + } + e := &Entry{ + Key: y.KeyWithTs(kv.Key, kv.Version), + Value: y.Copy(kv.Value), + UserMeta: userMeta, + ExpiresAt: kv.ExpiresAt, + meta: meta, + } + // If the value can be collocated with the key in LSM tree, we can skip + // writing the value to value log. + req := streamReqs[kv.StreamId] + if req == nil { + req = &request{} + streamReqs[kv.StreamId] = req + } + req.Entries = append(req.Entries, e) + return nil + }) + if err != nil { + return err + } + + all := make([]*request, 0, len(streamReqs)) + for _, req := range streamReqs { + all = append(all, req) + } + + sw.writeLock.Lock() + defer sw.writeLock.Unlock() + + // We are writing all requests to vlog even if some request belongs to already closed stream. + // It is safe to do because we are panicking while writing to sorted writer, which will be nil + // for closed stream. At restart, stream writer will drop all the data in Prepare function. + if err := sw.db.vlog.write(all); err != nil { + return err + } + + for streamID, req := range streamReqs { + writer, ok := sw.writers[streamID] + if !ok { + var err error + writer, err = sw.newWriter(streamID) + if err != nil { + return y.Wrapf(err, "failed to create writer with ID %d", streamID) + } + sw.writers[streamID] = writer + } + + if writer == nil { + panic(fmt.Sprintf("write performed on closed stream: %d", streamID)) + } + + writer.reqCh <- req + } + + // Now we can close any streams if required. We will make writer for + // the closed streams as nil. + for streamId := range closedStreams { + writer, ok := sw.writers[streamId] + if !ok { + sw.db.opt.Warningf("Trying to close stream: %d, but no sorted "+ + "writer found for it", streamId) + continue + } + + writer.closer.SignalAndWait() + if err := writer.Done(); err != nil { + return err + } + + sw.writers[streamId] = nil + } + return nil +} + +// Flush is called once we are done writing all the entries. It syncs DB directories. It also +// updates Oracle with maxVersion found in all entries (if DB is not managed). +func (sw *StreamWriter) Flush() error { + sw.writeLock.Lock() + defer sw.writeLock.Unlock() + + defer sw.done() + + for _, writer := range sw.writers { + if writer != nil { + writer.closer.SignalAndWait() + } + } + + for _, writer := range sw.writers { + if writer == nil { + continue + } + if err := writer.Done(); err != nil { + return err + } + } + + if !sw.db.opt.managedTxns { + if sw.db.orc != nil { + sw.db.orc.Stop() + } + + if curMax := sw.db.orc.readTs(); curMax >= sw.maxVersion { + sw.maxVersion = curMax + } + + sw.db.orc = newOracle(sw.db.opt) + sw.db.orc.nextTxnTs = sw.maxVersion + sw.db.orc.txnMark.Done(sw.maxVersion) + sw.db.orc.readMark.Done(sw.maxVersion) + sw.db.orc.incrementNextTs() + } + + // Wait for all files to be written. + if err := sw.throttle.Finish(); err != nil { + return err + } + + // Sort tables at the end. + for _, l := range sw.db.lc.levels { + l.sortTables() + } + + // Now sync the directories, so all the files are registered. + if sw.db.opt.ValueDir != sw.db.opt.Dir { + if err := sw.db.syncDir(sw.db.opt.ValueDir); err != nil { + return err + } + } + if err := sw.db.syncDir(sw.db.opt.Dir); err != nil { + return err + } + return sw.db.lc.validate() +} + +// Cancel signals all goroutines to exit. Calling defer sw.Cancel() immediately after creating a new StreamWriter +// ensures that writes are unblocked even upon early return. Note that dropAll() is not called here, so any +// partially written data will not be erased until a new StreamWriter is initialized. +func (sw *StreamWriter) Cancel() { + sw.writeLock.Lock() + defer sw.writeLock.Unlock() + + for _, writer := range sw.writers { + if writer != nil { + writer.closer.Signal() + } + } + for _, writer := range sw.writers { + if writer != nil { + writer.closer.Wait() + } + } + + if err := sw.throttle.Finish(); err != nil { + sw.db.opt.Errorf("error in throttle.Finish: %+v", err) + } + + // Handle Cancel() being called before Prepare(). + if sw.done != nil { + sw.done() + } +} + +type sortedWriter struct { + db *DB + throttle *y.Throttle + opts table.Options + + builder *table.Builder + lastKey []byte + level int + streamID uint32 + reqCh chan *request + // Have separate closer for each writer, as it can be closed at any time. + closer *z.Closer +} + +func (sw *StreamWriter) newWriter(streamID uint32) (*sortedWriter, error) { + bopts := buildTableOptions(sw.db) + for i := 2; i < sw.db.opt.MaxLevels; i++ { + bopts.TableSize *= uint64(sw.db.opt.TableSizeMultiplier) + } + w := &sortedWriter{ + db: sw.db, + opts: bopts, + streamID: streamID, + throttle: sw.throttle, + builder: table.NewTableBuilder(bopts), + reqCh: make(chan *request, 3), + closer: z.NewCloser(1), + level: sw.prevLevel - 1, // Write at the level just above the one we were writing to. + } + + go w.handleRequests() + return w, nil +} + +func (w *sortedWriter) handleRequests() { + defer w.closer.Done() + + process := func(req *request) { + for i, e := range req.Entries { + // If badger is running in InMemory mode, len(req.Ptrs) == 0. + var vs y.ValueStruct + if e.skipVlogAndSetThreshold(w.db.valueThreshold()) { + vs = y.ValueStruct{ + Value: e.Value, + Meta: e.meta, + UserMeta: e.UserMeta, + ExpiresAt: e.ExpiresAt, + } + } else { + vptr := req.Ptrs[i] + vs = y.ValueStruct{ + Value: vptr.Encode(), + Meta: e.meta | bitValuePointer, + UserMeta: e.UserMeta, + ExpiresAt: e.ExpiresAt, + } + } + if err := w.Add(e.Key, vs); err != nil { + panic(err) + } + } + } + + for { + select { + case req := <-w.reqCh: + process(req) + case <-w.closer.HasBeenClosed(): + close(w.reqCh) + for req := range w.reqCh { + process(req) + } + return + } + } +} + +// Add adds key and vs to sortedWriter. +func (w *sortedWriter) Add(key []byte, vs y.ValueStruct) error { + if len(w.lastKey) > 0 && y.CompareKeys(key, w.lastKey) <= 0 { + return errors.Errorf("keys not in sorted order (last key: %s, key: %s)", + hex.Dump(w.lastKey), hex.Dump(key)) + } + + sameKey := y.SameKey(key, w.lastKey) + + // Same keys should go into the same SSTable. + if !sameKey && w.builder.ReachedCapacity() { + if err := w.send(false); err != nil { + return err + } + } + + w.lastKey = y.SafeCopy(w.lastKey, key) + var vp valuePointer + if vs.Meta&bitValuePointer > 0 { + vp.Decode(vs.Value) + } + + w.builder.Add(key, vs, vp.Len) + return nil +} + +func (w *sortedWriter) send(done bool) error { + if err := w.throttle.Do(); err != nil { + return err + } + go func(builder *table.Builder) { + err := w.createTable(builder) + w.throttle.Done(err) + }(w.builder) + // If done is true, this indicates we can close the writer. + // No need to allocate underlying TableBuilder now. + if done { + w.builder = nil + return nil + } + + w.builder = table.NewTableBuilder(w.opts) + return nil +} + +// Done is called once we are done writing all keys and valueStructs +// to sortedWriter. It completes writing current SST to disk. +func (w *sortedWriter) Done() error { + if w.builder.Empty() { + w.builder.Close() + // Assign builder as nil, so that underlying memory can be garbage collected. + w.builder = nil + return nil + } + + return w.send(true) +} + +func (w *sortedWriter) createTable(builder *table.Builder) error { + defer builder.Close() + if builder.Empty() { + builder.Finish() + return nil + } + + fileID := w.db.lc.reserveFileID() + var tbl *table.Table + if w.db.opt.InMemory { + data := builder.Finish() + var err error + if tbl, err = table.OpenInMemoryTable(data, fileID, builder.Opts()); err != nil { + return err + } + } else { + var err error + fname := table.NewFilename(fileID, w.db.opt.Dir) + if tbl, err = table.CreateTable(fname, builder); err != nil { + return err + } + } + lc := w.db.lc + + lhandler := lc.levels[w.level] + // Now that table can be opened successfully, let's add this to the MANIFEST. + change := &pb.ManifestChange{ + Id: tbl.ID(), + KeyId: tbl.KeyID(), + Op: pb.ManifestChange_CREATE, + Level: uint32(lhandler.level), + Compression: uint32(tbl.CompressionType()), + } + if err := w.db.manifest.addChanges([]*pb.ManifestChange{change}); err != nil { + return err + } + + // We are not calling lhandler.replaceTables() here, as it sorts tables on every addition. + // We can sort all tables only once during Flush() call. + lhandler.addTable(tbl) + + // Release the ref held by OpenTable. + _ = tbl.DecrRef() + w.db.opt.Infof("Table created: %d at level: %d for stream: %d. Size: %s\n", + fileID, lhandler.level, w.streamID, humanize.IBytes(uint64(tbl.Size()))) + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/structs.go b/vendor/github.com/dgraph-io/badger/v4/structs.go new file mode 100644 index 00000000000..7a06d4f13e1 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/structs.go @@ -0,0 +1,226 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "encoding/binary" + "fmt" + "time" + "unsafe" +) + +type valuePointer struct { + Fid uint32 + Len uint32 + Offset uint32 +} + +const vptrSize = unsafe.Sizeof(valuePointer{}) + +func (p valuePointer) Less(o valuePointer) bool { + if p.Fid != o.Fid { + return p.Fid < o.Fid + } + if p.Offset != o.Offset { + return p.Offset < o.Offset + } + return p.Len < o.Len +} + +func (p valuePointer) IsZero() bool { + return p.Fid == 0 && p.Offset == 0 && p.Len == 0 +} + +// Encode encodes Pointer into byte buffer. +func (p valuePointer) Encode() []byte { + b := make([]byte, vptrSize) + // Copy over the content from p to b. + *(*valuePointer)(unsafe.Pointer(&b[0])) = p + return b +} + +// Decode decodes the value pointer into the provided byte buffer. +func (p *valuePointer) Decode(b []byte) { + // Copy over data from b into p. Using *p=unsafe.pointer(...) leads to + // pointer alignment issues. See https://github.com/dgraph-io/badger/issues/1096 + // and comment https://github.com/dgraph-io/badger/pull/1097#pullrequestreview-307361714 + copy(((*[vptrSize]byte)(unsafe.Pointer(p))[:]), b[:vptrSize]) +} + +// header is used in value log as a header before Entry. +type header struct { + klen uint32 + vlen uint32 + expiresAt uint64 + meta byte + userMeta byte +} + +const ( + // Maximum possible size of the header. The maximum size of header struct will be 18 but the + // maximum size of varint encoded header will be 22. + maxHeaderSize = 22 +) + +// Encode encodes the header into []byte. The provided []byte should be atleast 5 bytes. The +// function will panic if out []byte isn't large enough to hold all the values. +// The encoded header looks like +// +------+----------+------------+--------------+-----------+ +// | Meta | UserMeta | Key Length | Value Length | ExpiresAt | +// +------+----------+------------+--------------+-----------+ +func (h header) Encode(out []byte) int { + out[0], out[1] = h.meta, h.userMeta + index := 2 + index += binary.PutUvarint(out[index:], uint64(h.klen)) + index += binary.PutUvarint(out[index:], uint64(h.vlen)) + index += binary.PutUvarint(out[index:], h.expiresAt) + return index +} + +// Decode decodes the given header from the provided byte slice. +// Returns the number of bytes read. +func (h *header) Decode(buf []byte) int { + h.meta, h.userMeta = buf[0], buf[1] + index := 2 + klen, count := binary.Uvarint(buf[index:]) + h.klen = uint32(klen) + index += count + vlen, count := binary.Uvarint(buf[index:]) + h.vlen = uint32(vlen) + index += count + h.expiresAt, count = binary.Uvarint(buf[index:]) + return index + count +} + +// DecodeFrom reads the header from the hashReader. +// Returns the number of bytes read. +func (h *header) DecodeFrom(reader *hashReader) (int, error) { + var err error + h.meta, err = reader.ReadByte() + if err != nil { + return 0, err + } + h.userMeta, err = reader.ReadByte() + if err != nil { + return 0, err + } + klen, err := binary.ReadUvarint(reader) + if err != nil { + return 0, err + } + h.klen = uint32(klen) + vlen, err := binary.ReadUvarint(reader) + if err != nil { + return 0, err + } + h.vlen = uint32(vlen) + h.expiresAt, err = binary.ReadUvarint(reader) + if err != nil { + return 0, err + } + return reader.bytesRead, nil +} + +// Entry provides Key, Value, UserMeta and ExpiresAt. This struct can be used by +// the user to set data. +type Entry struct { + Key []byte + Value []byte + ExpiresAt uint64 // time.Unix + version uint64 + offset uint32 // offset is an internal field. + UserMeta byte + meta byte + + // Fields maintained internally. + hlen int // Length of the header. + valThreshold int64 +} + +func (e *Entry) isZero() bool { + return len(e.Key) == 0 +} + +func (e *Entry) estimateSizeAndSetThreshold(threshold int64) int64 { + if e.valThreshold == 0 { + e.valThreshold = threshold + } + k := int64(len(e.Key)) + v := int64(len(e.Value)) + if v < e.valThreshold { + return k + v + 2 // Meta, UserMeta + } + return k + 12 + 2 // 12 for ValuePointer, 2 for metas. +} + +func (e *Entry) skipVlogAndSetThreshold(threshold int64) bool { + if e.valThreshold == 0 { + e.valThreshold = threshold + } + return int64(len(e.Value)) < e.valThreshold +} + +//nolint:unused +func (e Entry) print(prefix string) { + fmt.Printf("%s Key: %s Meta: %d UserMeta: %d Offset: %d len(val)=%d", + prefix, e.Key, e.meta, e.UserMeta, e.offset, len(e.Value)) +} + +// NewEntry creates a new entry with key and value passed in args. This newly created entry can be +// set in a transaction by calling txn.SetEntry(). All other properties of Entry can be set by +// calling WithMeta, WithDiscard, WithTTL methods on it. +// This function uses key and value reference, hence users must +// not modify key and value until the end of transaction. +func NewEntry(key, value []byte) *Entry { + return &Entry{ + Key: key, + Value: value, + } +} + +// WithMeta adds meta data to Entry e. This byte is stored alongside the key +// and can be used as an aid to interpret the value or store other contextual +// bits corresponding to the key-value pair of entry. +func (e *Entry) WithMeta(meta byte) *Entry { + e.UserMeta = meta + return e +} + +// WithDiscard adds a marker to Entry e. This means all the previous versions of the key (of the +// Entry) will be eligible for garbage collection. +// This method is only useful if you have set a higher limit for options.NumVersionsToKeep. The +// default setting is 1, in which case, this function doesn't add any more benefit. If however, you +// have a higher setting for NumVersionsToKeep (in Dgraph, we set it to infinity), you can use this +// method to indicate that all the older versions can be discarded and removed during compactions. +func (e *Entry) WithDiscard() *Entry { + e.meta = bitDiscardEarlierVersions + return e +} + +// WithTTL adds time to live duration to Entry e. Entry stored with a TTL would automatically expire +// after the time has elapsed, and will be eligible for garbage collection. +func (e *Entry) WithTTL(dur time.Duration) *Entry { + e.ExpiresAt = uint64(time.Now().Add(dur).Unix()) + return e +} + +// withMergeBit sets merge bit in entry's metadata. This +// function is called by MergeOperator's Add method. +func (e *Entry) withMergeBit() *Entry { + e.meta = bitMergeEntry + return e +} diff --git a/vendor/github.com/dgraph-io/badger/v4/table/README.md b/vendor/github.com/dgraph-io/badger/v4/table/README.md new file mode 100644 index 00000000000..19276079ef1 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/table/README.md @@ -0,0 +1,108 @@ +Size of table is 123,217,667 bytes for all benchmarks. + +# BenchmarkRead +``` +$ go test -bench ^BenchmarkRead$ -run ^$ -count 3 +goos: linux +goarch: amd64 +pkg: github.com/dgraph-io/badger/table +BenchmarkRead-16 10 154074944 ns/op +BenchmarkRead-16 10 154340411 ns/op +BenchmarkRead-16 10 151914489 ns/op +PASS +ok github.com/dgraph-io/badger/table 22.467s +``` + +Size of table is 123,217,667 bytes, which is ~118MB. + +The rate is ~762MB/s using LoadToRAM (when table is in RAM). + +To read a 64MB table, this would take ~0.084s, which is negligible. + +# BenchmarkReadAndBuild +```go +$ go test -bench BenchmarkReadAndBuild -run ^$ -count 3 +goos: linux +goarch: amd64 +pkg: github.com/dgraph-io/badger/table +BenchmarkReadAndBuild-16 1 1026755231 ns/op +BenchmarkReadAndBuild-16 1 1009543316 ns/op +BenchmarkReadAndBuild-16 1 1039920546 ns/op +PASS +ok github.com/dgraph-io/badger/table 12.081s +``` + +The rate is ~123MB/s. To build a 64MB table, this would take ~0.56s. Note that this +does NOT include the flushing of the table to disk. All we are doing above is +reading one table (which is in RAM) and write one table in memory. + +The table building takes 0.56-0.084s ~ 0.4823s. + +# BenchmarkReadMerged +Below, we merge 5 tables. The total size remains unchanged at ~122M. + +```go +$ go test -bench ReadMerged -run ^$ -count 3 +goos: linux +goarch: amd64 +pkg: github.com/dgraph-io/badger/table +BenchmarkReadMerged-16 2 977588975 ns/op +BenchmarkReadMerged-16 2 982140738 ns/op +BenchmarkReadMerged-16 2 962046017 ns/op +PASS +ok github.com/dgraph-io/badger/table 27.433s +``` + +The rate is ~120MB/s. To read a 64MB table using merge iterator, this would take ~0.53s. + +# BenchmarkRandomRead + +```go +go test -bench BenchmarkRandomRead$ -run ^$ -count 3 +goos: linux +goarch: amd64 +pkg: github.com/dgraph-io/badger/table +BenchmarkRandomRead-16 500000 2645 ns/op +BenchmarkRandomRead-16 500000 2648 ns/op +BenchmarkRandomRead-16 500000 2614 ns/op +PASS +ok github.com/dgraph-io/badger/table 50.850s +``` +For random read benchmarking, we are randomly reading a key and verifying its value. + +# DB Open benchmark +1. Create badger DB with 2 billion key-value pairs (about 380GB of data) +``` +badger fill -m 2000 --dir="/tmp/data" --sorted +``` +2. Clear buffers and swap memory +``` +free -mh && sync && echo 3 | sudo tee /proc/sys/vm/drop_caches && sudo swapoff -a && sudo swapon -a && free -mh +``` +Also flush disk buffers +``` +blockdev --flushbufs /dev/nvme0n1p4 +``` +3. Run the benchmark +``` +go test -run=^$ github.com/dgraph-io/badger -bench ^BenchmarkDBOpen$ -benchdir="/tmp/data" -v + +badger 2019/06/04 17:15:56 INFO: 126 tables out of 1028 opened in 3.017s +badger 2019/06/04 17:15:59 INFO: 257 tables out of 1028 opened in 6.014s +badger 2019/06/04 17:16:02 INFO: 387 tables out of 1028 opened in 9.017s +badger 2019/06/04 17:16:05 INFO: 516 tables out of 1028 opened in 12.025s +badger 2019/06/04 17:16:08 INFO: 645 tables out of 1028 opened in 15.013s +badger 2019/06/04 17:16:11 INFO: 775 tables out of 1028 opened in 18.008s +badger 2019/06/04 17:16:14 INFO: 906 tables out of 1028 opened in 21.003s +badger 2019/06/04 17:16:17 INFO: All 1028 tables opened in 23.851s +badger 2019/06/04 17:16:17 INFO: Replaying file id: 1998 at offset: 332000 +badger 2019/06/04 17:16:17 INFO: Replay took: 9.81µs +goos: linux +goarch: amd64 +pkg: github.com/dgraph-io/badger +BenchmarkDBOpen-16 1 23930082140 ns/op +PASS +ok github.com/dgraph-io/badger 24.076s + +``` +It takes about 23.851s to open a DB with 2 billion sorted key-value entries. diff --git a/vendor/github.com/dgraph-io/badger/v4/table/builder.go b/vendor/github.com/dgraph-io/badger/v4/table/builder.go new file mode 100644 index 00000000000..5c9e065e0cf --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/table/builder.go @@ -0,0 +1,599 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "crypto/aes" + "math" + "runtime" + "sync" + "sync/atomic" + "unsafe" + + "github.com/golang/protobuf/proto" + "github.com/golang/snappy" + fbs "github.com/google/flatbuffers/go" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/fb" + "github.com/dgraph-io/badger/v4/options" + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +const ( + KB = 1024 + MB = KB * 1024 + + // When a block is encrypted, it's length increases. We add 256 bytes of padding to + // handle cases when block size increases. This is an approximate number. + padding = 256 +) + +type header struct { + overlap uint16 // Overlap with base key. + diff uint16 // Length of the diff. +} + +const headerSize = uint16(unsafe.Sizeof(header{})) + +// Encode encodes the header. +func (h header) Encode() []byte { + var b [4]byte + *(*header)(unsafe.Pointer(&b[0])) = h + return b[:] +} + +// Decode decodes the header. +func (h *header) Decode(buf []byte) { + // Copy over data from buf into h. Using *h=unsafe.pointer(...) leads to + // pointer alignment issues. See https://github.com/dgraph-io/badger/issues/1096 + // and comment https://github.com/dgraph-io/badger/pull/1097#pullrequestreview-307361714 + copy(((*[headerSize]byte)(unsafe.Pointer(h))[:]), buf[:headerSize]) +} + +// bblock represents a block that is being compressed/encrypted in the background. +type bblock struct { + data []byte + baseKey []byte // Base key for the current block. + entryOffsets []uint32 // Offsets of entries present in current block. + end int // Points to the end offset of the block. +} + +// Builder is used in building a table. +type Builder struct { + // Typically tens or hundreds of meg. This is for one single file. + alloc *z.Allocator + curBlock *bblock + compressedSize atomic.Uint32 + uncompressedSize atomic.Uint32 + + lenOffsets uint32 + keyHashes []uint32 // Used for building the bloomfilter. + opts *Options + maxVersion uint64 + onDiskSize uint32 + staleDataSize int + + // Used to concurrently compress/encrypt blocks. + wg sync.WaitGroup + blockChan chan *bblock + blockList []*bblock +} + +func (b *Builder) allocate(need int) []byte { + bb := b.curBlock + if len(bb.data[bb.end:]) < need { + // We need to reallocate. 1GB is the max size that the allocator can allocate. + // While reallocating, if doubling exceeds that limit, then put the upper bound on it. + sz := 2 * len(bb.data) + if sz > (1 << 30) { + sz = 1 << 30 + } + if bb.end+need > sz { + sz = bb.end + need + } + tmp := b.alloc.Allocate(sz) + copy(tmp, bb.data) + bb.data = tmp + } + bb.end += need + return bb.data[bb.end-need : bb.end] +} + +// append appends to curBlock.data +func (b *Builder) append(data []byte) { + dst := b.allocate(len(data)) + y.AssertTrue(len(data) == copy(dst, data)) +} + +const maxAllocatorInitialSz = 256 << 20 + +// NewTableBuilder makes a new TableBuilder. +func NewTableBuilder(opts Options) *Builder { + sz := 2 * int(opts.TableSize) + if sz > maxAllocatorInitialSz { + sz = maxAllocatorInitialSz + } + b := &Builder{ + alloc: opts.AllocPool.Get(sz, "TableBuilder"), + opts: &opts, + } + b.alloc.Tag = "Builder" + b.curBlock = &bblock{ + data: b.alloc.Allocate(opts.BlockSize + padding), + } + b.opts.tableCapacity = uint64(float64(b.opts.TableSize) * 0.95) + + // If encryption or compression is not enabled, do not start compression/encryption goroutines + // and write directly to the buffer. + if b.opts.Compression == options.None && b.opts.DataKey == nil { + return b + } + + count := 2 * runtime.NumCPU() + b.blockChan = make(chan *bblock, count*2) + + b.wg.Add(count) + for i := 0; i < count; i++ { + go b.handleBlock() + } + return b +} + +func maxEncodedLen(ctype options.CompressionType, sz int) int { + switch ctype { + case options.Snappy: + return snappy.MaxEncodedLen(sz) + case options.ZSTD: + return y.ZSTDCompressBound(sz) + } + return sz +} + +func (b *Builder) handleBlock() { + defer b.wg.Done() + + doCompress := b.opts.Compression != options.None + for item := range b.blockChan { + // Extract the block. + blockBuf := item.data[:item.end] + // Compress the block. + if doCompress { + out, err := b.compressData(blockBuf) + y.Check(err) + blockBuf = out + } + if b.shouldEncrypt() { + out, err := b.encrypt(blockBuf) + y.Check(y.Wrapf(err, "Error while encrypting block in table builder.")) + blockBuf = out + } + + // BlockBuf should always less than or equal to allocated space. If the blockBuf is greater + // than allocated space that means the data from this block cannot be stored in its + // existing location. + allocatedSpace := maxEncodedLen(b.opts.Compression, (item.end)) + padding + 1 + y.AssertTrue(len(blockBuf) <= allocatedSpace) + + // blockBuf was allocated on allocator. So, we don't need to copy it over. + item.data = blockBuf + item.end = len(blockBuf) + b.compressedSize.Add(uint32(len(blockBuf))) + } +} + +// Close closes the TableBuilder. +func (b *Builder) Close() { + b.opts.AllocPool.Return(b.alloc) +} + +// Empty returns whether it's empty. +func (b *Builder) Empty() bool { return len(b.keyHashes) == 0 } + +// keyDiff returns a suffix of newKey that is different from b.baseKey. +func (b *Builder) keyDiff(newKey []byte) []byte { + var i int + for i = 0; i < len(newKey) && i < len(b.curBlock.baseKey); i++ { + if newKey[i] != b.curBlock.baseKey[i] { + break + } + } + return newKey[i:] +} + +func (b *Builder) addHelper(key []byte, v y.ValueStruct, vpLen uint32) { + b.keyHashes = append(b.keyHashes, y.Hash(y.ParseKey(key))) + + if version := y.ParseTs(key); version > b.maxVersion { + b.maxVersion = version + } + + // diffKey stores the difference of key with baseKey. + var diffKey []byte + if len(b.curBlock.baseKey) == 0 { + // Make a copy. Builder should not keep references. Otherwise, caller has to be very careful + // and will have to make copies of keys every time they add to builder, which is even worse. + b.curBlock.baseKey = append(b.curBlock.baseKey[:0], key...) + diffKey = key + } else { + diffKey = b.keyDiff(key) + } + + y.AssertTrue(len(key)-len(diffKey) <= math.MaxUint16) + y.AssertTrue(len(diffKey) <= math.MaxUint16) + + h := header{ + overlap: uint16(len(key) - len(diffKey)), + diff: uint16(len(diffKey)), + } + + // store current entry's offset + b.curBlock.entryOffsets = append(b.curBlock.entryOffsets, uint32(b.curBlock.end)) + + // Layout: header, diffKey, value. + b.append(h.Encode()) + b.append(diffKey) + + dst := b.allocate(int(v.EncodedSize())) + v.Encode(dst) + + // Add the vpLen to the onDisk size. We'll add the size of the block to + // onDisk size in Finish() function. + b.onDiskSize += vpLen +} + +/* +Structure of Block. ++-------------------+---------------------+--------------------+--------------+------------------+ +| Entry1 | Entry2 | Entry3 | Entry4 | Entry5 | ++-------------------+---------------------+--------------------+--------------+------------------+ +| Entry6 | ... | ... | ... | EntryN | ++-------------------+---------------------+--------------------+--------------+------------------+ +| Block Meta(contains list of offsets used| Block Meta Size | Block | Checksum Size | +| to perform binary search in the block) | (4 Bytes) | Checksum | (4 Bytes) | ++-----------------------------------------+--------------------+--------------+------------------+ +*/ +// In case the data is encrypted, the "IV" is added to the end of the block. +func (b *Builder) finishBlock() { + if len(b.curBlock.entryOffsets) == 0 { + return + } + // Append the entryOffsets and its length. + b.append(y.U32SliceToBytes(b.curBlock.entryOffsets)) + b.append(y.U32ToBytes(uint32(len(b.curBlock.entryOffsets)))) + + checksum := b.calculateChecksum(b.curBlock.data[:b.curBlock.end]) + + // Append the block checksum and its length. + b.append(checksum) + b.append(y.U32ToBytes(uint32(len(checksum)))) + + b.blockList = append(b.blockList, b.curBlock) + b.uncompressedSize.Add(uint32(b.curBlock.end)) + + // Add length of baseKey (rounded to next multiple of 4 because of alignment). + // Add another 40 Bytes, these additional 40 bytes consists of + // 12 bytes of metadata of flatbuffer + // 8 bytes for Key in flat buffer + // 8 bytes for offset + // 8 bytes for the len + // 4 bytes for the size of slice while SliceAllocate + b.lenOffsets += uint32(int(math.Ceil(float64(len(b.curBlock.baseKey))/4))*4) + 40 + + // If compression/encryption is enabled, we need to send the block to the blockChan. + if b.blockChan != nil { + b.blockChan <- b.curBlock + } +} + +func (b *Builder) shouldFinishBlock(key []byte, value y.ValueStruct) bool { + // If there is no entry till now, we will return false. + if len(b.curBlock.entryOffsets) <= 0 { + return false + } + + // Integer overflow check for statements below. + y.AssertTrue((uint32(len(b.curBlock.entryOffsets))+1)*4+4+8+4 < math.MaxUint32) + // We should include current entry also in size, that's why +1 to len(b.entryOffsets). + entriesOffsetsSize := uint32((len(b.curBlock.entryOffsets)+1)*4 + + 4 + // size of list + 8 + // Sum64 in checksum proto + 4) // checksum length + estimatedSize := uint32(b.curBlock.end) + uint32(6 /*header size for entry*/) + + uint32(len(key)) + value.EncodedSize() + entriesOffsetsSize + + if b.shouldEncrypt() { + // IV is added at the end of the block, while encrypting. + // So, size of IV is added to estimatedSize. + estimatedSize += aes.BlockSize + } + + // Integer overflow check for table size. + y.AssertTrue(uint64(b.curBlock.end)+uint64(estimatedSize) < math.MaxUint32) + + return estimatedSize > uint32(b.opts.BlockSize) +} + +// AddStaleKey is same is Add function but it also increments the internal +// staleDataSize counter. This value will be used to prioritize this table for +// compaction. +func (b *Builder) AddStaleKey(key []byte, v y.ValueStruct, valueLen uint32) { + // Rough estimate based on how much space it will occupy in the SST. + b.staleDataSize += len(key) + len(v.Value) + 4 /* entry offset */ + 4 /* header size */ + b.addInternal(key, v, valueLen, true) +} + +// Add adds a key-value pair to the block. +func (b *Builder) Add(key []byte, value y.ValueStruct, valueLen uint32) { + b.addInternal(key, value, valueLen, false) +} + +func (b *Builder) addInternal(key []byte, value y.ValueStruct, valueLen uint32, isStale bool) { + if b.shouldFinishBlock(key, value) { + if isStale { + // This key will be added to tableIndex and it is stale. + b.staleDataSize += len(key) + 4 /* len */ + 4 /* offset */ + } + b.finishBlock() + // Create a new block and start writing. + b.curBlock = &bblock{ + data: b.alloc.Allocate(b.opts.BlockSize + padding), + } + } + b.addHelper(key, value, valueLen) +} + +// TODO: vvv this was the comment on ReachedCapacity. +// FinalSize returns the *rough* final size of the array, counting the header which is +// not yet written. +// TODO: Look into why there is a discrepancy. I suspect it is because of Write(empty, empty) +// at the end. The diff can vary. + +// ReachedCapacity returns true if we... roughly (?) reached capacity? +func (b *Builder) ReachedCapacity() bool { + // If encryption/compression is enabled then use the compresssed size. + sumBlockSizes := b.compressedSize.Load() + if b.opts.Compression == options.None && b.opts.DataKey == nil { + sumBlockSizes = b.uncompressedSize.Load() + } + blocksSize := sumBlockSizes + // actual length of current buffer + uint32(len(b.curBlock.entryOffsets)*4) + // all entry offsets size + 4 + // count of all entry offsets + 8 + // checksum bytes + 4 // checksum length + + estimateSz := blocksSize + + 4 + // Index length + b.lenOffsets + + return uint64(estimateSz) > b.opts.tableCapacity +} + +// Finish finishes the table by appending the index. +/* +The table structure looks like ++---------+------------+-----------+---------------+ +| Block 1 | Block 2 | Block 3 | Block 4 | ++---------+------------+-----------+---------------+ +| Block 5 | Block 6 | Block ... | Block N | ++---------+------------+-----------+---------------+ +| Index | Index Size | Checksum | Checksum Size | ++---------+------------+-----------+---------------+ +*/ +// In case the data is encrypted, the "IV" is added to the end of the index. +func (b *Builder) Finish() []byte { + bd := b.Done() + buf := make([]byte, bd.Size) + written := bd.Copy(buf) + y.AssertTrue(written == len(buf)) + return buf +} + +type buildData struct { + blockList []*bblock + index []byte + checksum []byte + Size int + alloc *z.Allocator +} + +func (bd *buildData) Copy(dst []byte) int { + var written int + for _, bl := range bd.blockList { + written += copy(dst[written:], bl.data[:bl.end]) + } + written += copy(dst[written:], bd.index) + written += copy(dst[written:], y.U32ToBytes(uint32(len(bd.index)))) + + written += copy(dst[written:], bd.checksum) + written += copy(dst[written:], y.U32ToBytes(uint32(len(bd.checksum)))) + return written +} + +func (b *Builder) Done() buildData { + b.finishBlock() // This will never start a new block. + if b.blockChan != nil { + close(b.blockChan) + } + // Wait for block handler to finish. + b.wg.Wait() + + if len(b.blockList) == 0 { + return buildData{} + } + bd := buildData{ + blockList: b.blockList, + alloc: b.alloc, + } + + var f y.Filter + if b.opts.BloomFalsePositive > 0 { + bits := y.BloomBitsPerKey(len(b.keyHashes), b.opts.BloomFalsePositive) + f = y.NewFilter(b.keyHashes, bits) + } + index, dataSize := b.buildIndex(f) + + var err error + if b.shouldEncrypt() { + index, err = b.encrypt(index) + y.Check(err) + } + checksum := b.calculateChecksum(index) + + bd.index = index + bd.checksum = checksum + bd.Size = int(dataSize) + len(index) + len(checksum) + 4 + 4 + return bd +} + +func (b *Builder) calculateChecksum(data []byte) []byte { + // Build checksum for the index. + checksum := pb.Checksum{ + // TODO: The checksum type should be configurable from the + // options. + // We chose to use CRC32 as the default option because + // it performed better compared to xxHash64. + // See the BenchmarkChecksum in table_test.go file + // Size => 1024 B 2048 B + // CRC32 => 63.7 ns/op 112 ns/op + // xxHash64 => 87.5 ns/op 158 ns/op + Sum: y.CalculateChecksum(data, pb.Checksum_CRC32C), + Algo: pb.Checksum_CRC32C, + } + + // Write checksum to the file. + chksum, err := proto.Marshal(&checksum) + y.Check(err) + // Write checksum size. + return chksum +} + +// DataKey returns datakey of the builder. +func (b *Builder) DataKey() *pb.DataKey { + return b.opts.DataKey +} + +func (b *Builder) Opts() *Options { + return b.opts +} + +// encrypt will encrypt the given data and appends IV to the end of the encrypted data. +// This should be only called only after checking shouldEncrypt method. +func (b *Builder) encrypt(data []byte) ([]byte, error) { + iv, err := y.GenerateIV() + if err != nil { + return data, y.Wrapf(err, "Error while generating IV in Builder.encrypt") + } + needSz := len(data) + len(iv) + dst := b.alloc.Allocate(needSz) + + if err = y.XORBlock(dst[:len(data)], data, b.DataKey().Data, iv); err != nil { + return data, y.Wrapf(err, "Error while encrypting in Builder.encrypt") + } + + y.AssertTrue(len(iv) == copy(dst[len(data):], iv)) + return dst, nil +} + +// shouldEncrypt tells us whether to encrypt the data or not. +// We encrypt only if the data key exist. Otherwise, not. +func (b *Builder) shouldEncrypt() bool { + return b.opts.DataKey != nil +} + +// compressData compresses the given data. +func (b *Builder) compressData(data []byte) ([]byte, error) { + switch b.opts.Compression { + case options.None: + return data, nil + case options.Snappy: + sz := snappy.MaxEncodedLen(len(data)) + dst := b.alloc.Allocate(sz) + return snappy.Encode(dst, data), nil + case options.ZSTD: + sz := y.ZSTDCompressBound(len(data)) + dst := b.alloc.Allocate(sz) + return y.ZSTDCompress(dst, data, b.opts.ZSTDCompressionLevel) + } + return nil, errors.New("Unsupported compression type") +} + +func (b *Builder) buildIndex(bloom []byte) ([]byte, uint32) { + builder := fbs.NewBuilder(3 << 20) + + boList, dataSize := b.writeBlockOffsets(builder) + // Write block offset vector the the idxBuilder. + fb.TableIndexStartOffsetsVector(builder, len(boList)) + + // Write individual block offsets in reverse order to work around how Flatbuffers expects it. + for i := len(boList) - 1; i >= 0; i-- { + builder.PrependUOffsetT(boList[i]) + } + boEnd := builder.EndVector(len(boList)) + + var bfoff fbs.UOffsetT + // Write the bloom filter. + if len(bloom) > 0 { + bfoff = builder.CreateByteVector(bloom) + } + b.onDiskSize += dataSize + fb.TableIndexStart(builder) + fb.TableIndexAddOffsets(builder, boEnd) + fb.TableIndexAddBloomFilter(builder, bfoff) + fb.TableIndexAddMaxVersion(builder, b.maxVersion) + fb.TableIndexAddUncompressedSize(builder, b.uncompressedSize.Load()) + fb.TableIndexAddKeyCount(builder, uint32(len(b.keyHashes))) + fb.TableIndexAddOnDiskSize(builder, b.onDiskSize) + fb.TableIndexAddStaleDataSize(builder, uint32(b.staleDataSize)) + builder.Finish(fb.TableIndexEnd(builder)) + + buf := builder.FinishedBytes() + index := fb.GetRootAsTableIndex(buf, 0) + // Mutate the ondisk size to include the size of the index as well. + y.AssertTrue(index.MutateOnDiskSize(index.OnDiskSize() + uint32(len(buf)))) + return buf, dataSize +} + +// writeBlockOffsets writes all the blockOffets in b.offsets and returns the +// offsets for the newly written items. +func (b *Builder) writeBlockOffsets(builder *fbs.Builder) ([]fbs.UOffsetT, uint32) { + var startOffset uint32 + var uoffs []fbs.UOffsetT + for _, bl := range b.blockList { + uoff := b.writeBlockOffset(builder, bl, startOffset) + uoffs = append(uoffs, uoff) + startOffset += uint32(bl.end) + } + return uoffs, startOffset +} + +// writeBlockOffset writes the given key,offset,len triple to the indexBuilder. +// It returns the offset of the newly written blockoffset. +func (b *Builder) writeBlockOffset( + builder *fbs.Builder, bl *bblock, startOffset uint32) fbs.UOffsetT { + // Write the key to the buffer. + k := builder.CreateByteVector(bl.baseKey) + + // Build the blockOffset. + fb.BlockOffsetStart(builder) + fb.BlockOffsetAddKey(builder, k) + fb.BlockOffsetAddOffset(builder, startOffset) + fb.BlockOffsetAddLen(builder, uint32(bl.end)) + return fb.BlockOffsetEnd(builder) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/table/iterator.go b/vendor/github.com/dgraph-io/badger/v4/table/iterator.go new file mode 100644 index 00000000000..10985658273 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/table/iterator.go @@ -0,0 +1,568 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "bytes" + "fmt" + "io" + "sort" + + "github.com/dgraph-io/badger/v4/fb" + "github.com/dgraph-io/badger/v4/y" +) + +type blockIterator struct { + data []byte + idx int // Idx of the entry inside a block + err error + baseKey []byte + key []byte + val []byte + entryOffsets []uint32 + block *block + + tableID uint64 + blockID int + // prevOverlap stores the overlap of the previous key with the base key. + // This avoids unnecessary copy of base key when the overlap is same for multiple keys. + prevOverlap uint16 +} + +func (itr *blockIterator) setBlock(b *block) { + // Decrement the ref for the old block. If the old block was compressed, we + // might be able to reuse it. + itr.block.decrRef() + + itr.block = b + itr.err = nil + itr.idx = 0 + itr.baseKey = itr.baseKey[:0] + itr.prevOverlap = 0 + itr.key = itr.key[:0] + itr.val = itr.val[:0] + // Drop the index from the block. We don't need it anymore. + itr.data = b.data[:b.entriesIndexStart] + itr.entryOffsets = b.entryOffsets +} + +// setIdx sets the iterator to the entry at index i and set it's key and value. +func (itr *blockIterator) setIdx(i int) { + itr.idx = i + if i >= len(itr.entryOffsets) || i < 0 { + itr.err = io.EOF + return + } + itr.err = nil + startOffset := int(itr.entryOffsets[i]) + + // Set base key. + if len(itr.baseKey) == 0 { + var baseHeader header + baseHeader.Decode(itr.data) + itr.baseKey = itr.data[headerSize : headerSize+baseHeader.diff] + } + + var endOffset int + // idx points to the last entry in the block. + if itr.idx+1 == len(itr.entryOffsets) { + endOffset = len(itr.data) + } else { + // idx point to some entry other than the last one in the block. + // EndOffset of the current entry is the start offset of the next entry. + endOffset = int(itr.entryOffsets[itr.idx+1]) + } + defer func() { + if r := recover(); r != nil { + var debugBuf bytes.Buffer + fmt.Fprintf(&debugBuf, "==== Recovered====\n") + fmt.Fprintf(&debugBuf, "Table ID: %d\nBlock ID: %d\nEntry Idx: %d\nData len: %d\n"+ + "StartOffset: %d\nEndOffset: %d\nEntryOffsets len: %d\nEntryOffsets: %v\n", + itr.tableID, itr.blockID, itr.idx, len(itr.data), startOffset, endOffset, + len(itr.entryOffsets), itr.entryOffsets) + panic(debugBuf.String()) + } + }() + + entryData := itr.data[startOffset:endOffset] + var h header + h.Decode(entryData) + // Header contains the length of key overlap and difference compared to the base key. If the key + // before this one had the same or better key overlap, we can avoid copying that part into + // itr.key. But, if the overlap was lesser, we could copy over just that portion. + if h.overlap > itr.prevOverlap { + itr.key = append(itr.key[:itr.prevOverlap], itr.baseKey[itr.prevOverlap:h.overlap]...) + } + itr.prevOverlap = h.overlap + valueOff := headerSize + h.diff + diffKey := entryData[headerSize:valueOff] + itr.key = append(itr.key[:h.overlap], diffKey...) + itr.val = entryData[valueOff:] +} + +func (itr *blockIterator) Valid() bool { + return itr != nil && itr.err == nil +} + +func (itr *blockIterator) Error() error { + return itr.err +} + +func (itr *blockIterator) Close() { + itr.block.decrRef() +} + +var ( + origin = 0 + current = 1 +) + +// seek brings us to the first block element that is >= input key. +func (itr *blockIterator) seek(key []byte, whence int) { + itr.err = nil + startIndex := 0 // This tells from which index we should start binary search. + + switch whence { + case origin: + // We don't need to do anything. startIndex is already at 0 + case current: + startIndex = itr.idx + } + + foundEntryIdx := sort.Search(len(itr.entryOffsets), func(idx int) bool { + // If idx is less than start index then just return false. + if idx < startIndex { + return false + } + itr.setIdx(idx) + return y.CompareKeys(itr.key, key) >= 0 + }) + itr.setIdx(foundEntryIdx) +} + +// seekToFirst brings us to the first element. +func (itr *blockIterator) seekToFirst() { + itr.setIdx(0) +} + +// seekToLast brings us to the last element. +func (itr *blockIterator) seekToLast() { + itr.setIdx(len(itr.entryOffsets) - 1) +} + +func (itr *blockIterator) next() { + itr.setIdx(itr.idx + 1) +} + +func (itr *blockIterator) prev() { + itr.setIdx(itr.idx - 1) +} + +// Iterator is an iterator for a Table. +type Iterator struct { + t *Table + bpos int + bi blockIterator + err error + + // Internally, Iterator is bidirectional. However, we only expose the + // unidirectional functionality for now. + opt int // Valid options are REVERSED and NOCACHE. +} + +// NewIterator returns a new iterator of the Table +func (t *Table) NewIterator(opt int) *Iterator { + t.IncrRef() // Important. + ti := &Iterator{t: t, opt: opt} + return ti +} + +// Close closes the iterator (and it must be called). +func (itr *Iterator) Close() error { + itr.bi.Close() + return itr.t.DecrRef() +} + +func (itr *Iterator) reset() { + itr.bpos = 0 + itr.err = nil +} + +// Valid follows the y.Iterator interface +func (itr *Iterator) Valid() bool { + return itr.err == nil +} + +func (itr *Iterator) useCache() bool { + return itr.opt&NOCACHE == 0 +} + +func (itr *Iterator) seekToFirst() { + numBlocks := itr.t.offsetsLength() + if numBlocks == 0 { + itr.err = io.EOF + return + } + itr.bpos = 0 + block, err := itr.t.block(itr.bpos, itr.useCache()) + if err != nil { + itr.err = err + return + } + itr.bi.tableID = itr.t.id + itr.bi.blockID = itr.bpos + itr.bi.setBlock(block) + itr.bi.seekToFirst() + itr.err = itr.bi.Error() +} + +func (itr *Iterator) seekToLast() { + numBlocks := itr.t.offsetsLength() + if numBlocks == 0 { + itr.err = io.EOF + return + } + itr.bpos = numBlocks - 1 + block, err := itr.t.block(itr.bpos, itr.useCache()) + if err != nil { + itr.err = err + return + } + itr.bi.tableID = itr.t.id + itr.bi.blockID = itr.bpos + itr.bi.setBlock(block) + itr.bi.seekToLast() + itr.err = itr.bi.Error() +} + +func (itr *Iterator) seekHelper(blockIdx int, key []byte) { + itr.bpos = blockIdx + block, err := itr.t.block(blockIdx, itr.useCache()) + if err != nil { + itr.err = err + return + } + itr.bi.tableID = itr.t.id + itr.bi.blockID = itr.bpos + itr.bi.setBlock(block) + itr.bi.seek(key, origin) + itr.err = itr.bi.Error() +} + +// seekFrom brings us to a key that is >= input key. +func (itr *Iterator) seekFrom(key []byte, whence int) { + itr.err = nil + switch whence { + case origin: + itr.reset() + case current: + } + + var ko fb.BlockOffset + idx := sort.Search(itr.t.offsetsLength(), func(idx int) bool { + // Offsets should never return false since we're iterating within the OffsetsLength. + y.AssertTrue(itr.t.offsets(&ko, idx)) + return y.CompareKeys(ko.KeyBytes(), key) > 0 + }) + if idx == 0 { + // The smallest key in our table is already strictly > key. We can return that. + // This is like a SeekToFirst. + itr.seekHelper(0, key) + return + } + + // block[idx].smallest is > key. + // Since idx>0, we know block[idx-1].smallest is <= key. + // There are two cases. + // 1) Everything in block[idx-1] is strictly < key. In this case, we should go to the first + // element of block[idx]. + // 2) Some element in block[idx-1] is >= key. We should go to that element. + itr.seekHelper(idx-1, key) + if itr.err == io.EOF { + // Case 1. Need to visit block[idx]. + if idx == itr.t.offsetsLength() { + // If idx == len(itr.t.blockIndex), then input key is greater than ANY element of table. + // There's nothing we can do. Valid() should return false as we seek to end of table. + return + } + // Since block[idx].smallest is > key. This is essentially a block[idx].SeekToFirst. + itr.seekHelper(idx, key) + } + // Case 2: No need to do anything. We already did the seek in block[idx-1]. +} + +// seek will reset iterator and seek to >= key. +func (itr *Iterator) seek(key []byte) { + itr.seekFrom(key, origin) +} + +// seekForPrev will reset iterator and seek to <= key. +func (itr *Iterator) seekForPrev(key []byte) { + // TODO: Optimize this. We shouldn't have to take a Prev step. + itr.seekFrom(key, origin) + if !bytes.Equal(itr.Key(), key) { + itr.prev() + } +} + +func (itr *Iterator) next() { + itr.err = nil + + if itr.bpos >= itr.t.offsetsLength() { + itr.err = io.EOF + return + } + + if len(itr.bi.data) == 0 { + block, err := itr.t.block(itr.bpos, itr.useCache()) + if err != nil { + itr.err = err + return + } + itr.bi.tableID = itr.t.id + itr.bi.blockID = itr.bpos + itr.bi.setBlock(block) + itr.bi.seekToFirst() + itr.err = itr.bi.Error() + return + } + + itr.bi.next() + if !itr.bi.Valid() { + itr.bpos++ + itr.bi.data = nil + itr.next() + return + } +} + +func (itr *Iterator) prev() { + itr.err = nil + if itr.bpos < 0 { + itr.err = io.EOF + return + } + + if len(itr.bi.data) == 0 { + block, err := itr.t.block(itr.bpos, itr.useCache()) + if err != nil { + itr.err = err + return + } + itr.bi.tableID = itr.t.id + itr.bi.blockID = itr.bpos + itr.bi.setBlock(block) + itr.bi.seekToLast() + itr.err = itr.bi.Error() + return + } + + itr.bi.prev() + if !itr.bi.Valid() { + itr.bpos-- + itr.bi.data = nil + itr.prev() + return + } +} + +// Key follows the y.Iterator interface. +// Returns the key with timestamp. +func (itr *Iterator) Key() []byte { + return itr.bi.key +} + +// Value follows the y.Iterator interface +func (itr *Iterator) Value() (ret y.ValueStruct) { + ret.Decode(itr.bi.val) + return +} + +// ValueCopy copies the current value and returns it as decoded +// ValueStruct. +func (itr *Iterator) ValueCopy() (ret y.ValueStruct) { + dst := y.Copy(itr.bi.val) + ret.Decode(dst) + return +} + +// Next follows the y.Iterator interface +func (itr *Iterator) Next() { + if itr.opt&REVERSED == 0 { + itr.next() + } else { + itr.prev() + } +} + +// Rewind follows the y.Iterator interface +func (itr *Iterator) Rewind() { + if itr.opt&REVERSED == 0 { + itr.seekToFirst() + } else { + itr.seekToLast() + } +} + +// Seek follows the y.Iterator interface +func (itr *Iterator) Seek(key []byte) { + if itr.opt&REVERSED == 0 { + itr.seek(key) + } else { + itr.seekForPrev(key) + } +} + +var ( + REVERSED int = 2 + NOCACHE int = 4 +) + +// ConcatIterator concatenates the sequences defined by several iterators. (It only works with +// TableIterators, probably just because it's faster to not be so generic.) +type ConcatIterator struct { + idx int // Which iterator is active now. + cur *Iterator + iters []*Iterator // Corresponds to tables. + tables []*Table // Disregarding reversed, this is in ascending order. + options int // Valid options are REVERSED and NOCACHE. +} + +// NewConcatIterator creates a new concatenated iterator +func NewConcatIterator(tbls []*Table, opt int) *ConcatIterator { + iters := make([]*Iterator, len(tbls)) + for i := 0; i < len(tbls); i++ { + // Increment the reference count. Since, we're not creating the iterator right now. + // Here, We'll hold the reference of the tables, till the lifecycle of the iterator. + tbls[i].IncrRef() + + // Save cycles by not initializing the iterators until needed. + // iters[i] = tbls[i].NewIterator(reversed) + } + return &ConcatIterator{ + options: opt, + iters: iters, + tables: tbls, + idx: -1, // Not really necessary because s.it.Valid()=false, but good to have. + } +} + +func (s *ConcatIterator) setIdx(idx int) { + s.idx = idx + if idx < 0 || idx >= len(s.iters) { + s.cur = nil + return + } + if s.iters[idx] == nil { + s.iters[idx] = s.tables[idx].NewIterator(s.options) + } + s.cur = s.iters[s.idx] +} + +// Rewind implements y.Interface +func (s *ConcatIterator) Rewind() { + if len(s.iters) == 0 { + return + } + if s.options&REVERSED == 0 { + s.setIdx(0) + } else { + s.setIdx(len(s.iters) - 1) + } + s.cur.Rewind() +} + +// Valid implements y.Interface +func (s *ConcatIterator) Valid() bool { + return s.cur != nil && s.cur.Valid() +} + +// Key implements y.Interface +func (s *ConcatIterator) Key() []byte { + return s.cur.Key() +} + +// Value implements y.Interface +func (s *ConcatIterator) Value() y.ValueStruct { + return s.cur.Value() +} + +// Seek brings us to element >= key if reversed is false. Otherwise, <= key. +func (s *ConcatIterator) Seek(key []byte) { + var idx int + if s.options&REVERSED == 0 { + idx = sort.Search(len(s.tables), func(i int) bool { + return y.CompareKeys(s.tables[i].Biggest(), key) >= 0 + }) + } else { + n := len(s.tables) + idx = n - 1 - sort.Search(n, func(i int) bool { + return y.CompareKeys(s.tables[n-1-i].Smallest(), key) <= 0 + }) + } + if idx >= len(s.tables) || idx < 0 { + s.setIdx(-1) + return + } + // For reversed=false, we know s.tables[i-1].Biggest() < key. Thus, the + // previous table cannot possibly contain key. + s.setIdx(idx) + s.cur.Seek(key) +} + +// Next advances our concat iterator. +func (s *ConcatIterator) Next() { + s.cur.Next() + if s.cur.Valid() { + // Nothing to do. Just stay with the current table. + return + } + for { // In case there are empty tables. + if s.options&REVERSED == 0 { + s.setIdx(s.idx + 1) + } else { + s.setIdx(s.idx - 1) + } + if s.cur == nil { + // End of list. Valid will become false. + return + } + s.cur.Rewind() + if s.cur.Valid() { + break + } + } +} + +// Close implements y.Interface. +func (s *ConcatIterator) Close() error { + for _, t := range s.tables { + // DeReference the tables while closing the iterator. + if err := t.DecrRef(); err != nil { + return err + } + } + for _, it := range s.iters { + if it == nil { + continue + } + if err := it.Close(); err != nil { + return y.Wrap(err, "ConcatIterator") + } + } + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/table/merge_iterator.go b/vendor/github.com/dgraph-io/badger/v4/table/merge_iterator.go new file mode 100644 index 00000000000..6e89fbbc241 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/table/merge_iterator.go @@ -0,0 +1,231 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "bytes" + + "github.com/dgraph-io/badger/v4/y" +) + +// MergeIterator merges multiple iterators. +// NOTE: MergeIterator owns the array of iterators and is responsible for closing them. +type MergeIterator struct { + left node + right node + small *node + + curKey []byte + reverse bool +} + +type node struct { + valid bool + key []byte + iter y.Iterator + + // The two iterators are type asserted from `y.Iterator`, used to inline more function calls. + // Calling functions on concrete types is much faster (about 25-30%) than calling the + // interface's function. + merge *MergeIterator + concat *ConcatIterator +} + +func (n *node) setIterator(iter y.Iterator) { + n.iter = iter + // It's okay if the type assertion below fails and n.merge/n.concat are set to nil. + // We handle the nil values of merge and concat in all the methods. + n.merge, _ = iter.(*MergeIterator) + n.concat, _ = iter.(*ConcatIterator) +} + +func (n *node) setKey() { + switch { + case n.merge != nil: + n.valid = n.merge.small.valid + if n.valid { + n.key = n.merge.small.key + } + case n.concat != nil: + n.valid = n.concat.Valid() + if n.valid { + n.key = n.concat.Key() + } + default: + n.valid = n.iter.Valid() + if n.valid { + n.key = n.iter.Key() + } + } +} + +func (n *node) next() { + switch { + case n.merge != nil: + n.merge.Next() + case n.concat != nil: + n.concat.Next() + default: + n.iter.Next() + } + n.setKey() +} + +func (n *node) rewind() { + n.iter.Rewind() + n.setKey() +} + +func (n *node) seek(key []byte) { + n.iter.Seek(key) + n.setKey() +} + +func (mi *MergeIterator) fix() { + if !mi.bigger().valid { + return + } + if !mi.small.valid { + mi.swapSmall() + return + } + cmp := y.CompareKeys(mi.small.key, mi.bigger().key) + switch { + case cmp == 0: // Both the keys are equal. + // In case of same keys, move the right iterator ahead. + mi.right.next() + if &mi.right == mi.small { + mi.swapSmall() + } + return + case cmp < 0: // Small is less than bigger(). + if mi.reverse { + mi.swapSmall() + } else { //nolint:staticcheck + // we don't need to do anything. Small already points to the smallest. + } + return + default: // bigger() is less than small. + if mi.reverse { + // Do nothing since we're iterating in reverse. Small currently points to + // the bigger key and that's okay in reverse iteration. + } else { + mi.swapSmall() + } + return + } +} + +func (mi *MergeIterator) bigger() *node { + if mi.small == &mi.left { + return &mi.right + } + return &mi.left +} + +func (mi *MergeIterator) swapSmall() { + if mi.small == &mi.left { + mi.small = &mi.right + return + } + if mi.small == &mi.right { + mi.small = &mi.left + return + } +} + +// Next returns the next element. If it is the same as the current key, ignore it. +func (mi *MergeIterator) Next() { + for mi.Valid() { + if !bytes.Equal(mi.small.key, mi.curKey) { + break + } + mi.small.next() + mi.fix() + } + mi.setCurrent() +} + +func (mi *MergeIterator) setCurrent() { + mi.curKey = append(mi.curKey[:0], mi.small.key...) +} + +// Rewind seeks to first element (or last element for reverse iterator). +func (mi *MergeIterator) Rewind() { + mi.left.rewind() + mi.right.rewind() + mi.fix() + mi.setCurrent() +} + +// Seek brings us to element with key >= given key. +func (mi *MergeIterator) Seek(key []byte) { + mi.left.seek(key) + mi.right.seek(key) + mi.fix() + mi.setCurrent() +} + +// Valid returns whether the MergeIterator is at a valid element. +func (mi *MergeIterator) Valid() bool { + return mi.small.valid +} + +// Key returns the key associated with the current iterator. +func (mi *MergeIterator) Key() []byte { + return mi.small.key +} + +// Value returns the value associated with the iterator. +func (mi *MergeIterator) Value() y.ValueStruct { + return mi.small.iter.Value() +} + +// Close implements y.Iterator. +func (mi *MergeIterator) Close() error { + err1 := mi.left.iter.Close() + err2 := mi.right.iter.Close() + if err1 != nil { + return y.Wrap(err1, "MergeIterator") + } + return y.Wrap(err2, "MergeIterator") +} + +// NewMergeIterator creates a merge iterator. +func NewMergeIterator(iters []y.Iterator, reverse bool) y.Iterator { + switch len(iters) { + case 0: + return nil + case 1: + return iters[0] + case 2: + mi := &MergeIterator{ + reverse: reverse, + } + mi.left.setIterator(iters[0]) + mi.right.setIterator(iters[1]) + // Assign left iterator randomly. This will be fixed when user calls rewind/seek. + mi.small = &mi.left + return mi + } + mid := len(iters) / 2 + return NewMergeIterator( + []y.Iterator{ + NewMergeIterator(iters[:mid], reverse), + NewMergeIterator(iters[mid:], reverse), + }, reverse) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/table/table.go b/vendor/github.com/dgraph-io/badger/v4/table/table.go new file mode 100644 index 00000000000..0bbc9108920 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/table/table.go @@ -0,0 +1,842 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package table + +import ( + "bytes" + "crypto/aes" + "encoding/binary" + "fmt" + "math" + "os" + "path/filepath" + "strconv" + "strings" + "sync" + "sync/atomic" + "time" + "unsafe" + + "github.com/golang/protobuf/proto" + "github.com/golang/snappy" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/fb" + "github.com/dgraph-io/badger/v4/options" + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto" + "github.com/dgraph-io/ristretto/z" +) + +const fileSuffix = ".sst" +const intSize = int(unsafe.Sizeof(int(0))) + +// Options contains configurable options for Table/Builder. +type Options struct { + // Options for Opening/Building Table. + + // Open tables in read only mode. + ReadOnly bool + MetricsEnabled bool + + // Maximum size of the table. + TableSize uint64 + tableCapacity uint64 // 0.9x TableSize. + + // ChkMode is the checksum verification mode for Table. + ChkMode options.ChecksumVerificationMode + + // Options for Table builder. + + // BloomFalsePositive is the false positive probabiltiy of bloom filter. + BloomFalsePositive float64 + + // BlockSize is the size of each block inside SSTable in bytes. + BlockSize int + + // DataKey is the key used to decrypt the encrypted text. + DataKey *pb.DataKey + + // Compression indicates the compression algorithm used for block compression. + Compression options.CompressionType + + // Block cache is used to cache decompressed and decrypted blocks. + BlockCache *ristretto.Cache + IndexCache *ristretto.Cache + + AllocPool *z.AllocatorPool + + // ZSTDCompressionLevel is the ZSTD compression level used for compressing blocks. + ZSTDCompressionLevel int +} + +// TableInterface is useful for testing. +type TableInterface interface { + Smallest() []byte + Biggest() []byte + DoesNotHave(hash uint32) bool + MaxVersion() uint64 +} + +// Table represents a loaded table file with the info we have about it. +type Table struct { + sync.Mutex + *z.MmapFile + + tableSize int // Initialized in OpenTable, using fd.Stat(). + + _index *fb.TableIndex // Nil if encryption is enabled. Use fetchIndex to access. + _cheap *cheapIndex + ref atomic.Int32 // For file garbage collection + + // The following are initialized once and const. + smallest, biggest []byte // Smallest and largest keys (with timestamps). + id uint64 // file id, part of filename + + Checksum []byte + CreatedAt time.Time + indexStart int + indexLen int + hasBloomFilter bool + + IsInmemory bool // Set to true if the table is on level 0 and opened in memory. + opt *Options +} + +type cheapIndex struct { + MaxVersion uint64 + KeyCount uint32 + UncompressedSize uint32 + OnDiskSize uint32 + BloomFilterLength int + OffsetsLength int +} + +func (t *Table) cheapIndex() *cheapIndex { + return t._cheap +} +func (t *Table) offsetsLength() int { return t.cheapIndex().OffsetsLength } + +// MaxVersion returns the maximum version across all keys stored in this table. +func (t *Table) MaxVersion() uint64 { return t.cheapIndex().MaxVersion } + +// BloomFilterSize returns the size of the bloom filter in bytes stored in memory. +func (t *Table) BloomFilterSize() int { return t.cheapIndex().BloomFilterLength } + +// UncompressedSize is the size uncompressed data stored in this file. +func (t *Table) UncompressedSize() uint32 { return t.cheapIndex().UncompressedSize } + +// KeyCount is the total number of keys in this table. +func (t *Table) KeyCount() uint32 { return t.cheapIndex().KeyCount } + +// OnDiskSize returns the total size of key-values stored in this table (including the +// disk space occupied on the value log). +func (t *Table) OnDiskSize() uint32 { return t.cheapIndex().OnDiskSize } + +// CompressionType returns the compression algorithm used for block compression. +func (t *Table) CompressionType() options.CompressionType { + return t.opt.Compression +} + +// IncrRef increments the refcount (having to do with whether the file should be deleted) +func (t *Table) IncrRef() { + t.ref.Add(1) +} + +// DecrRef decrements the refcount and possibly deletes the table +func (t *Table) DecrRef() error { + newRef := t.ref.Add(-1) + if newRef == 0 { + // We can safely delete this file, because for all the current files, we always have + // at least one reference pointing to them. + + // Delete all blocks from the cache. + for i := 0; i < t.offsetsLength(); i++ { + t.opt.BlockCache.Del(t.blockCacheKey(i)) + } + if err := t.Delete(); err != nil { + return err + } + } + return nil +} + +// BlockEvictHandler is used to reuse the byte slice stored in the block on cache eviction. +func BlockEvictHandler(value interface{}) { + if b, ok := value.(*block); ok { + b.decrRef() + } +} + +type block struct { + offset int + data []byte + checksum []byte + entriesIndexStart int // start index of entryOffsets list + entryOffsets []uint32 // used to binary search an entry in the block. + chkLen int // checksum length. + freeMe bool // used to determine if the blocked should be reused. + ref atomic.Int32 +} + +var NumBlocks atomic.Int32 + +// incrRef increments the ref of a block and return a bool indicating if the +// increment was successful. A true value indicates that the block can be used. +func (b *block) incrRef() bool { + for { + // We can't blindly add 1 to ref. We need to check whether it has + // reached zero first, because if it did, then we should absolutely not + // use this block. + ref := b.ref.Load() + // The ref would not be equal to 0 unless the existing + // block get evicted before this line. If the ref is zero, it means that + // the block is already added the the blockPool and cannot be used + // anymore. The ref of a new block is 1 so the following condition will + // be true only if the block got reused before we could increment its + // ref. + if ref == 0 { + return false + } + // Increment the ref only if it is not zero and has not changed between + // the time we read it and we're updating it. + // + if b.ref.CompareAndSwap(ref, ref+1) { + return true + } + } +} +func (b *block) decrRef() { + if b == nil { + return + } + + // Insert the []byte into pool only if the block is resuable. When a block + // is reusable a new []byte is used for decompression and this []byte can + // be reused. + // In case of an uncompressed block, the []byte is a reference to the + // table.mmap []byte slice. Any attempt to write data to the mmap []byte + // will lead to SEGFAULT. + if b.ref.Add(-1) == 0 { + if b.freeMe { + z.Free(b.data) + } + NumBlocks.Add(-1) + // blockPool.Put(&b.data) + } + y.AssertTrue(b.ref.Load() >= 0) +} +func (b *block) size() int64 { + return int64(3*intSize /* Size of the offset, entriesIndexStart and chkLen */ + + cap(b.data) + cap(b.checksum) + cap(b.entryOffsets)*4) +} + +func (b *block) verifyCheckSum() error { + cs := &pb.Checksum{} + if err := proto.Unmarshal(b.checksum, cs); err != nil { + return y.Wrapf(err, "unable to unmarshal checksum for block") + } + return y.VerifyChecksum(b.data, cs) +} + +func CreateTable(fname string, builder *Builder) (*Table, error) { + bd := builder.Done() + mf, err := z.OpenMmapFile(fname, os.O_CREATE|os.O_RDWR|os.O_EXCL, bd.Size) + if err == z.NewFile { + // Expected. + } else if err != nil { + return nil, y.Wrapf(err, "while creating table: %s", fname) + } else { + return nil, errors.Errorf("file already exists: %s", fname) + } + + written := bd.Copy(mf.Data) + y.AssertTrue(written == len(mf.Data)) + if err := z.Msync(mf.Data); err != nil { + return nil, y.Wrapf(err, "while calling msync on %s", fname) + } + return OpenTable(mf, *builder.opts) +} + +// OpenTable assumes file has only one table and opens it. Takes ownership of fd upon function +// entry. Returns a table with one reference count on it (decrementing which may delete the file! +// -- consider t.Close() instead). The fd has to writeable because we call Truncate on it before +// deleting. Checksum for all blocks of table is verified based on value of chkMode. +func OpenTable(mf *z.MmapFile, opts Options) (*Table, error) { + // BlockSize is used to compute the approximate size of the decompressed + // block. It should not be zero if the table is compressed. + if opts.BlockSize == 0 && opts.Compression != options.None { + return nil, errors.New("Block size cannot be zero") + } + fileInfo, err := mf.Fd.Stat() + if err != nil { + mf.Close(-1) + return nil, y.Wrap(err, "") + } + + filename := fileInfo.Name() + id, ok := ParseFileID(filename) + if !ok { + mf.Close(-1) + return nil, errors.Errorf("Invalid filename: %s", filename) + } + t := &Table{ + MmapFile: mf, + id: id, + opt: &opts, + IsInmemory: false, + tableSize: int(fileInfo.Size()), + CreatedAt: fileInfo.ModTime(), + } + // Caller is given one reference. + t.ref.Store(1) + + if err := t.initBiggestAndSmallest(); err != nil { + return nil, y.Wrapf(err, "failed to initialize table") + } + + if opts.ChkMode == options.OnTableRead || opts.ChkMode == options.OnTableAndBlockRead { + if err := t.VerifyChecksum(); err != nil { + mf.Close(-1) + return nil, y.Wrapf(err, "failed to verify checksum") + } + } + + return t, nil +} + +// OpenInMemoryTable is similar to OpenTable but it opens a new table from the provided data. +// OpenInMemoryTable is used for L0 tables. +func OpenInMemoryTable(data []byte, id uint64, opt *Options) (*Table, error) { + mf := &z.MmapFile{ + Data: data, + Fd: nil, + } + t := &Table{ + MmapFile: mf, + opt: opt, + tableSize: len(data), + IsInmemory: true, + id: id, // It is important that each table gets a unique ID. + } + // Caller is given one reference. + t.ref.Store(1) + + if err := t.initBiggestAndSmallest(); err != nil { + return nil, err + } + return t, nil +} + +func (t *Table) initBiggestAndSmallest() error { + // This defer will help gathering debugging info incase initIndex crashes. + defer func() { + if r := recover(); r != nil { + // Use defer for printing info because there may be an intermediate panic. + var debugBuf bytes.Buffer + defer func() { + panic(fmt.Sprintf("%s\n== Recovered ==\n", debugBuf.String())) + }() + + // Get the count of null bytes at the end of file. This is to make sure if there was an + // issue with mmap sync or file copy. + count := 0 + for i := len(t.Data) - 1; i >= 0; i-- { + if t.Data[i] != 0 { + break + } + count++ + } + + fmt.Fprintf(&debugBuf, "\n== Recovering from initIndex crash ==\n") + fmt.Fprintf(&debugBuf, "File Info: [ID: %d, Size: %d, Zeros: %d]\n", + t.id, t.tableSize, count) + + fmt.Fprintf(&debugBuf, "isEnrypted: %v ", t.shouldDecrypt()) + + readPos := t.tableSize + + // Read checksum size. + readPos -= 4 + buf := t.readNoFail(readPos, 4) + checksumLen := int(y.BytesToU32(buf)) + fmt.Fprintf(&debugBuf, "checksumLen: %d ", checksumLen) + + // Read checksum. + checksum := &pb.Checksum{} + readPos -= checksumLen + buf = t.readNoFail(readPos, checksumLen) + _ = proto.Unmarshal(buf, checksum) + fmt.Fprintf(&debugBuf, "checksum: %+v ", checksum) + + // Read index size from the footer. + readPos -= 4 + buf = t.readNoFail(readPos, 4) + indexLen := int(y.BytesToU32(buf)) + fmt.Fprintf(&debugBuf, "indexLen: %d ", indexLen) + + // Read index. + readPos -= t.indexLen + t.indexStart = readPos + indexData := t.readNoFail(readPos, t.indexLen) + fmt.Fprintf(&debugBuf, "index: %v ", indexData) + } + }() + + var err error + var ko *fb.BlockOffset + if ko, err = t.initIndex(); err != nil { + return y.Wrapf(err, "failed to read index.") + } + + t.smallest = y.Copy(ko.KeyBytes()) + + it2 := t.NewIterator(REVERSED | NOCACHE) + defer it2.Close() + it2.Rewind() + if !it2.Valid() { + return y.Wrapf(it2.err, "failed to initialize biggest for table %s", t.Filename()) + } + t.biggest = y.Copy(it2.Key()) + return nil +} + +func (t *Table) read(off, sz int) ([]byte, error) { + return t.Bytes(off, sz) +} + +func (t *Table) readNoFail(off, sz int) []byte { + res, err := t.read(off, sz) + y.Check(err) + return res +} + +// initIndex reads the index and populate the necessary table fields and returns +// first block offset +func (t *Table) initIndex() (*fb.BlockOffset, error) { + readPos := t.tableSize + + // Read checksum len from the last 4 bytes. + readPos -= 4 + buf := t.readNoFail(readPos, 4) + checksumLen := int(y.BytesToU32(buf)) + if checksumLen < 0 { + return nil, errors.New("checksum length less than zero. Data corrupted") + } + + // Read checksum. + expectedChk := &pb.Checksum{} + readPos -= checksumLen + buf = t.readNoFail(readPos, checksumLen) + if err := proto.Unmarshal(buf, expectedChk); err != nil { + return nil, err + } + + // Read index size from the footer. + readPos -= 4 + buf = t.readNoFail(readPos, 4) + t.indexLen = int(y.BytesToU32(buf)) + + // Read index. + readPos -= t.indexLen + t.indexStart = readPos + data := t.readNoFail(readPos, t.indexLen) + + if err := y.VerifyChecksum(data, expectedChk); err != nil { + return nil, y.Wrapf(err, "failed to verify checksum for table: %s", t.Filename()) + } + + index, err := t.readTableIndex() + if err != nil { + return nil, err + } + if !t.shouldDecrypt() { + // If there's no encryption, this points to the mmap'ed buffer. + t._index = index + } + t._cheap = &cheapIndex{ + MaxVersion: index.MaxVersion(), + KeyCount: index.KeyCount(), + UncompressedSize: index.UncompressedSize(), + OnDiskSize: index.OnDiskSize(), + OffsetsLength: index.OffsetsLength(), + BloomFilterLength: index.BloomFilterLength(), + } + + t.hasBloomFilter = len(index.BloomFilterBytes()) > 0 + + var bo fb.BlockOffset + y.AssertTrue(index.Offsets(&bo, 0)) + return &bo, nil +} + +// KeySplits splits the table into at least n ranges based on the block offsets. +func (t *Table) KeySplits(n int, prefix []byte) []string { + if n == 0 { + return nil + } + + oLen := t.offsetsLength() + jump := oLen / n + if jump == 0 { + jump = 1 + } + + var bo fb.BlockOffset + var res []string + for i := 0; i < oLen; i += jump { + if i >= oLen { + i = oLen - 1 + } + y.AssertTrue(t.offsets(&bo, i)) + if bytes.HasPrefix(bo.KeyBytes(), prefix) { + res = append(res, string(bo.KeyBytes())) + } + } + return res +} + +func (t *Table) fetchIndex() *fb.TableIndex { + if !t.shouldDecrypt() { + return t._index + } + + if t.opt.IndexCache == nil { + panic("Index Cache must be set for encrypted workloads") + } + if val, ok := t.opt.IndexCache.Get(t.indexKey()); ok && val != nil { + return val.(*fb.TableIndex) + } + + index, err := t.readTableIndex() + y.Check(err) + t.opt.IndexCache.Set(t.indexKey(), index, int64(t.indexLen)) + return index +} + +func (t *Table) offsets(ko *fb.BlockOffset, i int) bool { + return t.fetchIndex().Offsets(ko, i) +} + +// block function return a new block. Each block holds a ref and the byte +// slice stored in the block will be reused when the ref becomes zero. The +// caller should release the block by calling block.decrRef() on it. +func (t *Table) block(idx int, useCache bool) (*block, error) { + y.AssertTruef(idx >= 0, "idx=%d", idx) + if idx >= t.offsetsLength() { + return nil, errors.New("block out of index") + } + if t.opt.BlockCache != nil { + key := t.blockCacheKey(idx) + blk, ok := t.opt.BlockCache.Get(key) + if ok && blk != nil { + // Use the block only if the increment was successful. The block + // could get evicted from the cache between the Get() call and the + // incrRef() call. + if b := blk.(*block); b.incrRef() { + return b, nil + } + } + } + + var ko fb.BlockOffset + y.AssertTrue(t.offsets(&ko, idx)) + blk := &block{offset: int(ko.Offset())} + blk.ref.Store(1) + defer blk.decrRef() // Deal with any errors, where blk would not be returned. + NumBlocks.Add(1) + + var err error + if blk.data, err = t.read(blk.offset, int(ko.Len())); err != nil { + return nil, y.Wrapf(err, + "failed to read from file: %s at offset: %d, len: %d", + t.Fd.Name(), blk.offset, ko.Len()) + } + + if t.shouldDecrypt() { + // Decrypt the block if it is encrypted. + if blk.data, err = t.decrypt(blk.data, true); err != nil { + return nil, err + } + // blk.data is allocated via Calloc. So, do free. + blk.freeMe = true + } + + if err = t.decompress(blk); err != nil { + return nil, y.Wrapf(err, + "failed to decode compressed data in file: %s at offset: %d, len: %d", + t.Fd.Name(), blk.offset, ko.Len()) + } + + // Read meta data related to block. + readPos := len(blk.data) - 4 // First read checksum length. + blk.chkLen = int(y.BytesToU32(blk.data[readPos : readPos+4])) + + // Checksum length greater than block size could happen if the table was compressed and + // it was opened with an incorrect compression algorithm (or the data was corrupted). + if blk.chkLen > len(blk.data) { + return nil, errors.New("invalid checksum length. Either the data is " + + "corrupted or the table options are incorrectly set") + } + + // Read checksum and store it + readPos -= blk.chkLen + blk.checksum = blk.data[readPos : readPos+blk.chkLen] + // Move back and read numEntries in the block. + readPos -= 4 + numEntries := int(y.BytesToU32(blk.data[readPos : readPos+4])) + entriesIndexStart := readPos - (numEntries * 4) + entriesIndexEnd := entriesIndexStart + numEntries*4 + + blk.entryOffsets = y.BytesToU32Slice(blk.data[entriesIndexStart:entriesIndexEnd]) + + blk.entriesIndexStart = entriesIndexStart + + // Drop checksum and checksum length. + // The checksum is calculated for actual data + entry index + index length + blk.data = blk.data[:readPos+4] + + // Verify checksum on if checksum verification mode is OnRead on OnStartAndRead. + if t.opt.ChkMode == options.OnBlockRead || t.opt.ChkMode == options.OnTableAndBlockRead { + if err = blk.verifyCheckSum(); err != nil { + return nil, err + } + } + + blk.incrRef() + if useCache && t.opt.BlockCache != nil { + key := t.blockCacheKey(idx) + // incrRef should never return false here because we're calling it on a + // new block with ref=1. + y.AssertTrue(blk.incrRef()) + + // Decrement the block ref if we could not insert it in the cache. + if !t.opt.BlockCache.Set(key, blk, blk.size()) { + blk.decrRef() + } + // We have added an OnReject func in our cache, which gets called in case the block is not + // admitted to the cache. So, every block would be accounted for. + } + return blk, nil +} + +// blockCacheKey is used to store blocks in the block cache. +func (t *Table) blockCacheKey(idx int) []byte { + y.AssertTrue(t.id < math.MaxUint32) + y.AssertTrue(uint32(idx) < math.MaxUint32) + + buf := make([]byte, 8) + // Assume t.ID does not overflow uint32. + binary.BigEndian.PutUint32(buf[:4], uint32(t.ID())) + binary.BigEndian.PutUint32(buf[4:], uint32(idx)) + return buf +} + +// indexKey returns the cache key for block offsets. blockOffsets +// are stored in the index cache. +func (t *Table) indexKey() uint64 { + return t.id +} + +// IndexSize is the size of table index in bytes. +func (t *Table) IndexSize() int { + return t.indexLen +} + +// Size is its file size in bytes +func (t *Table) Size() int64 { return int64(t.tableSize) } + +// StaleDataSize is the amount of stale data (that can be dropped by a compaction )in this SST. +func (t *Table) StaleDataSize() uint32 { return t.fetchIndex().StaleDataSize() } + +// Smallest is its smallest key, or nil if there are none +func (t *Table) Smallest() []byte { return t.smallest } + +// Biggest is its biggest key, or nil if there are none +func (t *Table) Biggest() []byte { return t.biggest } + +// Filename is NOT the file name. Just kidding, it is. +func (t *Table) Filename() string { return t.Fd.Name() } + +// ID is the table's ID number (used to make the file name). +func (t *Table) ID() uint64 { return t.id } + +// DoesNotHave returns true if and only if the table does not have the key hash. +// It does a bloom filter lookup. +func (t *Table) DoesNotHave(hash uint32) bool { + if !t.hasBloomFilter { + return false + } + + y.NumLSMBloomHitsAdd(t.opt.MetricsEnabled, "DoesNotHave_ALL", 1) + index := t.fetchIndex() + bf := index.BloomFilterBytes() + mayContain := y.Filter(bf).MayContain(hash) + if !mayContain { + y.NumLSMBloomHitsAdd(t.opt.MetricsEnabled, "DoesNotHave_HIT", 1) + } + return !mayContain +} + +// readTableIndex reads table index from the sst and returns its pb format. +func (t *Table) readTableIndex() (*fb.TableIndex, error) { + data := t.readNoFail(t.indexStart, t.indexLen) + var err error + // Decrypt the table index if it is encrypted. + if t.shouldDecrypt() { + if data, err = t.decrypt(data, false); err != nil { + return nil, y.Wrapf(err, + "Error while decrypting table index for the table %d in readTableIndex", t.id) + } + } + return fb.GetRootAsTableIndex(data, 0), nil +} + +// VerifyChecksum verifies checksum for all blocks of table. This function is called by +// OpenTable() function. This function is also called inside levelsController.VerifyChecksum(). +func (t *Table) VerifyChecksum() error { + ti := t.fetchIndex() + for i := 0; i < ti.OffsetsLength(); i++ { + b, err := t.block(i, true) + if err != nil { + return y.Wrapf(err, "checksum validation failed for table: %s, block: %d, offset:%d", + t.Filename(), i, b.offset) + } + // We should not call incrRef here, because the block already has one ref when created. + defer b.decrRef() + // OnBlockRead or OnTableAndBlockRead, we don't need to call verify checksum + // on block, verification would be done while reading block itself. + if !(t.opt.ChkMode == options.OnBlockRead || t.opt.ChkMode == options.OnTableAndBlockRead) { + if err = b.verifyCheckSum(); err != nil { + return y.Wrapf(err, + "checksum validation failed for table: %s, block: %d, offset:%d", + t.Filename(), i, b.offset) + } + } + } + return nil +} + +// shouldDecrypt tells whether to decrypt or not. We decrypt only if the datakey exist +// for the table. +func (t *Table) shouldDecrypt() bool { + return t.opt.DataKey != nil +} + +// KeyID returns data key id. +func (t *Table) KeyID() uint64 { + if t.opt.DataKey != nil { + return t.opt.DataKey.KeyId + } + // By default it's 0, if it is plain text. + return 0 +} + +// decrypt decrypts the given data. It should be called only after checking shouldDecrypt. +func (t *Table) decrypt(data []byte, viaCalloc bool) ([]byte, error) { + // Last BlockSize bytes of the data is the IV. + iv := data[len(data)-aes.BlockSize:] + // Rest all bytes are data. + data = data[:len(data)-aes.BlockSize] + + var dst []byte + if viaCalloc { + dst = z.Calloc(len(data), "Table.Decrypt") + } else { + dst = make([]byte, len(data)) + } + if err := y.XORBlock(dst, data, t.opt.DataKey.Data, iv); err != nil { + return nil, y.Wrapf(err, "while decrypt") + } + return dst, nil +} + +// ParseFileID reads the file id out of a filename. +func ParseFileID(name string) (uint64, bool) { + name = filepath.Base(name) + if !strings.HasSuffix(name, fileSuffix) { + return 0, false + } + // suffix := name[len(fileSuffix):] + name = strings.TrimSuffix(name, fileSuffix) + id, err := strconv.Atoi(name) + if err != nil { + return 0, false + } + y.AssertTrue(id >= 0) + return uint64(id), true +} + +// IDToFilename does the inverse of ParseFileID +func IDToFilename(id uint64) string { + return fmt.Sprintf("%06d", id) + fileSuffix +} + +// NewFilename should be named TableFilepath -- it combines the dir with the ID to make a table +// filepath. +func NewFilename(id uint64, dir string) string { + return filepath.Join(dir, IDToFilename(id)) +} + +// decompress decompresses the data stored in a block. +func (t *Table) decompress(b *block) error { + var dst []byte + var err error + + // Point to the original b.data + src := b.data + + switch t.opt.Compression { + case options.None: + // Nothing to be done here. + return nil + case options.Snappy: + if sz, err := snappy.DecodedLen(b.data); err == nil { + dst = z.Calloc(sz, "Table.Decompress") + } else { + dst = z.Calloc(len(b.data)*4, "Table.Decompress") // Take a guess. + } + b.data, err = snappy.Decode(dst, b.data) + if err != nil { + z.Free(dst) + return y.Wrap(err, "failed to decompress") + } + case options.ZSTD: + sz := int(float64(t.opt.BlockSize) * 1.2) + dst = z.Calloc(sz, "Table.Decompress") + b.data, err = y.ZSTDDecompress(dst, b.data) + if err != nil { + z.Free(dst) + return y.Wrap(err, "failed to decompress") + } + default: + return errors.New("Unsupported compression type") + } + + if b.freeMe { + z.Free(src) + b.freeMe = false + } + + if len(b.data) > 0 && len(dst) > 0 && &dst[0] != &b.data[0] { + z.Free(dst) + } else { + b.freeMe = true + } + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/test.sh b/vendor/github.com/dgraph-io/badger/v4/test.sh new file mode 100644 index 00000000000..53b5477aa6d --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/test.sh @@ -0,0 +1,110 @@ +#!/bin/bash + +set -eo pipefail + +go version + +# Check if Github Actions is running +if [ $CI = "true" ]; then + # Enable code coverage + # export because tests run in a subprocess + export covermode="-covermode=atomic" + export coverprofile="-coverprofile=cover_tmp.out" + echo "mode: atomic" >> cover.out +fi + +# Run `go list` BEFORE setting GOFLAGS so that the output is in the right +# format for grep. +# export packages because the test will run in a sub process. +export packages=$(go list ./... | grep "github.com/dgraph-io/badger/v4/") + +tags="-tags=jemalloc" + +# Compile the Badger binary +pushd badger +go build -v $tags . +popd + +# Run the memory intensive tests first. +manual() { + timeout="-timeout 2m" + echo "==> Running package tests for $packages" + set -e + for pkg in $packages; do + echo "===> Testing $pkg" + go test $tags -timeout=25m $covermode $coverprofile -failfast -race -parallel 16 $pkg && write_coverage || return 1 + done + echo "==> DONE package tests" + + echo "==> Running manual tests" + # Run the special Truncate test. + rm -rf p + set -e + go test $tags $timeout $covermode $coverprofile -run='TestTruncateVlogNoClose$' -failfast --manual=true && write_coverage || return 1 + truncate --size=4096 p/000000.vlog + go test $tags $timeout $covermode $coverprofile -run='TestTruncateVlogNoClose2$' -failfast --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -run='TestTruncateVlogNoClose3$' -failfast --manual=true && write_coverage || return 1 + rm -rf p + + # TODO(ibrahim): Let's make these tests have Manual prefix. + # go test $tags -run='TestManual' --manual=true --parallel=2 + # TestWriteBatch + # TestValueGCManaged + # TestDropPrefix + # TestDropAllManaged + go test $tags $timeout $covermode $coverprofile -failfast -run='TestBigKeyValuePairs$' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestPushValueLogLimit' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestKeyCount' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestIteratePrefix' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestIterateParallel' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestBigStream' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestGoroutineLeak' --manual=true && write_coverage || return 1 + go test $tags $timeout $covermode $coverprofile -failfast -run='TestGetMore' --manual=true && write_coverage || return 1 + + echo "==> DONE manual tests" +} + +root() { + # Run the normal tests. + # go test -timeout=25m -v -race github.com/dgraph-io/badger/v4/... + + echo "==> Running root level tests." + go test $tags -v -race -parallel=16 -timeout=25m -failfast $covermode $coverprofile . && write_coverage || return 1 + echo "==> DONE root level tests" +} + +stream() { + set -eo pipefail + pushd badger + baseDir=$(mktemp -d -p .) + ./badger benchmark write -s --dir=$baseDir/test | tee $baseDir/log.txt + ./badger benchmark read --dir=$baseDir/test --full-scan | tee --append $baseDir/log.txt + ./badger benchmark read --dir=$baseDir/test -d=30s | tee --append $baseDir/log.txt + ./badger stream --dir=$baseDir/test -o "$baseDir/test2" | tee --append $baseDir/log.txt + count=$(cat "$baseDir/log.txt" | grep "at program end: 0 B" | wc -l) + rm -rf $baseDir + if [ $count -ne 4 ]; then + echo "LEAK detected in Badger stream." + return 1 + fi + echo "==> DONE stream test" + popd + return 0 +} + +write_coverage() { + if [[ $CI = "true" ]]; then + if [[ -f cover_tmp.out ]]; then + sed -i '1d' cover_tmp.out + cat cover_tmp.out >> cover.out && rm cover_tmp.out + fi + fi + +} + +# parallel tests currently not working +# parallel --halt now,fail=1 --progress --line-buffer ::: stream manual root +# run tests in sequence +root +stream +manual diff --git a/vendor/github.com/dgraph-io/badger/v4/test_extensions.go b/vendor/github.com/dgraph-io/badger/v4/test_extensions.go new file mode 100644 index 00000000000..f9c025b43a6 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/test_extensions.go @@ -0,0 +1,79 @@ +/* + * Copyright 2023 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +// Important: Do NOT import the "testing" package, as otherwise, that +// will pull in imports into the production class that we do not want. + +// TODO: Consider using this with specific compilation tags so that it only +// shows up when performing testing (e.g., specify build tag=unit). +// We are not yet ready to do that, as it may impact customer usage as +// well as requiring us to update the CI build flags. Moreover, the +// current model does not actually incur any significant cost. +// If we do this, we will also want to introduce a parallel file that +// overrides some of these structs and functions with empty contents. + +// String constants for messages to be pushed to syncChan. +const ( + updateDiscardStatsMsg = "updateDiscardStats iteration done" + endVLogInitMsg = "End: vlog.init(db)" +) + +// testOnlyOptions specifies an extension to the type Options that we want to +// use only in the context of testing. +type testOnlyOptions struct { + // syncChan is used to listen for specific messages related to activities + // that can occur in a DB instance. Currently, this is only used in + // testing activities. + syncChan chan string +} + +// testOnlyDBExtensions specifies an extension to the type DB that we want to +// use only in the context of testing. +type testOnlyDBExtensions struct { + syncChan chan string + + // onCloseDiscardCapture will be populated by a DB instance during the + // process of performing the Close operation. Currently, we only consider + // using this during testing. + onCloseDiscardCapture map[uint64]uint64 +} + +// logToSyncChan sends a message to the DB's syncChan. Note that we expect +// that the DB never closes this channel; the responsibility for +// allocating and closing the channel belongs to the test module. +// if db.syncChan is nil or has never been initialized, ths will be +// silently ignored. +func (db *DB) logToSyncChan(msg string) { + if db.syncChan != nil { + db.syncChan <- msg + } +} + +// captureDiscardStats will copy the contents of the discardStats file +// maintained by vlog to the onCloseDiscardCapture map specified by +// db.opt. Of couse, if db.opt.onCloseDiscardCapture is nil (as expected +// for a production system as opposed to a test system), this is a no-op. +func (db *DB) captureDiscardStats() { + if db.onCloseDiscardCapture != nil { + db.vlog.discardStats.Lock() + db.vlog.discardStats.Iterate(func(id, val uint64) { + db.onCloseDiscardCapture[id] = val + }) + db.vlog.discardStats.Unlock() + } +} diff --git a/vendor/github.com/dgraph-io/badger/v4/trie/trie.go b/vendor/github.com/dgraph-io/badger/v4/trie/trie.go new file mode 100644 index 00000000000..2c250c3b821 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/trie/trie.go @@ -0,0 +1,263 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package trie + +import ( + "fmt" + "strconv" + "strings" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/badger/v4/y" +) + +type node struct { + children map[byte]*node + ignore *node + ids []uint64 +} + +func (n *node) isEmpty() bool { + return len(n.children) == 0 && len(n.ids) == 0 && n.ignore == nil +} + +func newNode() *node { + return &node{ + children: make(map[byte]*node), + ids: []uint64{}, + } +} + +// Trie datastructure. +type Trie struct { + root *node +} + +// NewTrie returns Trie. +func NewTrie() *Trie { + return &Trie{ + root: newNode(), + } +} + +// parseIgnoreBytes would parse the ignore string, and convert it into a list of bools, where +// bool[idx] = true implies that key[idx] can be ignored during comparison. +func parseIgnoreBytes(ig string) ([]bool, error) { + var out []bool + if ig == "" { + return out, nil + } + + for _, each := range strings.Split(strings.TrimSpace(ig), ",") { + r := strings.Split(strings.TrimSpace(each), "-") + if len(r) == 0 || len(r) > 2 { + return out, fmt.Errorf("Invalid range: %s", each) + } + start, end := -1, -1 //nolint:ineffassign + if len(r) == 2 { + idx, err := strconv.Atoi(strings.TrimSpace(r[1])) + if err != nil { + return out, err + } + end = idx + } + { + // Always consider r[0] + idx, err := strconv.Atoi(strings.TrimSpace(r[0])) + if err != nil { + return out, err + } + start = idx + } + if start == -1 { + return out, fmt.Errorf("Invalid range: %s", each) + } + for start >= len(out) { + out = append(out, false) + } + for end >= len(out) { // end could be -1, so do have the start loop above. + out = append(out, false) + } + if end == -1 { + out[start] = true + } else { + for i := start; i <= end; i++ { + out[i] = true + } + } + } + return out, nil +} + +// Add adds the id in the trie for the given prefix path. +func (t *Trie) Add(prefix []byte, id uint64) { + m := pb.Match{ + Prefix: prefix, + } + y.Check(t.AddMatch(m, id)) +} + +// AddMatch allows you to send in a prefix match, with "holes" in the prefix. The holes are +// specified via IgnoreBytes in a comma-separated list of indices starting from 0. A dash can be +// used to denote a range. Valid example is "3, 5-8, 10, 12-15". Length of IgnoreBytes does not need +// to match the length of the Prefix passed. +// +// Consider a prefix = "aaaa". If the IgnoreBytes is set to "0, 2", then along with key "aaaa...", +// a key "baba..." would also match. +func (t *Trie) AddMatch(m pb.Match, id uint64) error { + return t.fix(m, id, set) +} + +const ( + set = iota + del +) + +func (t *Trie) fix(m pb.Match, id uint64, op int) error { + curNode := t.root + + ignore, err := parseIgnoreBytes(m.IgnoreBytes) + if err != nil { + return errors.Wrapf(err, "while parsing ignore bytes: %s", m.IgnoreBytes) + } + for len(ignore) < len(m.Prefix) { + ignore = append(ignore, false) + } + for idx, byt := range m.Prefix { + var child *node + if ignore[idx] { + child = curNode.ignore + if child == nil { + if op == del { + // No valid node found for delete operation. Return immediately. + return nil + } + child = newNode() + curNode.ignore = child + } + } else { + child = curNode.children[byt] + if child == nil { + if op == del { + // No valid node found for delete operation. Return immediately. + return nil + } + child = newNode() + curNode.children[byt] = child + } + } + curNode = child + } + + // We only need to add the id to the last node of the given prefix. + if op == set { + curNode.ids = append(curNode.ids, id) + + } else if op == del { + out := curNode.ids[:0] + for _, cid := range curNode.ids { + if id != cid { + out = append(out, cid) + } + } + curNode.ids = out + } else { + y.AssertTrue(false) + } + return nil +} + +func (t *Trie) Get(key []byte) map[uint64]struct{} { + return t.get(t.root, key) +} + +// Get returns prefix matched ids for the given key. +func (t *Trie) get(curNode *node, key []byte) map[uint64]struct{} { + y.AssertTrue(curNode != nil) + + out := make(map[uint64]struct{}) + // If any node in the path of the key has ids, pick them up. + // This would also match nil prefixes. + for _, i := range curNode.ids { + out[i] = struct{}{} + } + if len(key) == 0 { + return out + } + + // If we found an ignore node, traverse that path. + if curNode.ignore != nil { + res := t.get(curNode.ignore, key[1:]) + for id := range res { + out[id] = struct{}{} + } + } + + if child := curNode.children[key[0]]; child != nil { + res := t.get(child, key[1:]) + for id := range res { + out[id] = struct{}{} + } + } + return out +} + +func removeEmpty(curNode *node) bool { + // Go depth first. + if curNode.ignore != nil { + if empty := removeEmpty(curNode.ignore); empty { + curNode.ignore = nil + } + } + + for byt, n := range curNode.children { + if empty := removeEmpty(n); empty { + delete(curNode.children, byt) + } + } + + return curNode.isEmpty() +} + +// Delete will delete the id if the id exist in the given index path. +func (t *Trie) Delete(prefix []byte, id uint64) error { + return t.DeleteMatch(pb.Match{Prefix: prefix}, id) +} + +func (t *Trie) DeleteMatch(m pb.Match, id uint64) error { + if err := t.fix(m, id, del); err != nil { + return err + } + // Would recursively delete empty nodes. + // Do not remove the t.root even if its empty. + removeEmpty(t.root) + return nil +} + +func numNodes(curNode *node) int { + if curNode == nil { + return 0 + } + + num := numNodes(curNode.ignore) + for _, n := range curNode.children { + num += numNodes(n) + } + return num + 1 +} diff --git a/vendor/github.com/dgraph-io/badger/v4/txn.go b/vendor/github.com/dgraph-io/badger/v4/txn.go new file mode 100644 index 00000000000..4a5fe476c4e --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/txn.go @@ -0,0 +1,828 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "context" + "encoding/hex" + "math" + "sort" + "strconv" + "sync" + "sync/atomic" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +type oracle struct { + isManaged bool // Does not change value, so no locking required. + detectConflicts bool // Determines if the txns should be checked for conflicts. + + sync.Mutex // For nextTxnTs and commits. + // writeChLock lock is for ensuring that transactions go to the write + // channel in the same order as their commit timestamps. + writeChLock sync.Mutex + nextTxnTs uint64 + + // Used to block NewTransaction, so all previous commits are visible to a new read. + txnMark *y.WaterMark + + // Either of these is used to determine which versions can be permanently + // discarded during compaction. + discardTs uint64 // Used by ManagedDB. + readMark *y.WaterMark // Used by DB. + + // committedTxns contains all committed writes (contains fingerprints + // of keys written and their latest commit counter). + committedTxns []committedTxn + lastCleanupTs uint64 + + // closer is used to stop watermarks. + closer *z.Closer +} + +type committedTxn struct { + ts uint64 + // ConflictKeys Keeps track of the entries written at timestamp ts. + conflictKeys map[uint64]struct{} +} + +func newOracle(opt Options) *oracle { + orc := &oracle{ + isManaged: opt.managedTxns, + detectConflicts: opt.DetectConflicts, + // We're not initializing nextTxnTs and readOnlyTs. It would be done after replay in Open. + // + // WaterMarks must be 64-bit aligned for atomic package, hence we must use pointers here. + // See https://golang.org/pkg/sync/atomic/#pkg-note-BUG. + readMark: &y.WaterMark{Name: "badger.PendingReads"}, + txnMark: &y.WaterMark{Name: "badger.TxnTimestamp"}, + closer: z.NewCloser(2), + } + orc.readMark.Init(orc.closer) + orc.txnMark.Init(orc.closer) + return orc +} + +func (o *oracle) Stop() { + o.closer.SignalAndWait() +} + +func (o *oracle) readTs() uint64 { + if o.isManaged { + panic("ReadTs should not be retrieved for managed DB") + } + + var readTs uint64 + o.Lock() + readTs = o.nextTxnTs - 1 + o.readMark.Begin(readTs) + o.Unlock() + + // Wait for all txns which have no conflicts, have been assigned a commit + // timestamp and are going through the write to value log and LSM tree + // process. Not waiting here could mean that some txns which have been + // committed would not be read. + y.Check(o.txnMark.WaitForMark(context.Background(), readTs)) + return readTs +} + +func (o *oracle) nextTs() uint64 { + o.Lock() + defer o.Unlock() + return o.nextTxnTs +} + +func (o *oracle) incrementNextTs() { + o.Lock() + defer o.Unlock() + o.nextTxnTs++ +} + +// Any deleted or invalid versions at or below ts would be discarded during +// compaction to reclaim disk space in LSM tree and thence value log. +func (o *oracle) setDiscardTs(ts uint64) { + o.Lock() + defer o.Unlock() + o.discardTs = ts + o.cleanupCommittedTransactions() +} + +func (o *oracle) discardAtOrBelow() uint64 { + if o.isManaged { + o.Lock() + defer o.Unlock() + return o.discardTs + } + return o.readMark.DoneUntil() +} + +// hasConflict must be called while having a lock. +func (o *oracle) hasConflict(txn *Txn) bool { + if len(txn.reads) == 0 { + return false + } + for _, committedTxn := range o.committedTxns { + // If the committedTxn.ts is less than txn.readTs that implies that the + // committedTxn finished before the current transaction started. + // We don't need to check for conflict in that case. + // This change assumes linearizability. Lack of linearizability could + // cause the read ts of a new txn to be lower than the commit ts of + // a txn before it (@mrjn). + if committedTxn.ts <= txn.readTs { + continue + } + + for _, ro := range txn.reads { + if _, has := committedTxn.conflictKeys[ro]; has { + return true + } + } + } + + return false +} + +func (o *oracle) newCommitTs(txn *Txn) (uint64, bool) { + o.Lock() + defer o.Unlock() + + if o.hasConflict(txn) { + return 0, true + } + + var ts uint64 + if !o.isManaged { + o.doneRead(txn) + o.cleanupCommittedTransactions() + + // This is the general case, when user doesn't specify the read and commit ts. + ts = o.nextTxnTs + o.nextTxnTs++ + o.txnMark.Begin(ts) + + } else { + // If commitTs is set, use it instead. + ts = txn.commitTs + } + + y.AssertTrue(ts >= o.lastCleanupTs) + + if o.detectConflicts { + // We should ensure that txns are not added to o.committedTxns slice when + // conflict detection is disabled otherwise this slice would keep growing. + o.committedTxns = append(o.committedTxns, committedTxn{ + ts: ts, + conflictKeys: txn.conflictKeys, + }) + } + + return ts, false +} + +func (o *oracle) doneRead(txn *Txn) { + if !txn.doneRead { + txn.doneRead = true + o.readMark.Done(txn.readTs) + } +} + +func (o *oracle) cleanupCommittedTransactions() { // Must be called under o.Lock + if !o.detectConflicts { + // When detectConflicts is set to false, we do not store any + // committedTxns and so there's nothing to clean up. + return + } + // Same logic as discardAtOrBelow but unlocked + var maxReadTs uint64 + if o.isManaged { + maxReadTs = o.discardTs + } else { + maxReadTs = o.readMark.DoneUntil() + } + + y.AssertTrue(maxReadTs >= o.lastCleanupTs) + + // do not run clean up if the maxReadTs (read timestamp of the + // oldest transaction that is still in flight) has not increased + if maxReadTs == o.lastCleanupTs { + return + } + o.lastCleanupTs = maxReadTs + + tmp := o.committedTxns[:0] + for _, txn := range o.committedTxns { + if txn.ts <= maxReadTs { + continue + } + tmp = append(tmp, txn) + } + o.committedTxns = tmp +} + +func (o *oracle) doneCommit(cts uint64) { + if o.isManaged { + // No need to update anything. + return + } + o.txnMark.Done(cts) +} + +// Txn represents a Badger transaction. +type Txn struct { + readTs uint64 + commitTs uint64 + size int64 + count int64 + db *DB + + reads []uint64 // contains fingerprints of keys read. + // contains fingerprints of keys written. This is used for conflict detection. + conflictKeys map[uint64]struct{} + readsLock sync.Mutex // guards the reads slice. See addReadKey. + + pendingWrites map[string]*Entry // cache stores any writes done by txn. + duplicateWrites []*Entry // Used in managed mode to store duplicate entries. + + numIterators atomic.Int32 + discarded bool + doneRead bool + update bool // update is used to conditionally keep track of reads. +} + +type pendingWritesIterator struct { + entries []*Entry + nextIdx int + readTs uint64 + reversed bool +} + +func (pi *pendingWritesIterator) Next() { + pi.nextIdx++ +} + +func (pi *pendingWritesIterator) Rewind() { + pi.nextIdx = 0 +} + +func (pi *pendingWritesIterator) Seek(key []byte) { + key = y.ParseKey(key) + pi.nextIdx = sort.Search(len(pi.entries), func(idx int) bool { + cmp := bytes.Compare(pi.entries[idx].Key, key) + if !pi.reversed { + return cmp >= 0 + } + return cmp <= 0 + }) +} + +func (pi *pendingWritesIterator) Key() []byte { + y.AssertTrue(pi.Valid()) + entry := pi.entries[pi.nextIdx] + return y.KeyWithTs(entry.Key, pi.readTs) +} + +func (pi *pendingWritesIterator) Value() y.ValueStruct { + y.AssertTrue(pi.Valid()) + entry := pi.entries[pi.nextIdx] + return y.ValueStruct{ + Value: entry.Value, + Meta: entry.meta, + UserMeta: entry.UserMeta, + ExpiresAt: entry.ExpiresAt, + Version: pi.readTs, + } +} + +func (pi *pendingWritesIterator) Valid() bool { + return pi.nextIdx < len(pi.entries) +} + +func (pi *pendingWritesIterator) Close() error { + return nil +} + +func (txn *Txn) newPendingWritesIterator(reversed bool) *pendingWritesIterator { + if !txn.update || len(txn.pendingWrites) == 0 { + return nil + } + entries := make([]*Entry, 0, len(txn.pendingWrites)) + for _, e := range txn.pendingWrites { + entries = append(entries, e) + } + // Number of pending writes per transaction shouldn't be too big in general. + sort.Slice(entries, func(i, j int) bool { + cmp := bytes.Compare(entries[i].Key, entries[j].Key) + if !reversed { + return cmp < 0 + } + return cmp > 0 + }) + return &pendingWritesIterator{ + readTs: txn.readTs, + entries: entries, + reversed: reversed, + } +} + +func (txn *Txn) checkSize(e *Entry) error { + count := txn.count + 1 + // Extra bytes for the version in key. + size := txn.size + e.estimateSizeAndSetThreshold(txn.db.valueThreshold()) + 10 + if count >= txn.db.opt.maxBatchCount || size >= txn.db.opt.maxBatchSize { + return ErrTxnTooBig + } + txn.count, txn.size = count, size + return nil +} + +func exceedsSize(prefix string, max int64, key []byte) error { + return errors.Errorf("%s with size %d exceeded %d limit. %s:\n%s", + prefix, len(key), max, prefix, hex.Dump(key[:1<<10])) +} + +func (txn *Txn) modify(e *Entry) error { + const maxKeySize = 65000 + + switch { + case !txn.update: + return ErrReadOnlyTxn + case txn.discarded: + return ErrDiscardedTxn + case len(e.Key) == 0: + return ErrEmptyKey + case bytes.HasPrefix(e.Key, badgerPrefix): + return ErrInvalidKey + case len(e.Key) > maxKeySize: + // Key length can't be more than uint16, as determined by table::header. To + // keep things safe and allow badger move prefix and a timestamp suffix, let's + // cut it down to 65000, instead of using 65536. + return exceedsSize("Key", maxKeySize, e.Key) + case int64(len(e.Value)) > txn.db.opt.ValueLogFileSize: + return exceedsSize("Value", txn.db.opt.ValueLogFileSize, e.Value) + case txn.db.opt.InMemory && int64(len(e.Value)) > txn.db.valueThreshold(): + return exceedsSize("Value", txn.db.valueThreshold(), e.Value) + } + + if err := txn.db.isBanned(e.Key); err != nil { + return err + } + + if err := txn.checkSize(e); err != nil { + return err + } + + // The txn.conflictKeys is used for conflict detection. If conflict detection + // is disabled, we don't need to store key hashes in this map. + if txn.db.opt.DetectConflicts { + fp := z.MemHash(e.Key) // Avoid dealing with byte arrays. + txn.conflictKeys[fp] = struct{}{} + } + // If a duplicate entry was inserted in managed mode, move it to the duplicate writes slice. + // Add the entry to duplicateWrites only if both the entries have different versions. For + // same versions, we will overwrite the existing entry. + if oldEntry, ok := txn.pendingWrites[string(e.Key)]; ok && oldEntry.version != e.version { + txn.duplicateWrites = append(txn.duplicateWrites, oldEntry) + } + txn.pendingWrites[string(e.Key)] = e + return nil +} + +// Set adds a key-value pair to the database. +// It will return ErrReadOnlyTxn if update flag was set to false when creating the transaction. +// +// The current transaction keeps a reference to the key and val byte slice +// arguments. Users must not modify key and val until the end of the transaction. +func (txn *Txn) Set(key, val []byte) error { + return txn.SetEntry(NewEntry(key, val)) +} + +// SetEntry takes an Entry struct and adds the key-value pair in the struct, +// along with other metadata to the database. +// +// The current transaction keeps a reference to the entry passed in argument. +// Users must not modify the entry until the end of the transaction. +func (txn *Txn) SetEntry(e *Entry) error { + return txn.modify(e) +} + +// Delete deletes a key. +// +// This is done by adding a delete marker for the key at commit timestamp. Any +// reads happening before this timestamp would be unaffected. Any reads after +// this commit would see the deletion. +// +// The current transaction keeps a reference to the key byte slice argument. +// Users must not modify the key until the end of the transaction. +func (txn *Txn) Delete(key []byte) error { + e := &Entry{ + Key: key, + meta: bitDelete, + } + return txn.modify(e) +} + +// Get looks for key and returns corresponding Item. +// If key is not found, ErrKeyNotFound is returned. +func (txn *Txn) Get(key []byte) (item *Item, rerr error) { + if len(key) == 0 { + return nil, ErrEmptyKey + } else if txn.discarded { + return nil, ErrDiscardedTxn + } + + if err := txn.db.isBanned(key); err != nil { + return nil, err + } + + item = new(Item) + if txn.update { + if e, has := txn.pendingWrites[string(key)]; has && bytes.Equal(key, e.Key) { + if isDeletedOrExpired(e.meta, e.ExpiresAt) { + return nil, ErrKeyNotFound + } + // Fulfill from cache. + item.meta = e.meta + item.val = e.Value + item.userMeta = e.UserMeta + item.key = key + item.status = prefetched + item.version = txn.readTs + item.expiresAt = e.ExpiresAt + // We probably don't need to set db on item here. + return item, nil + } + // Only track reads if this is update txn. No need to track read if txn serviced it + // internally. + txn.addReadKey(key) + } + + seek := y.KeyWithTs(key, txn.readTs) + vs, err := txn.db.get(seek) + if err != nil { + return nil, y.Wrapf(err, "DB::Get key: %q", key) + } + if vs.Value == nil && vs.Meta == 0 { + return nil, ErrKeyNotFound + } + if isDeletedOrExpired(vs.Meta, vs.ExpiresAt) { + return nil, ErrKeyNotFound + } + + item.key = key + item.version = vs.Version + item.meta = vs.Meta + item.userMeta = vs.UserMeta + item.vptr = y.SafeCopy(item.vptr, vs.Value) + item.txn = txn + item.expiresAt = vs.ExpiresAt + return item, nil +} + +func (txn *Txn) addReadKey(key []byte) { + if txn.update { + fp := z.MemHash(key) + + // Because of the possibility of multiple iterators it is now possible + // for multiple threads within a read-write transaction to read keys at + // the same time. The reads slice is not currently thread-safe and + // needs to be locked whenever we mark a key as read. + txn.readsLock.Lock() + txn.reads = append(txn.reads, fp) + txn.readsLock.Unlock() + } +} + +// Discard discards a created transaction. This method is very important and must be called. Commit +// method calls this internally, however, calling this multiple times doesn't cause any issues. So, +// this can safely be called via a defer right when transaction is created. +// +// NOTE: If any operations are run on a discarded transaction, ErrDiscardedTxn is returned. +func (txn *Txn) Discard() { + if txn.discarded { // Avoid a re-run. + return + } + if txn.numIterators.Load() > 0 { + panic("Unclosed iterator at time of Txn.Discard.") + } + txn.discarded = true + if !txn.db.orc.isManaged { + txn.db.orc.doneRead(txn) + } +} + +func (txn *Txn) commitAndSend() (func() error, error) { + orc := txn.db.orc + // Ensure that the order in which we get the commit timestamp is the same as + // the order in which we push these updates to the write channel. So, we + // acquire a writeChLock before getting a commit timestamp, and only release + // it after pushing the entries to it. + orc.writeChLock.Lock() + defer orc.writeChLock.Unlock() + + commitTs, conflict := orc.newCommitTs(txn) + if conflict { + return nil, ErrConflict + } + + keepTogether := true + setVersion := func(e *Entry) { + if e.version == 0 { + e.version = commitTs + } else { + keepTogether = false + } + } + for _, e := range txn.pendingWrites { + setVersion(e) + } + // The duplicateWrites slice will be non-empty only if there are duplicate + // entries with different versions. + for _, e := range txn.duplicateWrites { + setVersion(e) + } + + entries := make([]*Entry, 0, len(txn.pendingWrites)+len(txn.duplicateWrites)+1) + + processEntry := func(e *Entry) { + // Suffix the keys with commit ts, so the key versions are sorted in + // descending order of commit timestamp. + e.Key = y.KeyWithTs(e.Key, e.version) + // Add bitTxn only if these entries are part of a transaction. We + // support SetEntryAt(..) in managed mode which means a single + // transaction can have entries with different timestamps. If entries + // in a single transaction have different timestamps, we don't add the + // transaction markers. + if keepTogether { + e.meta |= bitTxn + } + entries = append(entries, e) + } + + // The following debug information is what led to determining the cause of + // bank txn violation bug, and it took a whole bunch of effort to narrow it + // down to here. So, keep this around for at least a couple of months. + // var b strings.Builder + // fmt.Fprintf(&b, "Read: %d. Commit: %d. reads: %v. writes: %v. Keys: ", + // txn.readTs, commitTs, txn.reads, txn.conflictKeys) + for _, e := range txn.pendingWrites { + processEntry(e) + } + for _, e := range txn.duplicateWrites { + processEntry(e) + } + + if keepTogether { + // CommitTs should not be zero if we're inserting transaction markers. + y.AssertTrue(commitTs != 0) + e := &Entry{ + Key: y.KeyWithTs(txnKey, commitTs), + Value: []byte(strconv.FormatUint(commitTs, 10)), + meta: bitFinTxn, + } + entries = append(entries, e) + } + + req, err := txn.db.sendToWriteCh(entries) + if err != nil { + orc.doneCommit(commitTs) + return nil, err + } + ret := func() error { + err := req.Wait() + // Wait before marking commitTs as done. + // We can't defer doneCommit above, because it is being called from a + // callback here. + orc.doneCommit(commitTs) + return err + } + return ret, nil +} + +func (txn *Txn) commitPrecheck() error { + if txn.discarded { + return errors.New("Trying to commit a discarded txn") + } + keepTogether := true + for _, e := range txn.pendingWrites { + if e.version != 0 { + keepTogether = false + } + } + + // If keepTogether is True, it implies transaction markers will be added. + // In that case, commitTs should not be never be zero. This might happen if + // someone uses txn.Commit instead of txn.CommitAt in managed mode. This + // should happen only in managed mode. In normal mode, keepTogether will + // always be true. + if keepTogether && txn.db.opt.managedTxns && txn.commitTs == 0 { + return errors.New("CommitTs cannot be zero. Please use commitAt instead") + } + return nil +} + +// Commit commits the transaction, following these steps: +// +// 1. If there are no writes, return immediately. +// +// 2. Check if read rows were updated since txn started. If so, return ErrConflict. +// +// 3. If no conflict, generate a commit timestamp and update written rows' commit ts. +// +// 4. Batch up all writes, write them to value log and LSM tree. +// +// 5. If callback is provided, Badger will return immediately after checking +// for conflicts. Writes to the database will happen in the background. If +// there is a conflict, an error will be returned and the callback will not +// run. If there are no conflicts, the callback will be called in the +// background upon successful completion of writes or any error during write. +// +// If error is nil, the transaction is successfully committed. In case of a non-nil error, the LSM +// tree won't be updated, so there's no need for any rollback. +func (txn *Txn) Commit() error { + // txn.conflictKeys can be zero if conflict detection is turned off. So we + // should check txn.pendingWrites. + if len(txn.pendingWrites) == 0 { + return nil // Nothing to do. + } + // Precheck before discarding txn. + if err := txn.commitPrecheck(); err != nil { + return err + } + defer txn.Discard() + + txnCb, err := txn.commitAndSend() + if err != nil { + return err + } + // If batchSet failed, LSM would not have been updated. So, no need to rollback anything. + + // TODO: What if some of the txns successfully make it to value log, but others fail. + // Nothing gets updated to LSM, until a restart happens. + return txnCb() +} + +type txnCb struct { + commit func() error + user func(error) + err error +} + +func runTxnCallback(cb *txnCb) { + switch { + case cb == nil: + panic("txn callback is nil") + case cb.user == nil: + panic("Must have caught a nil callback for txn.CommitWith") + case cb.err != nil: + cb.user(cb.err) + case cb.commit != nil: + err := cb.commit() + cb.user(err) + default: + cb.user(nil) + } +} + +// CommitWith acts like Commit, but takes a callback, which gets run via a +// goroutine to avoid blocking this function. The callback is guaranteed to run, +// so it is safe to increment sync.WaitGroup before calling CommitWith, and +// decrementing it in the callback; to block until all callbacks are run. +func (txn *Txn) CommitWith(cb func(error)) { + if cb == nil { + panic("Nil callback provided to CommitWith") + } + + if len(txn.pendingWrites) == 0 { + // Do not run these callbacks from here, because the CommitWith and the + // callback might be acquiring the same locks. Instead run the callback + // from another goroutine. + go runTxnCallback(&txnCb{user: cb, err: nil}) + return + } + + // Precheck before discarding txn. + if err := txn.commitPrecheck(); err != nil { + cb(err) + return + } + + defer txn.Discard() + + commitCb, err := txn.commitAndSend() + if err != nil { + go runTxnCallback(&txnCb{user: cb, err: err}) + return + } + + go runTxnCallback(&txnCb{user: cb, commit: commitCb}) +} + +// ReadTs returns the read timestamp of the transaction. +func (txn *Txn) ReadTs() uint64 { + return txn.readTs +} + +// NewTransaction creates a new transaction. Badger supports concurrent execution of transactions, +// providing serializable snapshot isolation, avoiding write skews. Badger achieves this by tracking +// the keys read and at Commit time, ensuring that these read keys weren't concurrently modified by +// another transaction. +// +// For read-only transactions, set update to false. In this mode, we don't track the rows read for +// any changes. Thus, any long running iterations done in this mode wouldn't pay this overhead. +// +// Running transactions concurrently is OK. However, a transaction itself isn't thread safe, and +// should only be run serially. It doesn't matter if a transaction is created by one goroutine and +// passed down to other, as long as the Txn APIs are called serially. +// +// When you create a new transaction, it is absolutely essential to call +// Discard(). This should be done irrespective of what the update param is set +// to. Commit API internally runs Discard, but running it twice wouldn't cause +// any issues. +// +// txn := db.NewTransaction(false) +// defer txn.Discard() +// // Call various APIs. +func (db *DB) NewTransaction(update bool) *Txn { + return db.newTransaction(update, false) +} + +func (db *DB) newTransaction(update, isManaged bool) *Txn { + if db.opt.ReadOnly && update { + // DB is read-only, force read-only transaction. + update = false + } + + txn := &Txn{ + update: update, + db: db, + count: 1, // One extra entry for BitFin. + size: int64(len(txnKey) + 10), // Some buffer for the extra entry. + } + if update { + if db.opt.DetectConflicts { + txn.conflictKeys = make(map[uint64]struct{}) + } + txn.pendingWrites = make(map[string]*Entry) + } + if !isManaged { + txn.readTs = db.orc.readTs() + } + return txn +} + +// View executes a function creating and managing a read-only transaction for the user. Error +// returned by the function is relayed by the View method. +// If View is used with managed transactions, it would assume a read timestamp of MaxUint64. +func (db *DB) View(fn func(txn *Txn) error) error { + if db.IsClosed() { + return ErrDBClosed + } + var txn *Txn + if db.opt.managedTxns { + txn = db.NewTransactionAt(math.MaxUint64, false) + } else { + txn = db.NewTransaction(false) + } + defer txn.Discard() + + return fn(txn) +} + +// Update executes a function, creating and managing a read-write transaction +// for the user. Error returned by the function is relayed by the Update method. +// Update cannot be used with managed transactions. +func (db *DB) Update(fn func(txn *Txn) error) error { + if db.IsClosed() { + return ErrDBClosed + } + if db.opt.managedTxns { + panic("Update can only be used with managedDB=false.") + } + txn := db.NewTransaction(true) + defer txn.Discard() + + if err := fn(txn); err != nil { + return err + } + + return txn.Commit() +} diff --git a/vendor/github.com/dgraph-io/badger/v4/util.go b/vendor/github.com/dgraph-io/badger/v4/util.go new file mode 100644 index 00000000000..8192594c27d --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/util.go @@ -0,0 +1,117 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "encoding/hex" + "math/rand" + "os" + "time" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/table" + "github.com/dgraph-io/badger/v4/y" +) + +func (s *levelsController) validate() error { + for _, l := range s.levels { + if err := l.validate(); err != nil { + return y.Wrap(err, "Levels Controller") + } + } + return nil +} + +// Check does some sanity check on one level of data or in-memory index. +func (s *levelHandler) validate() error { + if s.level == 0 { + return nil + } + + s.RLock() + defer s.RUnlock() + numTables := len(s.tables) + for j := 1; j < numTables; j++ { + if j >= len(s.tables) { + return errors.Errorf("Level %d, j=%d numTables=%d", s.level, j, numTables) + } + + if y.CompareKeys(s.tables[j-1].Biggest(), s.tables[j].Smallest()) >= 0 { + return errors.Errorf( + "Inter: Biggest(j-1)[%d] \n%s\n vs Smallest(j)[%d]: \n%s\n: "+ + "level=%d j=%d numTables=%d", + s.tables[j-1].ID(), hex.Dump(s.tables[j-1].Biggest()), s.tables[j].ID(), + hex.Dump(s.tables[j].Smallest()), s.level, j, numTables) + } + + if y.CompareKeys(s.tables[j].Smallest(), s.tables[j].Biggest()) > 0 { + return errors.Errorf( + "Intra: \n%s\n vs \n%s\n: level=%d j=%d numTables=%d", + hex.Dump(s.tables[j].Smallest()), hex.Dump(s.tables[j].Biggest()), s.level, j, numTables) + } + } + return nil +} + +// func (s *KV) debugPrintMore() { s.lc.debugPrintMore() } + +// // debugPrintMore shows key ranges of each level. +// func (s *levelsController) debugPrintMore() { +// s.Lock() +// defer s.Unlock() +// for i := 0; i < s.kv.opt.MaxLevels; i++ { +// s.levels[i].debugPrintMore() +// } +// } + +// func (s *levelHandler) debugPrintMore() { +// s.RLock() +// defer s.RUnlock() +// s.elog.Printf("Level %d:", s.level) +// for _, t := range s.tables { +// y.Printf(" [%s, %s]", t.Smallest(), t.Biggest()) +// } +// y.Printf("\n") +// } + +// reserveFileID reserves a unique file id. +func (s *levelsController) reserveFileID() uint64 { + id := s.nextFileID.Add(1) + return id - 1 +} + +func getIDMap(dir string) map[uint64]struct{} { + fileInfos, err := os.ReadDir(dir) + y.Check(err) + idMap := make(map[uint64]struct{}) + for _, info := range fileInfos { + if info.IsDir() { + continue + } + fileID, ok := table.ParseFileID(info.Name()) + if !ok { + continue + } + idMap[fileID] = struct{}{} + } + return idMap +} + +func init() { + rand.Seed(time.Now().UnixNano()) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/value.go b/vendor/github.com/dgraph-io/badger/v4/value.go new file mode 100644 index 00000000000..6b2f81016fa --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/value.go @@ -0,0 +1,1197 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package badger + +import ( + "bytes" + "context" + "fmt" + "hash" + "hash/crc32" + "io" + "math" + "os" + "sort" + "strconv" + "strings" + "sync" + "sync/atomic" + + "github.com/pkg/errors" + otrace "go.opencensus.io/trace" + + "github.com/dgraph-io/badger/v4/y" + "github.com/dgraph-io/ristretto/z" +) + +// maxVlogFileSize is the maximum size of the vlog file which can be created. Vlog Offset is of +// uint32, so limiting at max uint32. +var maxVlogFileSize uint32 = math.MaxUint32 + +// Values have their first byte being byteData or byteDelete. This helps us distinguish between +// a key that has never been seen and a key that has been explicitly deleted. +const ( + bitDelete byte = 1 << 0 // Set if the key has been deleted. + bitValuePointer byte = 1 << 1 // Set if the value is NOT stored directly next to key. + bitDiscardEarlierVersions byte = 1 << 2 // Set if earlier versions can be discarded. + // Set if item shouldn't be discarded via compactions (used by merge operator) + bitMergeEntry byte = 1 << 3 + // The MSB 2 bits are for transactions. + bitTxn byte = 1 << 6 // Set if the entry is part of a txn. + bitFinTxn byte = 1 << 7 // Set if the entry is to indicate end of txn in value log. + + mi int64 = 1 << 20 //nolint:unused + + // size of vlog header. + // +----------------+------------------+ + // | keyID(8 bytes) | baseIV(12 bytes)| + // +----------------+------------------+ + vlogHeaderSize = 20 +) + +var errStop = errors.New("Stop iteration") +var errTruncate = errors.New("Do truncate") + +type logEntry func(e Entry, vp valuePointer) error + +type safeRead struct { + k []byte + v []byte + + recordOffset uint32 + lf *logFile +} + +// hashReader implements io.Reader, io.ByteReader interfaces. It also keeps track of the number +// bytes read. The hashReader writes to h (hash) what it reads from r. +type hashReader struct { + r io.Reader + h hash.Hash32 + bytesRead int // Number of bytes read. +} + +func newHashReader(r io.Reader) *hashReader { + hash := crc32.New(y.CastagnoliCrcTable) + return &hashReader{ + r: r, + h: hash, + } +} + +// Read reads len(p) bytes from the reader. Returns the number of bytes read, error on failure. +func (t *hashReader) Read(p []byte) (int, error) { + n, err := t.r.Read(p) + if err != nil { + return n, err + } + t.bytesRead += n + return t.h.Write(p[:n]) +} + +// ReadByte reads exactly one byte from the reader. Returns error on failure. +func (t *hashReader) ReadByte() (byte, error) { + b := make([]byte, 1) + _, err := t.Read(b) + return b[0], err +} + +// Sum32 returns the sum32 of the underlying hash. +func (t *hashReader) Sum32() uint32 { + return t.h.Sum32() +} + +// Entry reads an entry from the provided reader. It also validates the checksum for every entry +// read. Returns error on failure. +func (r *safeRead) Entry(reader io.Reader) (*Entry, error) { + tee := newHashReader(reader) + var h header + hlen, err := h.DecodeFrom(tee) + if err != nil { + return nil, err + } + if h.klen > uint32(1<<16) { // Key length must be below uint16. + return nil, errTruncate + } + kl := int(h.klen) + if cap(r.k) < kl { + r.k = make([]byte, 2*kl) + } + vl := int(h.vlen) + if cap(r.v) < vl { + r.v = make([]byte, 2*vl) + } + + e := &Entry{} + e.offset = r.recordOffset + e.hlen = hlen + buf := make([]byte, h.klen+h.vlen) + if _, err := io.ReadFull(tee, buf[:]); err != nil { + if err == io.EOF { + err = errTruncate + } + return nil, err + } + if r.lf.encryptionEnabled() { + if buf, err = r.lf.decryptKV(buf[:], r.recordOffset); err != nil { + return nil, err + } + } + e.Key = buf[:h.klen] + e.Value = buf[h.klen:] + var crcBuf [crc32.Size]byte + if _, err := io.ReadFull(reader, crcBuf[:]); err != nil { + if err == io.EOF { + err = errTruncate + } + return nil, err + } + crc := y.BytesToU32(crcBuf[:]) + if crc != tee.Sum32() { + return nil, errTruncate + } + e.meta = h.meta + e.UserMeta = h.userMeta + e.ExpiresAt = h.expiresAt + return e, nil +} + +func (vlog *valueLog) rewrite(f *logFile) error { + vlog.filesLock.RLock() + for _, fid := range vlog.filesToBeDeleted { + if fid == f.fid { + vlog.filesLock.RUnlock() + return errors.Errorf("value log file already marked for deletion fid: %d", fid) + } + } + maxFid := vlog.maxFid + y.AssertTruef(f.fid < maxFid, "fid to move: %d. Current max fid: %d", f.fid, maxFid) + vlog.filesLock.RUnlock() + + vlog.opt.Infof("Rewriting fid: %d", f.fid) + wb := make([]*Entry, 0, 1000) + var size int64 + + y.AssertTrue(vlog.db != nil) + var count, moved int + fe := func(e Entry) error { + count++ + if count%100000 == 0 { + vlog.opt.Debugf("Processing entry %d", count) + } + + vs, err := vlog.db.get(e.Key) + if err != nil { + return err + } + if discardEntry(e, vs, vlog.db) { + return nil + } + + // Value is still present in value log. + if len(vs.Value) == 0 { + return errors.Errorf("Empty value: %+v", vs) + } + var vp valuePointer + vp.Decode(vs.Value) + + // If the entry found from the LSM Tree points to a newer vlog file, don't do anything. + if vp.Fid > f.fid { + return nil + } + // If the entry found from the LSM Tree points to an offset greater than the one + // read from vlog, don't do anything. + if vp.Offset > e.offset { + return nil + } + // If the entry read from LSM Tree and vlog file point to the same vlog file and offset, + // insert them back into the DB. + // NOTE: It might be possible that the entry read from the LSM Tree points to + // an older vlog file. See the comments in the else part. + if vp.Fid == f.fid && vp.Offset == e.offset { + moved++ + // This new entry only contains the key, and a pointer to the value. + ne := new(Entry) + // Remove only the bitValuePointer and transaction markers. We + // should keep the other bits. + ne.meta = e.meta &^ (bitValuePointer | bitTxn | bitFinTxn) + ne.UserMeta = e.UserMeta + ne.ExpiresAt = e.ExpiresAt + ne.Key = append([]byte{}, e.Key...) + ne.Value = append([]byte{}, e.Value...) + es := ne.estimateSizeAndSetThreshold(vlog.db.valueThreshold()) + // Consider size of value as well while considering the total size + // of the batch. There have been reports of high memory usage in + // rewrite because we don't consider the value size. See #1292. + es += int64(len(e.Value)) + + // Ensure length and size of wb is within transaction limits. + if int64(len(wb)+1) >= vlog.opt.maxBatchCount || + size+es >= vlog.opt.maxBatchSize { + if err := vlog.db.batchSet(wb); err != nil { + return err + } + size = 0 + wb = wb[:0] + } + wb = append(wb, ne) + size += es + } else { //nolint:staticcheck + // It might be possible that the entry read from LSM Tree points to + // an older vlog file. This can happen in the following situation. + // Assume DB is opened with + // numberOfVersionsToKeep=1 + // + // Now, if we have ONLY one key in the system "FOO" which has been + // updated 3 times and the same key has been garbage collected 3 + // times, we'll have 3 versions of the movekey + // for the same key "FOO". + // + // NOTE: moveKeyi is the gc'ed version of the original key with version i + // We're calling the gc'ed keys as moveKey to simplify the + // explanantion. We used to add move keys but we no longer do that. + // + // Assume we have 3 move keys in L0. + // - moveKey1 (points to vlog file 10), + // - moveKey2 (points to vlog file 14) and + // - moveKey3 (points to vlog file 15). + // + // Also, assume there is another move key "moveKey1" (points to + // vlog file 6) (this is also a move Key for key "FOO" ) on upper + // levels (let's say 3). The move key "moveKey1" on level 0 was + // inserted because vlog file 6 was GCed. + // + // Here's what the arrangement looks like + // L0 => (moveKey1 => vlog10), (moveKey2 => vlog14), (moveKey3 => vlog15) + // L1 => .... + // L2 => .... + // L3 => (moveKey1 => vlog6) + // + // When L0 compaction runs, it keeps only moveKey3 because the number of versions + // to keep is set to 1. (we've dropped moveKey1's latest version) + // + // The new arrangement of keys is + // L0 => .... + // L1 => (moveKey3 => vlog15) + // L2 => .... + // L3 => (moveKey1 => vlog6) + // + // Now if we try to GC vlog file 10, the entry read from vlog file + // will point to vlog10 but the entry read from LSM Tree will point + // to vlog6. The move key read from LSM tree will point to vlog6 + // because we've asked for version 1 of the move key. + // + // This might seem like an issue but it's not really an issue + // because the user has set the number of versions to keep to 1 and + // the latest version of moveKey points to the correct vlog file + // and offset. The stale move key on L3 will be eventually dropped + // by compaction because there is a newer versions in the upper + // levels. + } + return nil + } + + _, err := f.iterate(vlog.opt.ReadOnly, 0, func(e Entry, vp valuePointer) error { + return fe(e) + }) + if err != nil { + return err + } + + batchSize := 1024 + var loops int + for i := 0; i < len(wb); { + loops++ + if batchSize == 0 { + vlog.db.opt.Warningf("We shouldn't reach batch size of zero.") + return ErrNoRewrite + } + end := i + batchSize + if end > len(wb) { + end = len(wb) + } + if err := vlog.db.batchSet(wb[i:end]); err != nil { + if err == ErrTxnTooBig { + // Decrease the batch size to half. + batchSize = batchSize / 2 + continue + } + return err + } + i += batchSize + } + vlog.opt.Infof("Processed %d entries in %d loops", len(wb), loops) + vlog.opt.Infof("Total entries: %d. Moved: %d", count, moved) + vlog.opt.Infof("Removing fid: %d", f.fid) + var deleteFileNow bool + // Entries written to LSM. Remove the older file now. + { + vlog.filesLock.Lock() + // Just a sanity-check. + if _, ok := vlog.filesMap[f.fid]; !ok { + vlog.filesLock.Unlock() + return errors.Errorf("Unable to find fid: %d", f.fid) + } + if vlog.iteratorCount() == 0 { + delete(vlog.filesMap, f.fid) + deleteFileNow = true + } else { + vlog.filesToBeDeleted = append(vlog.filesToBeDeleted, f.fid) + } + vlog.filesLock.Unlock() + } + + if deleteFileNow { + if err := vlog.deleteLogFile(f); err != nil { + return err + } + } + return nil +} + +func (vlog *valueLog) incrIteratorCount() { + vlog.numActiveIterators.Add(1) +} + +func (vlog *valueLog) iteratorCount() int { + return int(vlog.numActiveIterators.Load()) +} + +func (vlog *valueLog) decrIteratorCount() error { + num := vlog.numActiveIterators.Add(-1) + if num != 0 { + return nil + } + + vlog.filesLock.Lock() + lfs := make([]*logFile, 0, len(vlog.filesToBeDeleted)) + for _, id := range vlog.filesToBeDeleted { + lfs = append(lfs, vlog.filesMap[id]) + delete(vlog.filesMap, id) + } + vlog.filesToBeDeleted = nil + vlog.filesLock.Unlock() + + for _, lf := range lfs { + if err := vlog.deleteLogFile(lf); err != nil { + return err + } + } + return nil +} + +func (vlog *valueLog) deleteLogFile(lf *logFile) error { + if lf == nil { + return nil + } + lf.lock.Lock() + defer lf.lock.Unlock() + // Delete fid from discard stats as well. + vlog.discardStats.Update(lf.fid, -1) + + return lf.Delete() +} + +func (vlog *valueLog) dropAll() (int, error) { + // If db is opened in InMemory mode, we don't need to do anything since there are no vlog files. + if vlog.db.opt.InMemory { + return 0, nil + } + // We don't want to block dropAll on any pending transactions. So, don't worry about iterator + // count. + var count int + deleteAll := func() error { + vlog.filesLock.Lock() + defer vlog.filesLock.Unlock() + for _, lf := range vlog.filesMap { + if err := vlog.deleteLogFile(lf); err != nil { + return err + } + count++ + } + vlog.filesMap = make(map[uint32]*logFile) + vlog.maxFid = 0 + return nil + } + if err := deleteAll(); err != nil { + return count, err + } + + vlog.db.opt.Infof("Value logs deleted. Creating value log file: 1") + if _, err := vlog.createVlogFile(); err != nil { // Called while writes are stopped. + return count, err + } + return count, nil +} + +func (db *DB) valueThreshold() int64 { + return db.threshold.valueThreshold.Load() +} + +type valueLog struct { + dirPath string + + // guards our view of which files exist, which to be deleted, how many active iterators + filesLock sync.RWMutex + filesMap map[uint32]*logFile + maxFid uint32 + filesToBeDeleted []uint32 + // A refcount of iterators -- when this hits zero, we can delete the filesToBeDeleted. + numActiveIterators atomic.Int32 + + db *DB + writableLogOffset atomic.Uint32 // read by read, written by write + numEntriesWritten uint32 + opt Options + + garbageCh chan struct{} + discardStats *discardStats +} + +func vlogFilePath(dirPath string, fid uint32) string { + return fmt.Sprintf("%s%s%06d.vlog", dirPath, string(os.PathSeparator), fid) +} + +func (vlog *valueLog) fpath(fid uint32) string { + return vlogFilePath(vlog.dirPath, fid) +} + +func (vlog *valueLog) populateFilesMap() error { + vlog.filesMap = make(map[uint32]*logFile) + + files, err := os.ReadDir(vlog.dirPath) + if err != nil { + return errFile(err, vlog.dirPath, "Unable to open log dir.") + } + + found := make(map[uint64]struct{}) + for _, file := range files { + if !strings.HasSuffix(file.Name(), ".vlog") { + continue + } + fsz := len(file.Name()) + fid, err := strconv.ParseUint(file.Name()[:fsz-5], 10, 32) + if err != nil { + return errFile(err, file.Name(), "Unable to parse log id.") + } + if _, ok := found[fid]; ok { + return errFile(err, file.Name(), "Duplicate file found. Please delete one.") + } + found[fid] = struct{}{} + + lf := &logFile{ + fid: uint32(fid), + path: vlog.fpath(uint32(fid)), + registry: vlog.db.registry, + } + vlog.filesMap[uint32(fid)] = lf + if vlog.maxFid < uint32(fid) { + vlog.maxFid = uint32(fid) + } + } + return nil +} + +func (vlog *valueLog) createVlogFile() (*logFile, error) { + fid := vlog.maxFid + 1 + path := vlog.fpath(fid) + lf := &logFile{ + fid: fid, + path: path, + registry: vlog.db.registry, + writeAt: vlogHeaderSize, + opt: vlog.opt, + } + err := lf.open(path, os.O_RDWR|os.O_CREATE|os.O_EXCL, 2*vlog.opt.ValueLogFileSize) + if err != z.NewFile && err != nil { + return nil, err + } + + vlog.filesLock.Lock() + vlog.filesMap[fid] = lf + y.AssertTrue(vlog.maxFid < fid) + vlog.maxFid = fid + // writableLogOffset is only written by write func, by read by Read func. + // To avoid a race condition, all reads and updates to this variable must be + // done via atomics. + vlog.writableLogOffset.Store(vlogHeaderSize) + vlog.numEntriesWritten = 0 + vlog.filesLock.Unlock() + + return lf, nil +} + +func errFile(err error, path string, msg string) error { + return fmt.Errorf("%s. Path=%s. Error=%v", msg, path, err) +} + +// init initializes the value log struct. This initialization needs to happen +// before compactions start. +func (vlog *valueLog) init(db *DB) { + vlog.opt = db.opt + vlog.db = db + // We don't need to open any vlog files or collect stats for GC if DB is opened + // in InMemory mode. InMemory mode doesn't create any files/directories on disk. + if vlog.opt.InMemory { + return + } + vlog.dirPath = vlog.opt.ValueDir + + vlog.garbageCh = make(chan struct{}, 1) // Only allow one GC at a time. + lf, err := InitDiscardStats(vlog.opt) + y.Check(err) + vlog.discardStats = lf + // See TestPersistLFDiscardStats for purpose of statement below. + db.logToSyncChan(endVLogInitMsg) +} + +func (vlog *valueLog) open(db *DB) error { + // We don't need to open any vlog files or collect stats for GC if DB is opened + // in InMemory mode. InMemory mode doesn't create any files/directories on disk. + if db.opt.InMemory { + return nil + } + + if err := vlog.populateFilesMap(); err != nil { + return err + } + // If no files are found, then create a new file. + if len(vlog.filesMap) == 0 { + if vlog.opt.ReadOnly { + return nil + } + _, err := vlog.createVlogFile() + return y.Wrapf(err, "Error while creating log file in valueLog.open") + } + fids := vlog.sortedFids() + for _, fid := range fids { + lf, ok := vlog.filesMap[fid] + y.AssertTrue(ok) + + // Just open in RDWR mode. This should not create a new log file. + lf.opt = vlog.opt + if err := lf.open(vlog.fpath(fid), os.O_RDWR, + 2*vlog.opt.ValueLogFileSize); err != nil { + return y.Wrapf(err, "Open existing file: %q", lf.path) + } + // We shouldn't delete the maxFid file. + if lf.size.Load() == vlogHeaderSize && fid != vlog.maxFid { + vlog.opt.Infof("Deleting empty file: %s", lf.path) + if err := lf.Delete(); err != nil { + return y.Wrapf(err, "while trying to delete empty file: %s", lf.path) + } + delete(vlog.filesMap, fid) + } + } + + if vlog.opt.ReadOnly { + return nil + } + // Now we can read the latest value log file, and see if it needs truncation. We could + // technically do this over all the value log files, but that would mean slowing down the value + // log open. + last, ok := vlog.filesMap[vlog.maxFid] + y.AssertTrue(ok) + lastOff, err := last.iterate(vlog.opt.ReadOnly, vlogHeaderSize, + func(_ Entry, vp valuePointer) error { + return nil + }) + if err != nil { + return y.Wrapf(err, "while iterating over: %s", last.path) + } + if err := last.Truncate(int64(lastOff)); err != nil { + return y.Wrapf(err, "while truncating last value log file: %s", last.path) + } + + // Don't write to the old log file. Always create a new one. + if _, err := vlog.createVlogFile(); err != nil { + return y.Wrapf(err, "Error while creating log file in valueLog.open") + } + return nil +} + +func (vlog *valueLog) Close() error { + if vlog == nil || vlog.db == nil || vlog.db.opt.InMemory { + return nil + } + + vlog.opt.Debugf("Stopping garbage collection of values.") + var err error + for id, lf := range vlog.filesMap { + lf.lock.Lock() // We won’t release the lock. + offset := int64(-1) + + if !vlog.opt.ReadOnly && id == vlog.maxFid { + offset = int64(vlog.woffset()) + } + if terr := lf.Close(offset); terr != nil && err == nil { + err = terr + } + } + if vlog.discardStats != nil { + vlog.db.captureDiscardStats() + if terr := vlog.discardStats.Close(-1); terr != nil && err == nil { + err = terr + } + } + return err +} + +// sortedFids returns the file id's not pending deletion, sorted. Assumes we have shared access to +// filesMap. +func (vlog *valueLog) sortedFids() []uint32 { + toBeDeleted := make(map[uint32]struct{}) + for _, fid := range vlog.filesToBeDeleted { + toBeDeleted[fid] = struct{}{} + } + ret := make([]uint32, 0, len(vlog.filesMap)) + for fid := range vlog.filesMap { + if _, ok := toBeDeleted[fid]; !ok { + ret = append(ret, fid) + } + } + sort.Slice(ret, func(i, j int) bool { + return ret[i] < ret[j] + }) + return ret +} + +type request struct { + // Input values + Entries []*Entry + // Output values and wait group stuff below + Ptrs []valuePointer + Wg sync.WaitGroup + Err error + ref atomic.Int32 +} + +func (req *request) reset() { + req.Entries = req.Entries[:0] + req.Ptrs = req.Ptrs[:0] + req.Wg = sync.WaitGroup{} + req.Err = nil + req.ref.Store(0) +} + +func (req *request) IncrRef() { + req.ref.Add(1) +} + +func (req *request) DecrRef() { + nRef := req.ref.Add(-1) + if nRef > 0 { + return + } + req.Entries = nil + requestPool.Put(req) +} + +func (req *request) Wait() error { + req.Wg.Wait() + err := req.Err + req.DecrRef() // DecrRef after writing to DB. + return err +} + +type requests []*request + +func (reqs requests) DecrRef() { + for _, req := range reqs { + req.DecrRef() + } +} + +func (reqs requests) IncrRef() { + for _, req := range reqs { + req.IncrRef() + } +} + +// sync function syncs content of latest value log file to disk. Syncing of value log directory is +// not required here as it happens every time a value log file rotation happens(check createVlogFile +// function). During rotation, previous value log file also gets synced to disk. It only syncs file +// if fid >= vlog.maxFid. In some cases such as replay(while opening db), it might be called with +// fid < vlog.maxFid. To sync irrespective of file id just call it with math.MaxUint32. +func (vlog *valueLog) sync() error { + if vlog.opt.SyncWrites || vlog.opt.InMemory { + return nil + } + + vlog.filesLock.RLock() + maxFid := vlog.maxFid + curlf := vlog.filesMap[maxFid] + // Sometimes it is possible that vlog.maxFid has been increased but file creation + // with same id is still in progress and this function is called. In those cases + // entry for the file might not be present in vlog.filesMap. + if curlf == nil { + vlog.filesLock.RUnlock() + return nil + } + curlf.lock.RLock() + vlog.filesLock.RUnlock() + + err := curlf.Sync() + curlf.lock.RUnlock() + return err +} + +func (vlog *valueLog) woffset() uint32 { + return vlog.writableLogOffset.Load() +} + +// validateWrites will check whether the given requests can fit into 4GB vlog file. +// NOTE: 4GB is the maximum size we can create for vlog because value pointer offset is of type +// uint32. If we create more than 4GB, it will overflow uint32. So, limiting the size to 4GB. +func (vlog *valueLog) validateWrites(reqs []*request) error { + vlogOffset := uint64(vlog.woffset()) + for _, req := range reqs { + // calculate size of the request. + size := estimateRequestSize(req) + estimatedVlogOffset := vlogOffset + size + if estimatedVlogOffset > uint64(maxVlogFileSize) { + return errors.Errorf("Request size offset %d is bigger than maximum offset %d", + estimatedVlogOffset, maxVlogFileSize) + } + + if estimatedVlogOffset >= uint64(vlog.opt.ValueLogFileSize) { + // We'll create a new vlog file if the estimated offset is greater or equal to + // max vlog size. So, resetting the vlogOffset. + vlogOffset = 0 + continue + } + // Estimated vlog offset will become current vlog offset if the vlog is not rotated. + vlogOffset = estimatedVlogOffset + } + return nil +} + +// estimateRequestSize returns the size that needed to be written for the given request. +func estimateRequestSize(req *request) uint64 { + size := uint64(0) + for _, e := range req.Entries { + size += uint64(maxHeaderSize + len(e.Key) + len(e.Value) + crc32.Size) + } + return size +} + +// write is thread-unsafe by design and should not be called concurrently. +func (vlog *valueLog) write(reqs []*request) error { + if vlog.db.opt.InMemory { + return nil + } + // Validate writes before writing to vlog. Because, we don't want to partially write and return + // an error. + if err := vlog.validateWrites(reqs); err != nil { + return y.Wrapf(err, "while validating writes") + } + + vlog.filesLock.RLock() + maxFid := vlog.maxFid + curlf := vlog.filesMap[maxFid] + vlog.filesLock.RUnlock() + + defer func() { + if vlog.opt.SyncWrites { + if err := curlf.Sync(); err != nil { + vlog.opt.Errorf("Error while curlf sync: %v\n", err) + } + } + }() + + write := func(buf *bytes.Buffer) error { + if buf.Len() == 0 { + return nil + } + + n := uint32(buf.Len()) + endOffset := vlog.writableLogOffset.Add(n) + // Increase the file size if we cannot accommodate this entry. + // [Aman] Should this be >= or just >? Doesn't make sense to extend the file if it big enough already. + if int(endOffset) >= len(curlf.Data) { + if err := curlf.Truncate(int64(endOffset)); err != nil { + return err + } + } + + start := int(endOffset - n) + y.AssertTrue(copy(curlf.Data[start:], buf.Bytes()) == int(n)) + + curlf.size.Store(endOffset) + return nil + } + + toDisk := func() error { + if vlog.woffset() > uint32(vlog.opt.ValueLogFileSize) || + vlog.numEntriesWritten > vlog.opt.ValueLogMaxEntries { + if err := curlf.doneWriting(vlog.woffset()); err != nil { + return err + } + + newlf, err := vlog.createVlogFile() + if err != nil { + return err + } + curlf = newlf + } + return nil + } + + buf := new(bytes.Buffer) + for i := range reqs { + b := reqs[i] + b.Ptrs = b.Ptrs[:0] + var written, bytesWritten int + valueSizes := make([]int64, 0, len(b.Entries)) + for j := range b.Entries { + buf.Reset() + + e := b.Entries[j] + valueSizes = append(valueSizes, int64(len(e.Value))) + if e.skipVlogAndSetThreshold(vlog.db.valueThreshold()) { + b.Ptrs = append(b.Ptrs, valuePointer{}) + continue + } + var p valuePointer + + p.Fid = curlf.fid + p.Offset = vlog.woffset() + + // We should not store transaction marks in the vlog file because it will never have all + // the entries in a transaction. If we store entries with transaction marks then value + // GC will not be able to iterate on the entire vlog file. + // But, we still want the entry to stay intact for the memTable WAL. So, store the meta + // in a temporary variable and reassign it after writing to the value log. + tmpMeta := e.meta + e.meta = e.meta &^ (bitTxn | bitFinTxn) + plen, err := curlf.encodeEntry(buf, e, p.Offset) // Now encode the entry into buffer. + if err != nil { + return err + } + // Restore the meta. + e.meta = tmpMeta + + p.Len = uint32(plen) + b.Ptrs = append(b.Ptrs, p) + if err := write(buf); err != nil { + return err + } + written++ + bytesWritten += buf.Len() + // No need to flush anything, we write to file directly via mmap. + } + y.NumWritesVlogAdd(vlog.opt.MetricsEnabled, int64(written)) + y.NumBytesWrittenVlogAdd(vlog.opt.MetricsEnabled, int64(bytesWritten)) + + vlog.numEntriesWritten += uint32(written) + vlog.db.threshold.update(valueSizes) + // We write to disk here so that all entries that are part of the same transaction are + // written to the same vlog file. + if err := toDisk(); err != nil { + return err + } + } + return toDisk() +} + +// Gets the logFile and acquires and RLock() for the mmap. You must call RUnlock on the file +// (if non-nil) +func (vlog *valueLog) getFileRLocked(vp valuePointer) (*logFile, error) { + vlog.filesLock.RLock() + defer vlog.filesLock.RUnlock() + ret, ok := vlog.filesMap[vp.Fid] + if !ok { + // log file has gone away, we can't do anything. Return. + return nil, errors.Errorf("file with ID: %d not found", vp.Fid) + } + + // Check for valid offset if we are reading from writable log. + maxFid := vlog.maxFid + // In read-only mode we don't need to check for writable offset as we are not writing anything. + // Moreover, this offset is not set in readonly mode. + if !vlog.opt.ReadOnly && vp.Fid == maxFid { + currentOffset := vlog.woffset() + if vp.Offset >= currentOffset { + return nil, errors.Errorf( + "Invalid value pointer offset: %d greater than current offset: %d", + vp.Offset, currentOffset) + } + } + + ret.lock.RLock() + return ret, nil +} + +// Read reads the value log at a given location. +// TODO: Make this read private. +func (vlog *valueLog) Read(vp valuePointer, _ *y.Slice) ([]byte, func(), error) { + buf, lf, err := vlog.readValueBytes(vp) + // log file is locked so, decide whether to lock immediately or let the caller to + // unlock it, after caller uses it. + cb := vlog.getUnlockCallback(lf) + if err != nil { + return nil, cb, err + } + + if vlog.opt.VerifyValueChecksum { + hash := crc32.New(y.CastagnoliCrcTable) + if _, err := hash.Write(buf[:len(buf)-crc32.Size]); err != nil { + runCallback(cb) + return nil, nil, y.Wrapf(err, "failed to write hash for vp %+v", vp) + } + // Fetch checksum from the end of the buffer. + checksum := buf[len(buf)-crc32.Size:] + if hash.Sum32() != y.BytesToU32(checksum) { + runCallback(cb) + return nil, nil, y.Wrapf(y.ErrChecksumMismatch, "value corrupted for vp: %+v", vp) + } + } + var h header + headerLen := h.Decode(buf) + kv := buf[headerLen:] + if lf.encryptionEnabled() { + kv, err = lf.decryptKV(kv, vp.Offset) + if err != nil { + return nil, cb, err + } + } + if uint32(len(kv)) < h.klen+h.vlen { + vlog.db.opt.Errorf("Invalid read: vp: %+v", vp) + return nil, nil, errors.Errorf("Invalid read: Len: %d read at:[%d:%d]", + len(kv), h.klen, h.klen+h.vlen) + } + return kv[h.klen : h.klen+h.vlen], cb, nil +} + +// getUnlockCallback will returns a function which unlock the logfile if the logfile is mmaped. +// otherwise, it unlock the logfile and return nil. +func (vlog *valueLog) getUnlockCallback(lf *logFile) func() { + if lf == nil { + return nil + } + return lf.lock.RUnlock +} + +// readValueBytes return vlog entry slice and read locked log file. Caller should take care of +// logFile unlocking. +func (vlog *valueLog) readValueBytes(vp valuePointer) ([]byte, *logFile, error) { + lf, err := vlog.getFileRLocked(vp) + if err != nil { + return nil, nil, err + } + + buf, err := lf.read(vp) + y.NumReadsVlogAdd(vlog.db.opt.MetricsEnabled, 1) + y.NumBytesReadsVlogAdd(vlog.db.opt.MetricsEnabled, int64(len(buf))) + return buf, lf, err +} + +func (vlog *valueLog) pickLog(discardRatio float64) *logFile { + vlog.filesLock.RLock() + defer vlog.filesLock.RUnlock() + +LOOP: + // Pick a candidate that contains the largest amount of discardable data + fid, discard := vlog.discardStats.MaxDiscard() + + // MaxDiscard will return fid=0 if it doesn't have any discard data. The + // vlog files start from 1. + if fid == 0 { + vlog.opt.Debugf("No file with discard stats") + return nil + } + lf, ok := vlog.filesMap[fid] + // This file was deleted but it's discard stats increased because of compactions. The file + // doesn't exist so we don't need to do anything. Skip it and retry. + if !ok { + vlog.discardStats.Update(fid, -1) + goto LOOP + } + // We have a valid file. + fi, err := lf.Fd.Stat() + if err != nil { + vlog.opt.Errorf("Unable to get stats for value log fid: %d err: %+v", fi, err) + return nil + } + if thr := discardRatio * float64(fi.Size()); float64(discard) < thr { + vlog.opt.Debugf("Discard: %d less than threshold: %.0f for file: %s", + discard, thr, fi.Name()) + return nil + } + if fid < vlog.maxFid { + vlog.opt.Infof("Found value log max discard fid: %d discard: %d\n", fid, discard) + lf, ok := vlog.filesMap[fid] + y.AssertTrue(ok) + return lf + } + + // Don't randomly pick any value log file. + return nil +} + +func discardEntry(e Entry, vs y.ValueStruct, db *DB) bool { + if vs.Version != y.ParseTs(e.Key) { + // Version not found. Discard. + return true + } + if isDeletedOrExpired(vs.Meta, vs.ExpiresAt) { + return true + } + if (vs.Meta & bitValuePointer) == 0 { + // Key also stores the value in LSM. Discard. + return true + } + if (vs.Meta & bitFinTxn) > 0 { + // Just a txn finish entry. Discard. + return true + } + return false +} + +func (vlog *valueLog) doRunGC(lf *logFile) error { + _, span := otrace.StartSpan(context.Background(), "Badger.GC") + span.Annotatef(nil, "GC rewrite for: %v", lf.path) + defer span.End() + if err := vlog.rewrite(lf); err != nil { + return err + } + // Remove the file from discardStats. + vlog.discardStats.Update(lf.fid, -1) + return nil +} + +func (vlog *valueLog) waitOnGC(lc *z.Closer) { + defer lc.Done() + + <-lc.HasBeenClosed() // Wait for lc to be closed. + + // Block any GC in progress to finish, and don't allow any more writes to runGC by filling up + // the channel of size 1. + vlog.garbageCh <- struct{}{} +} + +func (vlog *valueLog) runGC(discardRatio float64) error { + select { + case vlog.garbageCh <- struct{}{}: + // Pick a log file for GC. + defer func() { + <-vlog.garbageCh + }() + + lf := vlog.pickLog(discardRatio) + if lf == nil { + return ErrNoRewrite + } + return vlog.doRunGC(lf) + default: + return ErrRejected + } +} + +func (vlog *valueLog) updateDiscardStats(stats map[uint32]int64) { + if vlog.opt.InMemory { + return + } + for fid, discard := range stats { + vlog.discardStats.Update(fid, discard) + } + // The following is to coordinate with some test cases where we want to + // verify that at least one iteration of updateDiscardStats has been completed. + vlog.db.logToSyncChan(updateDiscardStatsMsg) +} + +type vlogThreshold struct { + logger Logger + percentile float64 + valueThreshold atomic.Int64 + valueCh chan []int64 + clearCh chan bool + closer *z.Closer + // Metrics contains a running log of statistics like amount of data stored etc. + vlMetrics *z.HistogramData +} + +func initVlogThreshold(opt *Options) *vlogThreshold { + getBounds := func() []float64 { + mxbd := opt.maxValueThreshold + mnbd := float64(opt.ValueThreshold) + y.AssertTruef(mxbd >= mnbd, "maximum threshold bound is less than the min threshold") + size := math.Min(mxbd-mnbd+1, 1024.0) + bdstp := (mxbd - mnbd) / size + bounds := make([]float64, int64(size)) + for i := range bounds { + if i == 0 { + bounds[0] = mnbd + continue + } + if i == int(size-1) { + bounds[i] = mxbd + continue + } + bounds[i] = bounds[i-1] + bdstp + } + return bounds + } + lt := &vlogThreshold{ + logger: opt.Logger, + percentile: opt.VLogPercentile, + valueCh: make(chan []int64, 1000), + clearCh: make(chan bool, 1), + closer: z.NewCloser(1), + vlMetrics: z.NewHistogramData(getBounds()), + } + lt.valueThreshold.Store(opt.ValueThreshold) + return lt +} + +func (v *vlogThreshold) Clear(opt Options) { + v.valueThreshold.Store(opt.ValueThreshold) + v.clearCh <- true +} + +func (v *vlogThreshold) update(sizes []int64) { + v.valueCh <- sizes +} + +func (v *vlogThreshold) close() { + v.closer.SignalAndWait() +} + +func (v *vlogThreshold) listenForValueThresholdUpdate() { + defer v.closer.Done() + for { + select { + case <-v.closer.HasBeenClosed(): + return + case val := <-v.valueCh: + for _, e := range val { + v.vlMetrics.Update(e) + } + // we are making it to get Options.VlogPercentile so that values with sizes + // in range of Options.VlogPercentile will make it to the LSM tree and rest to the + // value log file. + p := int64(v.vlMetrics.Percentile(v.percentile)) + if v.valueThreshold.Load() != p { + if v.logger != nil { + v.logger.Infof("updating value of threshold to: %d", p) + } + v.valueThreshold.Store(p) + } + case <-v.clearCh: + v.vlMetrics.Clear() + } + } +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/bloom.go b/vendor/github.com/dgraph-io/badger/v4/y/bloom.go new file mode 100644 index 00000000000..65f4f087944 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/bloom.go @@ -0,0 +1,170 @@ +// Copyright 2013 The LevelDB-Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package y + +import "math" + +// Filter is an encoded set of []byte keys. +type Filter []byte + +func (f Filter) MayContainKey(k []byte) bool { + return f.MayContain(Hash(k)) +} + +// MayContain returns whether the filter may contain given key. False positives +// are possible, where it returns true for keys not in the original set. +func (f Filter) MayContain(h uint32) bool { + if len(f) < 2 { + return false + } + k := f[len(f)-1] + if k > 30 { + // This is reserved for potentially new encodings for short Bloom filters. + // Consider it a match. + return true + } + nBits := uint32(8 * (len(f) - 1)) + delta := h>>17 | h<<15 + for j := uint8(0); j < k; j++ { + bitPos := h % nBits + if f[bitPos/8]&(1<<(bitPos%8)) == 0 { + return false + } + h += delta + } + return true +} + +// NewFilter returns a new Bloom filter that encodes a set of []byte keys with +// the given number of bits per key, approximately. +// +// A good bitsPerKey value is 10, which yields a filter with ~ 1% false +// positive rate. +func NewFilter(keys []uint32, bitsPerKey int) Filter { + return Filter(appendFilter(nil, keys, bitsPerKey)) +} + +// BloomBitsPerKey returns the bits per key required by bloomfilter based on +// the false positive rate. +func BloomBitsPerKey(numEntries int, fp float64) int { + size := -1 * float64(numEntries) * math.Log(fp) / math.Pow(float64(0.69314718056), 2) + locs := math.Ceil(float64(0.69314718056) * size / float64(numEntries)) + return int(locs) +} + +func appendFilter(buf []byte, keys []uint32, bitsPerKey int) []byte { + if bitsPerKey < 0 { + bitsPerKey = 0 + } + // 0.69 is approximately ln(2). + k := uint32(float64(bitsPerKey) * 0.69) + if k < 1 { + k = 1 + } + if k > 30 { + k = 30 + } + + nBits := len(keys) * bitsPerKey + // For small len(keys), we can see a very high false positive rate. Fix it + // by enforcing a minimum bloom filter length. + if nBits < 64 { + nBits = 64 + } + nBytes := (nBits + 7) / 8 + nBits = nBytes * 8 + buf, filter := extend(buf, nBytes+1) + + for _, h := range keys { + delta := h>>17 | h<<15 + for j := uint32(0); j < k; j++ { + bitPos := h % uint32(nBits) + filter[bitPos/8] |= 1 << (bitPos % 8) + h += delta + } + } + filter[nBytes] = uint8(k) + + return buf +} + +// extend appends n zero bytes to b. It returns the overall slice (of length +// n+len(originalB)) and the slice of n trailing zeroes. +func extend(b []byte, n int) (overall, trailer []byte) { + want := n + len(b) + if want <= cap(b) { + overall = b[:want] + trailer = overall[len(b):] + for i := range trailer { + trailer[i] = 0 + } + } else { + // Grow the capacity exponentially, with a 1KiB minimum. + c := 1024 + for c < want { + c += c / 4 + } + overall = make([]byte, want, c) + trailer = overall[len(b):] + copy(overall, b) + } + return overall, trailer +} + +// hash implements a hashing algorithm similar to the Murmur hash. +func Hash(b []byte) uint32 { + const ( + seed = 0xbc9f1d34 + m = 0xc6a4a793 + ) + h := uint32(seed) ^ uint32(len(b))*m + for ; len(b) >= 4; b = b[4:] { + h += uint32(b[0]) | uint32(b[1])<<8 | uint32(b[2])<<16 | uint32(b[3])<<24 + h *= m + h ^= h >> 16 + } + switch len(b) { + case 3: + h += uint32(b[2]) << 16 + fallthrough + case 2: + h += uint32(b[1]) << 8 + fallthrough + case 1: + h += uint32(b[0]) + h *= m + h ^= h >> 24 + } + return h +} + +// FilterPolicy implements the db.FilterPolicy interface from the leveldb/db +// package. +// +// The integer value is the approximate number of bits used per key. A good +// value is 10, which yields a filter with ~ 1% false positive rate. +// +// It is valid to use the other API in this package (leveldb/bloom) without +// using this type or the leveldb/db package. + +// type FilterPolicy int + +// // Name implements the db.FilterPolicy interface. +// func (p FilterPolicy) Name() string { +// // This string looks arbitrary, but its value is written to LevelDB .ldb +// // files, and should be this exact value to be compatible with those files +// // and with the C++ LevelDB code. +// return "leveldb.BuiltinBloomFilter2" +// } + +// // AppendFilter implements the db.FilterPolicy interface. +// func (p FilterPolicy) AppendFilter(dst []byte, keys [][]byte) []byte { +// return appendFilter(dst, keys, int(p)) +// } + +// // MayContain implements the db.FilterPolicy interface. +// func (p FilterPolicy) MayContain(filter, key []byte) bool { +// return Filter(filter).MayContain(key) +// } diff --git a/vendor/github.com/dgraph-io/badger/v4/y/checksum.go b/vendor/github.com/dgraph-io/badger/v4/y/checksum.go new file mode 100644 index 00000000000..5a4ac11b6f9 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/checksum.go @@ -0,0 +1,50 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "hash/crc32" + + "github.com/cespare/xxhash/v2" + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" +) + +// ErrChecksumMismatch is returned at checksum mismatch. +var ErrChecksumMismatch = errors.New("checksum mismatch") + +// CalculateChecksum calculates checksum for data using ct checksum type. +func CalculateChecksum(data []byte, ct pb.Checksum_Algorithm) uint64 { + switch ct { + case pb.Checksum_CRC32C: + return uint64(crc32.Checksum(data, CastagnoliCrcTable)) + case pb.Checksum_XXHash64: + return xxhash.Sum64(data) + default: + panic("checksum type not supported") + } +} + +// VerifyChecksum validates the checksum for the data against the given expected checksum. +func VerifyChecksum(data []byte, expected *pb.Checksum) error { + actual := CalculateChecksum(data, expected.Algo) + if actual != expected.Sum { + return Wrapf(ErrChecksumMismatch, "actual: %d, expected: %d", actual, expected.Sum) + } + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/encrypt.go b/vendor/github.com/dgraph-io/badger/v4/y/encrypt.go new file mode 100644 index 00000000000..a238fcf1b0d --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/encrypt.go @@ -0,0 +1,67 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "bytes" + "crypto/aes" + "crypto/cipher" + "crypto/rand" + "io" +) + +// XORBlock encrypts the given data with AES and XOR's with IV. +// Can be used for both encryption and decryption. IV is of +// AES block size. +func XORBlock(dst, src, key, iv []byte) error { + block, err := aes.NewCipher(key) + if err != nil { + return err + } + stream := cipher.NewCTR(block, iv) + stream.XORKeyStream(dst, src) + return nil +} + +func XORBlockAllocate(src, key, iv []byte) ([]byte, error) { + block, err := aes.NewCipher(key) + if err != nil { + return nil, err + } + stream := cipher.NewCTR(block, iv) + dst := make([]byte, len(src)) + stream.XORKeyStream(dst, src) + return dst, nil +} + +func XORBlockStream(w io.Writer, src, key, iv []byte) error { + block, err := aes.NewCipher(key) + if err != nil { + return err + } + stream := cipher.NewCTR(block, iv) + sw := cipher.StreamWriter{S: stream, W: w} + _, err = io.Copy(sw, bytes.NewReader(src)) + return Wrapf(err, "XORBlockStream") +} + +// GenerateIV generates IV. +func GenerateIV() ([]byte, error) { + iv := make([]byte, aes.BlockSize) + _, err := rand.Read(iv) + return iv, err +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/error.go b/vendor/github.com/dgraph-io/badger/v4/y/error.go new file mode 100644 index 00000000000..f36181b4651 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/error.go @@ -0,0 +1,99 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +// This file contains some functions for error handling. Note that we are moving +// towards using x.Trace, i.e., rpc tracing using net/tracer. But for now, these +// functions are useful for simple checks logged on one machine. +// Some common use cases are: +// (1) You receive an error from external lib, and would like to check/log fatal. +// For this, use x.Check, x.Checkf. These will check for err != nil, which is +// more common in Go. If you want to check for boolean being true, use +// x.Assert, x.Assertf. +// (2) You receive an error from external lib, and would like to pass on with some +// stack trace information. In this case, use x.Wrap or x.Wrapf. +// (3) You want to generate a new error with stack trace info. Use x.Errorf. + +import ( + "fmt" + "log" + + "github.com/pkg/errors" +) + +var debugMode = false + +// Check logs fatal if err != nil. +func Check(err error) { + if err != nil { + log.Fatalf("%+v", Wrap(err, "")) + } +} + +// Check2 acts as convenience wrapper around Check, using the 2nd argument as error. +func Check2(_ interface{}, err error) { + Check(err) +} + +// AssertTrue asserts that b is true. Otherwise, it would log fatal. +func AssertTrue(b bool) { + if !b { + log.Fatalf("%+v", errors.Errorf("Assert failed")) + } +} + +// AssertTruef is AssertTrue with extra info. +func AssertTruef(b bool, format string, args ...interface{}) { + if !b { + log.Fatalf("%+v", errors.Errorf(format, args...)) + } +} + +// Wrap wraps errors from external lib. +func Wrap(err error, msg string) error { + if !debugMode { + if err == nil { + return nil + } + return fmt.Errorf("%s err: %+v", msg, err) + } + return errors.Wrap(err, msg) +} + +// Wrapf is Wrap with extra info. +func Wrapf(err error, format string, args ...interface{}) error { + if !debugMode { + if err == nil { + return nil + } + return fmt.Errorf(format+" error: %+v", append(args, err)...) + } + return errors.Wrapf(err, format, args...) +} + +func CombineErrors(one, other error) error { + if one != nil && other != nil { + return fmt.Errorf("%v; %v", one, other) + } + if one != nil && other == nil { + return fmt.Errorf("%v", one) + } + if one == nil && other != nil { + return fmt.Errorf("%v", other) + } + return nil +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/event_log.go b/vendor/github.com/dgraph-io/badger/v4/y/event_log.go new file mode 100644 index 00000000000..ba9dcb1f636 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/event_log.go @@ -0,0 +1,31 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import "golang.org/x/net/trace" + +var ( + NoEventLog trace.EventLog = nilEventLog{} +) + +type nilEventLog struct{} + +func (nel nilEventLog) Printf(format string, a ...interface{}) {} + +func (nel nilEventLog) Errorf(format string, a ...interface{}) {} + +func (nel nilEventLog) Finish() {} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/file_dsync.go b/vendor/github.com/dgraph-io/badger/v4/y/file_dsync.go new file mode 100644 index 00000000000..1e171733ebc --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/file_dsync.go @@ -0,0 +1,26 @@ +//go:build !dragonfly && !freebsd && !windows && !plan9 +// +build !dragonfly,!freebsd,!windows,!plan9 + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import "golang.org/x/sys/unix" + +func init() { + datasyncFileFlag = unix.O_DSYNC +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/file_nodsync.go b/vendor/github.com/dgraph-io/badger/v4/y/file_nodsync.go new file mode 100644 index 00000000000..99f6888c712 --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/file_nodsync.go @@ -0,0 +1,26 @@ +//go:build dragonfly || freebsd || windows || plan9 +// +build dragonfly freebsd windows plan9 + +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import "syscall" + +func init() { + datasyncFileFlag = syscall.O_SYNC +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/iterator.go b/vendor/github.com/dgraph-io/badger/v4/y/iterator.go new file mode 100644 index 00000000000..ef032ef4faa --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/iterator.go @@ -0,0 +1,95 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "bytes" + "encoding/binary" +) + +// ValueStruct represents the value info that can be associated with a key, but also the internal +// Meta field. +type ValueStruct struct { + Meta byte + UserMeta byte + ExpiresAt uint64 + Value []byte + + Version uint64 // This field is not serialized. Only for internal usage. +} + +func sizeVarint(x uint64) (n int) { + for { + n++ + x >>= 7 + if x == 0 { + break + } + } + return n +} + +// EncodedSize is the size of the ValueStruct when encoded +func (v *ValueStruct) EncodedSize() uint32 { + sz := len(v.Value) + 2 // meta, usermeta. + enc := sizeVarint(v.ExpiresAt) + return uint32(sz + enc) +} + +// Decode uses the length of the slice to infer the length of the Value field. +func (v *ValueStruct) Decode(b []byte) { + v.Meta = b[0] + v.UserMeta = b[1] + var sz int + v.ExpiresAt, sz = binary.Uvarint(b[2:]) + v.Value = b[2+sz:] +} + +// Encode expects a slice of length at least v.EncodedSize(). +func (v *ValueStruct) Encode(b []byte) uint32 { + b[0] = v.Meta + b[1] = v.UserMeta + sz := binary.PutUvarint(b[2:], v.ExpiresAt) + n := copy(b[2+sz:], v.Value) + return uint32(2 + sz + n) +} + +// EncodeTo should be kept in sync with the Encode function above. The reason +// this function exists is to avoid creating byte arrays per key-value pair in +// table/builder.go. +func (v *ValueStruct) EncodeTo(buf *bytes.Buffer) { + buf.WriteByte(v.Meta) + buf.WriteByte(v.UserMeta) + var enc [binary.MaxVarintLen64]byte + sz := binary.PutUvarint(enc[:], v.ExpiresAt) + + buf.Write(enc[:sz]) + buf.Write(v.Value) +} + +// Iterator is an interface for a basic iterator. +type Iterator interface { + Next() + Rewind() + Seek(key []byte) + Key() []byte + Value() ValueStruct + Valid() bool + + // All iterators should be closed so that file garbage collection works. + Close() error +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/metrics.go b/vendor/github.com/dgraph-io/badger/v4/y/metrics.go new file mode 100644 index 00000000000..026328d197c --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/metrics.go @@ -0,0 +1,223 @@ +/* + * Copyright (C) 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "expvar" +) + +const ( + BADGER_METRIC_PREFIX = "badger_" +) + +var ( + // lsmSize has size of the LSM in bytes + lsmSize *expvar.Map + // vlogSize has size of the value log in bytes + vlogSize *expvar.Map + // pendingWrites tracks the number of pending writes. + pendingWrites *expvar.Map + + // These are cumulative + + // VLOG METRICS + // numReads has cumulative number of reads from vlog + numReadsVlog *expvar.Int + // numWrites has cumulative number of writes into vlog + numWritesVlog *expvar.Int + // numBytesRead has cumulative number of bytes read from VLOG + numBytesReadVlog *expvar.Int + // numBytesVlogWritten has cumulative number of bytes written into VLOG + numBytesVlogWritten *expvar.Int + + // LSM METRICS + // numBytesRead has cumulative number of bytes read from LSM tree + numBytesReadLSM *expvar.Int + // numBytesWrittenToL0 has cumulative number of bytes written into LSM Tree + numBytesWrittenToL0 *expvar.Int + // numLSMGets is number of LSM gets + numLSMGets *expvar.Map + // numBytesCompactionWritten is the number of bytes written in the lsm tree due to compaction + numBytesCompactionWritten *expvar.Map + // numLSMBloomHits is number of LMS bloom hits + numLSMBloomHits *expvar.Map + + // DB METRICS + // numGets is number of gets -> Number of get requests made + numGets *expvar.Int + // number of get queries in which we actually get a result + numGetsWithResults *expvar.Int + // number of iterators created, these would be the number of range queries + numIteratorsCreated *expvar.Int + // numPuts is number of puts -> Number of puts requests made + numPuts *expvar.Int + // numMemtableGets is number of memtable gets -> Number of get requests made on memtable + numMemtableGets *expvar.Int + // numCompactionTables is the number of tables being compacted + numCompactionTables *expvar.Int + // Total writes by a user in bytes + numBytesWrittenUser *expvar.Int +) + +// These variables are global and have cumulative values for all kv stores. +// Naming convention of metrics: {badger_version}_{singular operation}_{granularity}_{component} +func init() { + numReadsVlog = expvar.NewInt(BADGER_METRIC_PREFIX + "read_num_vlog") + numBytesReadVlog = expvar.NewInt(BADGER_METRIC_PREFIX + "read_bytes_vlog") + numWritesVlog = expvar.NewInt(BADGER_METRIC_PREFIX + "write_num_vlog") + numBytesVlogWritten = expvar.NewInt(BADGER_METRIC_PREFIX + "write_bytes_vlog") + + numBytesReadLSM = expvar.NewInt(BADGER_METRIC_PREFIX + "read_bytes_lsm") + numBytesWrittenToL0 = expvar.NewInt(BADGER_METRIC_PREFIX + "write_bytes_l0") + numBytesCompactionWritten = expvar.NewMap(BADGER_METRIC_PREFIX + "write_bytes_compaction") + + numLSMGets = expvar.NewMap(BADGER_METRIC_PREFIX + "get_num_lsm") + numLSMBloomHits = expvar.NewMap(BADGER_METRIC_PREFIX + "hit_num_lsm_bloom_filter") + numMemtableGets = expvar.NewInt(BADGER_METRIC_PREFIX + "get_num_memtable") + + // User operations + numGets = expvar.NewInt(BADGER_METRIC_PREFIX + "get_num_user") + numPuts = expvar.NewInt(BADGER_METRIC_PREFIX + "put_num_user") + numBytesWrittenUser = expvar.NewInt(BADGER_METRIC_PREFIX + "write_bytes_user") + + // Required for Enabled + numGetsWithResults = expvar.NewInt(BADGER_METRIC_PREFIX + "get_with_result_num_user") + numIteratorsCreated = expvar.NewInt(BADGER_METRIC_PREFIX + "iterator_num_user") + + // Sizes + lsmSize = expvar.NewMap(BADGER_METRIC_PREFIX + "size_bytes_lsm") + vlogSize = expvar.NewMap(BADGER_METRIC_PREFIX + "size_bytes_vlog") + + pendingWrites = expvar.NewMap(BADGER_METRIC_PREFIX + "write_pending_num_memtable") + numCompactionTables = expvar.NewInt(BADGER_METRIC_PREFIX + "compaction_current_num_lsm") +} + +func NumIteratorsCreatedAdd(enabled bool, val int64) { + addInt(enabled, numIteratorsCreated, val) +} + +func NumGetsWithResultsAdd(enabled bool, val int64) { + addInt(enabled, numGetsWithResults, val) +} + +func NumReadsVlogAdd(enabled bool, val int64) { + addInt(enabled, numReadsVlog, val) +} + +func NumBytesWrittenUserAdd(enabled bool, val int64) { + addInt(enabled, numBytesWrittenUser, val) +} + +func NumWritesVlogAdd(enabled bool, val int64) { + addInt(enabled, numWritesVlog, val) +} + +func NumBytesReadsVlogAdd(enabled bool, val int64) { + addInt(enabled, numBytesReadVlog, val) +} + +func NumBytesReadsLSMAdd(enabled bool, val int64) { + addInt(enabled, numBytesReadLSM, val) +} + +func NumBytesWrittenVlogAdd(enabled bool, val int64) { + addInt(enabled, numBytesVlogWritten, val) +} + +func NumBytesWrittenToL0Add(enabled bool, val int64) { + addInt(enabled, numBytesWrittenToL0, val) +} + +func NumBytesCompactionWrittenAdd(enabled bool, key string, val int64) { + addToMap(enabled, numBytesCompactionWritten, key, val) +} + +func NumGetsAdd(enabled bool, val int64) { + addInt(enabled, numGets, val) +} + +func NumPutsAdd(enabled bool, val int64) { + addInt(enabled, numPuts, val) +} + +func NumMemtableGetsAdd(enabled bool, val int64) { + addInt(enabled, numMemtableGets, val) +} + +func NumCompactionTablesAdd(enabled bool, val int64) { + addInt(enabled, numCompactionTables, val) +} + +func LSMSizeSet(enabled bool, key string, val expvar.Var) { + storeToMap(enabled, lsmSize, key, val) +} + +func VlogSizeSet(enabled bool, key string, val expvar.Var) { + storeToMap(enabled, vlogSize, key, val) +} + +func PendingWritesSet(enabled bool, key string, val expvar.Var) { + storeToMap(enabled, pendingWrites, key, val) +} + +func NumLSMBloomHitsAdd(enabled bool, key string, val int64) { + addToMap(enabled, numLSMBloomHits, key, val) +} + +func NumLSMGetsAdd(enabled bool, key string, val int64) { + addToMap(enabled, numLSMGets, key, val) +} + +func LSMSizeGet(enabled bool, key string) expvar.Var { + return getFromMap(enabled, lsmSize, key) +} + +func VlogSizeGet(enabled bool, key string) expvar.Var { + return getFromMap(enabled, vlogSize, key) +} + +func addInt(enabled bool, metric *expvar.Int, val int64) { + if !enabled { + return + } + + metric.Add(val) +} + +func addToMap(enabled bool, metric *expvar.Map, key string, val int64) { + if !enabled { + return + } + + metric.Add(key, val) +} + +func storeToMap(enabled bool, metric *expvar.Map, key string, val expvar.Var) { + if !enabled { + return + } + + metric.Set(key, val) +} + +func getFromMap(enabled bool, metric *expvar.Map, key string) expvar.Var { + if !enabled { + return nil + } + + return metric.Get(key) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/watermark.go b/vendor/github.com/dgraph-io/badger/v4/y/watermark.go new file mode 100644 index 00000000000..cf2992b8b2c --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/watermark.go @@ -0,0 +1,240 @@ +/* + * Copyright 2016-2018 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "container/heap" + "context" + "sync/atomic" + + "github.com/dgraph-io/ristretto/z" +) + +type uint64Heap []uint64 + +func (u uint64Heap) Len() int { return len(u) } +func (u uint64Heap) Less(i, j int) bool { return u[i] < u[j] } +func (u uint64Heap) Swap(i, j int) { u[i], u[j] = u[j], u[i] } +func (u *uint64Heap) Push(x interface{}) { *u = append(*u, x.(uint64)) } +func (u *uint64Heap) Pop() interface{} { + old := *u + n := len(old) + x := old[n-1] + *u = old[0 : n-1] + return x +} + +// mark contains one of more indices, along with a done boolean to indicate the +// status of the index: begin or done. It also contains waiters, who could be +// waiting for the watermark to reach >= a certain index. +type mark struct { + // Either this is an (index, waiter) pair or (index, done) or (indices, done). + index uint64 + waiter chan struct{} + indices []uint64 + done bool // Set to true if the index is done. +} + +// WaterMark is used to keep track of the minimum un-finished index. Typically, an index k becomes +// finished or "done" according to a WaterMark once Done(k) has been called +// 1. as many times as Begin(k) has, AND +// 2. a positive number of times. +// +// An index may also become "done" by calling SetDoneUntil at a time such that it is not +// inter-mingled with Begin/Done calls. +// +// Since doneUntil and lastIndex addresses are passed to sync/atomic packages, we ensure that they +// are 64-bit aligned by putting them at the beginning of the structure. +type WaterMark struct { + doneUntil atomic.Uint64 + lastIndex atomic.Uint64 + Name string + markCh chan mark +} + +// Init initializes a WaterMark struct. MUST be called before using it. +func (w *WaterMark) Init(closer *z.Closer) { + w.markCh = make(chan mark, 100) + go w.process(closer) +} + +// Begin sets the last index to the given value. +func (w *WaterMark) Begin(index uint64) { + w.lastIndex.Store(index) + w.markCh <- mark{index: index, done: false} +} + +// BeginMany works like Begin but accepts multiple indices. +func (w *WaterMark) BeginMany(indices []uint64) { + w.lastIndex.Store(indices[len(indices)-1]) + w.markCh <- mark{index: 0, indices: indices, done: false} +} + +// Done sets a single index as done. +func (w *WaterMark) Done(index uint64) { + w.markCh <- mark{index: index, done: true} +} + +// DoneMany works like Done but accepts multiple indices. +func (w *WaterMark) DoneMany(indices []uint64) { + w.markCh <- mark{index: 0, indices: indices, done: true} +} + +// DoneUntil returns the maximum index that has the property that all indices +// less than or equal to it are done. +func (w *WaterMark) DoneUntil() uint64 { + return w.doneUntil.Load() +} + +// SetDoneUntil sets the maximum index that has the property that all indices +// less than or equal to it are done. +func (w *WaterMark) SetDoneUntil(val uint64) { + w.doneUntil.Store(val) +} + +// LastIndex returns the last index for which Begin has been called. +func (w *WaterMark) LastIndex() uint64 { + return w.lastIndex.Load() +} + +// WaitForMark waits until the given index is marked as done. +func (w *WaterMark) WaitForMark(ctx context.Context, index uint64) error { + if w.DoneUntil() >= index { + return nil + } + waitCh := make(chan struct{}) + w.markCh <- mark{index: index, waiter: waitCh} + + select { + case <-ctx.Done(): + return ctx.Err() + case <-waitCh: + return nil + } +} + +// process is used to process the Mark channel. This is not thread-safe, +// so only run one goroutine for process. One is sufficient, because +// all goroutine ops use purely memory and cpu. +// Each index has to emit atleast one begin watermark in serial order otherwise waiters +// can get blocked idefinitely. Example: We had an watermark at 100 and a waiter at 101, +// if no watermark is emitted at index 101 then waiter would get stuck indefinitely as it +// can't decide whether the task at 101 has decided not to emit watermark or it didn't get +// scheduled yet. +func (w *WaterMark) process(closer *z.Closer) { + defer closer.Done() + + var indices uint64Heap + // pending maps raft proposal index to the number of pending mutations for this proposal. + pending := make(map[uint64]int) + waiters := make(map[uint64][]chan struct{}) + + heap.Init(&indices) + + processOne := func(index uint64, done bool) { + // If not already done, then set. Otherwise, don't undo a done entry. + prev, present := pending[index] + if !present { + heap.Push(&indices, index) + } + + delta := 1 + if done { + delta = -1 + } + pending[index] = prev + delta + + // Update mark by going through all indices in order; and checking if they have + // been done. Stop at the first index, which isn't done. + doneUntil := w.DoneUntil() + if doneUntil > index { + AssertTruef(false, "Name: %s doneUntil: %d. Index: %d", w.Name, doneUntil, index) + } + + until := doneUntil + loops := 0 + + for len(indices) > 0 { + min := indices[0] + if done := pending[min]; done > 0 { + break // len(indices) will be > 0. + } + // Even if done is called multiple times causing it to become + // negative, we should still pop the index. + heap.Pop(&indices) + delete(pending, min) + until = min + loops++ + } + + if until != doneUntil { + AssertTrue(w.doneUntil.CompareAndSwap(doneUntil, until)) + } + + notifyAndRemove := func(idx uint64, toNotify []chan struct{}) { + for _, ch := range toNotify { + close(ch) + } + delete(waiters, idx) // Release the memory back. + } + + if until-doneUntil <= uint64(len(waiters)) { + // Issue #908 showed that if doneUntil is close to 2^60, while until is zero, this loop + // can hog up CPU just iterating over integers creating a busy-wait loop. So, only do + // this path if until - doneUntil is less than the number of waiters. + for idx := doneUntil + 1; idx <= until; idx++ { + if toNotify, ok := waiters[idx]; ok { + notifyAndRemove(idx, toNotify) + } + } + } else { + for idx, toNotify := range waiters { + if idx <= until { + notifyAndRemove(idx, toNotify) + } + } + } // end of notifying waiters. + } + + for { + select { + case <-closer.HasBeenClosed(): + return + case mark := <-w.markCh: + if mark.waiter != nil { + doneUntil := w.doneUntil.Load() + if doneUntil >= mark.index { + close(mark.waiter) + } else { + ws, ok := waiters[mark.index] + if !ok { + waiters[mark.index] = []chan struct{}{mark.waiter} + } else { + waiters[mark.index] = append(ws, mark.waiter) + } + } + } else { + if mark.index > 0 { + processOne(mark.index, mark.done) + } + for _, index := range mark.indices { + processOne(index, mark.done) + } + } + } + } +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/y.go b/vendor/github.com/dgraph-io/badger/v4/y/y.go new file mode 100644 index 00000000000..83da2c1d5aa --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/y.go @@ -0,0 +1,595 @@ +/* + * Copyright 2017 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "bytes" + "encoding/binary" + "fmt" + "hash/crc32" + "io" + "math" + "os" + "reflect" + "strconv" + "sync" + "time" + "unsafe" + + "github.com/pkg/errors" + + "github.com/dgraph-io/badger/v4/pb" + "github.com/dgraph-io/ristretto/z" +) + +var ( + // ErrEOF indicates an end of file when trying to read from a memory mapped file + // and encountering the end of slice. + ErrEOF = errors.New("ErrEOF: End of file") + + // ErrCommitAfterFinish indicates that write batch commit was called after + // finish + ErrCommitAfterFinish = errors.New("Batch commit not permitted after finish") +) + +type Flags int + +const ( + // Sync indicates that O_DSYNC should be set on the underlying file, + // ensuring that data writes do not return until the data is flushed + // to disk. + Sync Flags = 1 << iota + // ReadOnly opens the underlying file on a read-only basis. + ReadOnly +) + +var ( + // This is O_DSYNC (datasync) on platforms that support it -- see file_unix.go + datasyncFileFlag = 0x0 + + // CastagnoliCrcTable is a CRC32 polynomial table + CastagnoliCrcTable = crc32.MakeTable(crc32.Castagnoli) +) + +// OpenExistingFile opens an existing file, errors if it doesn't exist. +func OpenExistingFile(filename string, flags Flags) (*os.File, error) { + openFlags := os.O_RDWR + if flags&ReadOnly != 0 { + openFlags = os.O_RDONLY + } + + if flags&Sync != 0 { + openFlags |= datasyncFileFlag + } + return os.OpenFile(filename, openFlags, 0) +} + +// CreateSyncedFile creates a new file (using O_EXCL), errors if it already existed. +func CreateSyncedFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR | os.O_CREATE | os.O_EXCL + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0600) +} + +// OpenSyncedFile creates the file if one doesn't exist. +func OpenSyncedFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR | os.O_CREATE + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0600) +} + +// OpenTruncFile opens the file with O_RDWR | O_CREATE | O_TRUNC +func OpenTruncFile(filename string, sync bool) (*os.File, error) { + flags := os.O_RDWR | os.O_CREATE | os.O_TRUNC + if sync { + flags |= datasyncFileFlag + } + return os.OpenFile(filename, flags, 0600) +} + +// SafeCopy does append(a[:0], src...). +func SafeCopy(a, src []byte) []byte { + return append(a[:0], src...) +} + +// Copy copies a byte slice and returns the copied slice. +func Copy(a []byte) []byte { + b := make([]byte, len(a)) + copy(b, a) + return b +} + +// KeyWithTs generates a new key by appending ts to key. +func KeyWithTs(key []byte, ts uint64) []byte { + out := make([]byte, len(key)+8) + copy(out, key) + binary.BigEndian.PutUint64(out[len(key):], math.MaxUint64-ts) + return out +} + +// ParseTs parses the timestamp from the key bytes. +func ParseTs(key []byte) uint64 { + if len(key) <= 8 { + return 0 + } + return math.MaxUint64 - binary.BigEndian.Uint64(key[len(key)-8:]) +} + +// CompareKeys checks the key without timestamp and checks the timestamp if keyNoTs +// is same. +// a would be sorted higher than aa if we use bytes.compare +// All keys should have timestamp. +func CompareKeys(key1, key2 []byte) int { + if cmp := bytes.Compare(key1[:len(key1)-8], key2[:len(key2)-8]); cmp != 0 { + return cmp + } + return bytes.Compare(key1[len(key1)-8:], key2[len(key2)-8:]) +} + +// ParseKey parses the actual key from the key bytes. +func ParseKey(key []byte) []byte { + if key == nil { + return nil + } + + return key[:len(key)-8] +} + +// SameKey checks for key equality ignoring the version timestamp suffix. +func SameKey(src, dst []byte) bool { + if len(src) != len(dst) { + return false + } + return bytes.Equal(ParseKey(src), ParseKey(dst)) +} + +// Slice holds a reusable buf, will reallocate if you request a larger size than ever before. +// One problem is with n distinct sizes in random order it'll reallocate log(n) times. +type Slice struct { + buf []byte +} + +// Resize reuses the Slice's buffer (or makes a new one) and returns a slice in that buffer of +// length sz. +func (s *Slice) Resize(sz int) []byte { + if cap(s.buf) < sz { + s.buf = make([]byte, sz) + } + return s.buf[0:sz] +} + +// FixedDuration returns a string representation of the given duration with the +// hours, minutes, and seconds. +func FixedDuration(d time.Duration) string { + str := fmt.Sprintf("%02ds", int(d.Seconds())%60) + if d >= time.Minute { + str = fmt.Sprintf("%02dm", int(d.Minutes())%60) + str + } + if d >= time.Hour { + str = fmt.Sprintf("%02dh", int(d.Hours())) + str + } + return str +} + +// Throttle allows a limited number of workers to run at a time. It also +// provides a mechanism to check for errors encountered by workers and wait for +// them to finish. +type Throttle struct { + once sync.Once + wg sync.WaitGroup + ch chan struct{} + errCh chan error + finishErr error +} + +// NewThrottle creates a new throttle with a max number of workers. +func NewThrottle(max int) *Throttle { + return &Throttle{ + ch: make(chan struct{}, max), + errCh: make(chan error, max), + } +} + +// Do should be called by workers before they start working. It blocks if there +// are already maximum number of workers working. If it detects an error from +// previously Done workers, it would return it. +func (t *Throttle) Do() error { + for { + select { + case t.ch <- struct{}{}: + t.wg.Add(1) + return nil + case err := <-t.errCh: + if err != nil { + return err + } + } + } +} + +// Done should be called by workers when they finish working. They can also +// pass the error status of work done. +func (t *Throttle) Done(err error) { + if err != nil { + t.errCh <- err + } + select { + case <-t.ch: + default: + panic("Throttle Do Done mismatch") + } + t.wg.Done() +} + +// Finish waits until all workers have finished working. It would return any error passed by Done. +// If Finish is called multiple time, it will wait for workers to finish only once(first time). +// From next calls, it will return same error as found on first call. +func (t *Throttle) Finish() error { + t.once.Do(func() { + t.wg.Wait() + close(t.ch) + close(t.errCh) + for err := range t.errCh { + if err != nil { + t.finishErr = err + return + } + } + }) + + return t.finishErr +} + +// U16ToBytes converts the given Uint16 to bytes +func U16ToBytes(v uint16) []byte { + var uBuf [2]byte + binary.BigEndian.PutUint16(uBuf[:], v) + return uBuf[:] +} + +// BytesToU16 converts the given byte slice to uint16 +func BytesToU16(b []byte) uint16 { + return binary.BigEndian.Uint16(b) +} + +// U32ToBytes converts the given Uint32 to bytes +func U32ToBytes(v uint32) []byte { + var uBuf [4]byte + binary.BigEndian.PutUint32(uBuf[:], v) + return uBuf[:] +} + +// BytesToU32 converts the given byte slice to uint32 +func BytesToU32(b []byte) uint32 { + return binary.BigEndian.Uint32(b) +} + +// U32SliceToBytes converts the given Uint32 slice to byte slice +func U32SliceToBytes(u32s []uint32) []byte { + if len(u32s) == 0 { + return nil + } + var b []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&b)) + hdr.Len = len(u32s) * 4 + hdr.Cap = hdr.Len + hdr.Data = uintptr(unsafe.Pointer(&u32s[0])) + return b +} + +// BytesToU32Slice converts the given byte slice to uint32 slice +func BytesToU32Slice(b []byte) []uint32 { + if len(b) == 0 { + return nil + } + var u32s []uint32 + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&u32s)) + hdr.Len = len(b) / 4 + hdr.Cap = hdr.Len + hdr.Data = uintptr(unsafe.Pointer(&b[0])) + return u32s +} + +// U64ToBytes converts the given Uint64 to bytes +func U64ToBytes(v uint64) []byte { + var uBuf [8]byte + binary.BigEndian.PutUint64(uBuf[:], v) + return uBuf[:] +} + +// BytesToU64 converts the given byte slice to uint64 +func BytesToU64(b []byte) uint64 { + return binary.BigEndian.Uint64(b) +} + +// U64SliceToBytes converts the given Uint64 slice to byte slice +func U64SliceToBytes(u64s []uint64) []byte { + if len(u64s) == 0 { + return nil + } + var b []byte + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&b)) + hdr.Len = len(u64s) * 8 + hdr.Cap = hdr.Len + hdr.Data = uintptr(unsafe.Pointer(&u64s[0])) + return b +} + +// BytesToU64Slice converts the given byte slice to uint64 slice +func BytesToU64Slice(b []byte) []uint64 { + if len(b) == 0 { + return nil + } + var u64s []uint64 + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&u64s)) + hdr.Len = len(b) / 8 + hdr.Cap = hdr.Len + hdr.Data = uintptr(unsafe.Pointer(&b[0])) + return u64s +} + +// page struct contains one underlying buffer. +type page struct { + buf []byte +} + +// PageBuffer consists of many pages. A page is a wrapper over []byte. PageBuffer can act as a +// replacement of bytes.Buffer. Instead of having single underlying buffer, it has multiple +// underlying buffers. Hence it avoids any copy during relocation(as happens in bytes.Buffer). +// PageBuffer allocates memory in pages. Once a page is full, it will allocate page with double the +// size of previous page. Its function are not thread safe. +type PageBuffer struct { + pages []*page + + length int // Length of PageBuffer. + nextPageSize int // Size of next page to be allocated. +} + +// NewPageBuffer returns a new PageBuffer with first page having size pageSize. +func NewPageBuffer(pageSize int) *PageBuffer { + b := &PageBuffer{} + b.pages = append(b.pages, &page{buf: make([]byte, 0, pageSize)}) + b.nextPageSize = pageSize * 2 + return b +} + +// Write writes data to PageBuffer b. It returns number of bytes written and any error encountered. +func (b *PageBuffer) Write(data []byte) (int, error) { + dataLen := len(data) + for { + cp := b.pages[len(b.pages)-1] // Current page. + + n := copy(cp.buf[len(cp.buf):cap(cp.buf)], data) + cp.buf = cp.buf[:len(cp.buf)+n] + b.length += n + + if len(data) == n { + break + } + data = data[n:] + + b.pages = append(b.pages, &page{buf: make([]byte, 0, b.nextPageSize)}) + b.nextPageSize *= 2 + } + + return dataLen, nil +} + +// WriteByte writes data byte to PageBuffer and returns any encountered error. +func (b *PageBuffer) WriteByte(data byte) error { + _, err := b.Write([]byte{data}) + return err +} + +// Len returns length of PageBuffer. +func (b *PageBuffer) Len() int { + return b.length +} + +// pageForOffset returns pageIdx and startIdx for the offset. +func (b *PageBuffer) pageForOffset(offset int) (int, int) { + AssertTrue(offset < b.length) + + var pageIdx, startIdx, sizeNow int + for i := 0; i < len(b.pages); i++ { + cp := b.pages[i] + + if sizeNow+len(cp.buf)-1 < offset { + sizeNow += len(cp.buf) + } else { + pageIdx = i + startIdx = offset - sizeNow + break + } + } + + return pageIdx, startIdx +} + +// Truncate truncates PageBuffer to length n. +func (b *PageBuffer) Truncate(n int) { + pageIdx, startIdx := b.pageForOffset(n) + // For simplicity of the code reject extra pages. These pages can be kept. + b.pages = b.pages[:pageIdx+1] + cp := b.pages[len(b.pages)-1] + cp.buf = cp.buf[:startIdx] + b.length = n +} + +// Bytes returns whole Buffer data as single []byte. +func (b *PageBuffer) Bytes() []byte { + buf := make([]byte, b.length) + written := 0 + for i := 0; i < len(b.pages); i++ { + written += copy(buf[written:], b.pages[i].buf) + } + + return buf +} + +// WriteTo writes whole buffer to w. It returns number of bytes written and any error encountered. +func (b *PageBuffer) WriteTo(w io.Writer) (int64, error) { + written := int64(0) + for i := 0; i < len(b.pages); i++ { + n, err := w.Write(b.pages[i].buf) + written += int64(n) + if err != nil { + return written, err + } + } + + return written, nil +} + +// NewReaderAt returns a reader which starts reading from offset in page buffer. +func (b *PageBuffer) NewReaderAt(offset int) *PageBufferReader { + pageIdx, startIdx := b.pageForOffset(offset) + + return &PageBufferReader{ + buf: b, + pageIdx: pageIdx, + startIdx: startIdx, + } +} + +// PageBufferReader is a reader for PageBuffer. +type PageBufferReader struct { + buf *PageBuffer // Underlying page buffer. + pageIdx int // Idx of page from where it will start reading. + startIdx int // Idx inside page - buf.pages[pageIdx] from where it will start reading. +} + +// Read reads upto len(p) bytes. It returns number of bytes read and any error encountered. +func (r *PageBufferReader) Read(p []byte) (int, error) { + // Check if there is enough to Read. + pc := len(r.buf.pages) + + read := 0 + for r.pageIdx < pc && read < len(p) { + cp := r.buf.pages[r.pageIdx] // Current Page. + endIdx := len(cp.buf) // Last Idx up to which we can read from this page. + + n := copy(p[read:], cp.buf[r.startIdx:endIdx]) + read += n + r.startIdx += n + + // Instead of len(cp.buf), we comparing with cap(cp.buf). This ensures that we move to next + // page only when we have read all data. Reading from last page is an edge case. We don't + // want to move to next page until last page is full to its capacity. + if r.startIdx >= cap(cp.buf) { + // We should move to next page. + r.pageIdx++ + r.startIdx = 0 + continue + } + + // When last page in not full to its capacity and we have read all data up to its + // length, just break out of the loop. + if r.pageIdx == pc-1 { + break + } + } + + if read == 0 && len(p) > 0 { + return read, io.EOF + } + + return read, nil +} + +const kvsz = int(unsafe.Sizeof(pb.KV{})) + +func NewKV(alloc *z.Allocator) *pb.KV { + if alloc == nil { + return &pb.KV{} + } + b := alloc.AllocateAligned(kvsz) + return (*pb.KV)(unsafe.Pointer(&b[0])) +} + +// IBytesToString converts size in bytes to human readable format. +// The code is taken from humanize library and changed to provide +// value upto custom decimal precision. +// IBytesToString(12312412, 1) -> 11.7 MiB +func IBytesToString(size uint64, precision int) string { + sizes := []string{"B", "KiB", "MiB", "GiB", "TiB", "PiB", "EiB"} + base := float64(1024) + if size < 10 { + return fmt.Sprintf("%d B", size) + } + e := math.Floor(math.Log(float64(size)) / math.Log(base)) + suffix := sizes[int(e)] + val := float64(size) / math.Pow(base, e) + f := "%." + strconv.Itoa(precision) + "f %s" + + return fmt.Sprintf(f, val, suffix) +} + +type RateMonitor struct { + start time.Time + lastSent uint64 + lastCapture time.Time + rates []float64 + idx int +} + +func NewRateMonitor(numSamples int) *RateMonitor { + return &RateMonitor{ + start: time.Now(), + rates: make([]float64, numSamples), + } +} + +const minRate = 0.0001 + +// Capture captures the current number of sent bytes. This number should be monotonically +// increasing. +func (rm *RateMonitor) Capture(sent uint64) { + diff := sent - rm.lastSent + dur := time.Since(rm.lastCapture) + rm.lastCapture, rm.lastSent = time.Now(), sent + + rate := float64(diff) / dur.Seconds() + if rate < minRate { + rate = minRate + } + rm.rates[rm.idx] = rate + rm.idx = (rm.idx + 1) % len(rm.rates) +} + +// Rate returns the average rate of transmission smoothed out by the number of samples. +func (rm *RateMonitor) Rate() uint64 { + var total float64 + var den float64 + for _, r := range rm.rates { + if r < minRate { + // Ignore this. We always set minRate, so this is a zero. + // Typically at the start of the rate monitor, we'd have zeros. + continue + } + total += r + den += 1.0 + } + if den < minRate { + return 0 + } + return uint64(total / den) +} diff --git a/vendor/github.com/dgraph-io/badger/v4/y/zstd.go b/vendor/github.com/dgraph-io/badger/v4/y/zstd.go new file mode 100644 index 00000000000..57018680a7f --- /dev/null +++ b/vendor/github.com/dgraph-io/badger/v4/y/zstd.go @@ -0,0 +1,64 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package y + +import ( + "sync" + + "github.com/klauspost/compress/zstd" +) + +var ( + decoder *zstd.Decoder + encoder *zstd.Encoder + + encOnce, decOnce sync.Once +) + +// ZSTDDecompress decompresses a block using ZSTD algorithm. +func ZSTDDecompress(dst, src []byte) ([]byte, error) { + decOnce.Do(func() { + var err error + decoder, err = zstd.NewReader(nil) + Check(err) + }) + return decoder.DecodeAll(src, dst[:0]) +} + +// ZSTDCompress compresses a block using ZSTD algorithm. +func ZSTDCompress(dst, src []byte, compressionLevel int) ([]byte, error) { + encOnce.Do(func() { + var err error + level := zstd.EncoderLevelFromZstd(compressionLevel) + encoder, err = zstd.NewWriter(nil, zstd.WithEncoderLevel(level)) + Check(err) + }) + return encoder.EncodeAll(src, dst[:0]), nil +} + +// ZSTDCompressBound returns the worst case size needed for a destination buffer. +// Klauspost ZSTD library does not provide any API for Compression Bound. This +// calculation is based on the DataDog ZSTD library. +// See https://pkg.go.dev/github.com/DataDog/zstd#CompressBound +func ZSTDCompressBound(srcSize int) int { + lowLimit := 128 << 10 // 128 kB + var margin int + if srcSize < lowLimit { + margin = (lowLimit - srcSize) >> 11 + } + return srcSize + (srcSize >> 8) + margin +} diff --git a/vendor/github.com/dgraph-io/ristretto/.deepsource.toml b/vendor/github.com/dgraph-io/ristretto/.deepsource.toml new file mode 100644 index 00000000000..40609eff3f1 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/.deepsource.toml @@ -0,0 +1,17 @@ +version = 1 + +test_patterns = [ + '**/*_test.go' +] + +exclude_patterns = [ + +] + +[[analyzers]] +name = 'go' +enabled = true + + + [analyzers.meta] + import_path = 'github.com/dgraph-io/ristretto' diff --git a/vendor/github.com/dgraph-io/ristretto/.go-version b/vendor/github.com/dgraph-io/ristretto/.go-version new file mode 100644 index 00000000000..b8f1e3fd3ee --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/.go-version @@ -0,0 +1 @@ +1.17.11 diff --git a/vendor/github.com/dgraph-io/ristretto/.golangci.yml b/vendor/github.com/dgraph-io/ristretto/.golangci.yml new file mode 100644 index 00000000000..7318e9a3b6a --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/.golangci.yml @@ -0,0 +1,23 @@ +run: + tests: false + skip-dirs: + - contrib + - sim + +linters-settings: + lll: + line-length: 120 + +linters: + disable-all: true + enable: + #- errcheck + #- ineffassign + - gas + #- gofmt + #- golint + #- gosimple + #- govet + - lll + #- varcheck + #- unused diff --git a/vendor/github.com/dgraph-io/ristretto/CHANGELOG.md b/vendor/github.com/dgraph-io/ristretto/CHANGELOG.md new file mode 100644 index 00000000000..3d18e39ed65 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/CHANGELOG.md @@ -0,0 +1,187 @@ +# Changelog +All notable changes to this project will be documented in this file. + +The format is based on [Keep a Changelog](http://keepachangelog.com/en/1.0.0/) +and this project will adhere to [Semantic Versioning](http://semver.org/spec/v2.0.0.html) starting v1.0.0. + +## [0.1.1] - 2022-10-12 + +[0.1.1]: https://github.com/dgraph-io/ristretto/compare/v0.1.0..v0.1.1 +This release fixes certain arm64 build issues in the z package. It also +incorporates CI steps in our repository. + +### Changed +- [chore(docs): Include SpiceDB in the list of projects using Ristretto (#285)](https://github.com/dgraph-io/ristretto/pull/311) + +### Added +- [Run CI Jobs via Github Actions #304](https://github.com/dgraph-io/ristretto/pull/304) + +### Fixed +- [fix(build): update x/sys dependency](https://github.com/dgraph-io/ristretto/pull/308) +- [fix(z): Address inconsistent mremap return arguments with arm64](https://github.com/dgraph-io/ristretto/pull/309) +- [fix(z): runtime error: index out of range for !amd64 env #287](https://github.com/dgraph-io/ristretto/pull/307) + +## [0.1.0] - 2021-06-03 + +[0.1.0]: https://github.com/dgraph-io/ristretto/compare/v0.0.3..v0.1.0 +This release contains bug fixes and improvements to Ristretto. It also contains +major updates to the z package. The z package contains types such as Tree (B+ +tree), Buffer, Mmap file, etc. All these types are used in Badger and Dgraph to +improve performance and reduce memory requirements. + +### Changed +- Make item public. Add a new onReject call for rejected items. (#180) + +### Added +- Use z.Buffer backing for B+ tree (#268) +- expose GetTTL function (#270) +- docs(README): Ristretto is production-ready. (#267) +- Add IterateKV (#265) +- feat(super-flags): Add GetPath method in superflags (#258) +- add GetDuration to SuperFlag (#248) +- add Has, GetFloat64, and GetInt64 to SuperFlag (#247) +- move SuperFlag to Ristretto (#246) +- add SuperFlagHelp tool to generate flag help text (#251) +- allow empty defaults in SuperFlag (#254) +- add mmaped b+ tree (#207) +- Add API to allow the MaxCost of an existing cache to be updated. (#200) +- Add OnExit handler which can be used for manual memory management (#183) +- Add life expectancy histogram (#182) +- Add mechanism to wait for items to be processed. (#184) + +### Fixed +- change expiration type from int64 to time.Time (#277) +- fix(buffer): make buffer capacity atleast defaultCapacity (#273) +- Fixes for z.PersistentTree (#272) +- Initialize persistent tree correctly (#271) +- use xxhash v2 (#266) +- update comments to correctly reflect counter space usage (#189) +- enable riscv64 builds (#264) +- Switch from log to glog (#263) +- Use Fibonacci for latency numbers +- cache: fix race when clearning a cache (#261) +- Check for keys without values in superflags (#259) +- chore(perf): using tags instead of runtime callers to improve the performance of leak detection (#255) +- fix(Flags): panic on user errors (#256) +- fix SuperFlagHelp newline (#252) +- fix(arm): Fix crashing under ARMv6 due to memory mis-alignment (#239) +- Fix incorrect unit test coverage depiction (#245) +- chore(histogram): adding percentile in histogram (#241) +- fix(windows): use filepath instead of path (#244) +- fix(MmapFile): Close the fd before deleting the file (#242) +- Fixes CGO_ENABLED=0 compilation error (#240) +- fix(build): fix build on non-amd64 architectures (#238) +- fix(b+tree): Do not double the size of btree (#237) +- fix(jemalloc): Fix the stats of jemalloc (#236) +- Don't print stuff, only return strings. +- Bring memclrNoHeapPointers to z (#235) +- increase number of buffers from 32 to 64 in allocator (#234) +- Set minSize to 1MB. +- Opt(btree): Use Go memory instead of mmap files +- Opt(btree): Lightweight stats calculation +- Put padding internally to z.Buffer +- Chore(z): Add SetTmpDir API to set the temp directory (#233) +- Add a BufferFrom +- Bring z.Allocator and z.AllocatorPool back +- Fix(z.Allocator): Make Allocator use Go memory +- Updated ZeroOut to use a simple for loop. (#231) +- Add concurrency back +- Add a test to check concurrency of Allocator. +- Fix(buffer): Expose padding by z.Buffer's APIs and fix test (#222) +- AllocateSlice should Truncate if the file is not big enough (#226) +- Zero out allocations for structs now that we're reusing Allocators. +- Fix the ristretto substring +- Deal with nil z.AllocatorPool +- Create an AllocatorPool class. +- chore(btree): clean NewTree API (#225) +- fix(MmapFile): Don't error out if fileSize > sz (#224) +- feat(btree): allow option to reset btree and mmaping it to specified file. (#223) +- Use mremap on Linux instead of munmap+mmap (#221) +- Reuse pages in B+ tree (#220) +- fix(allocator): make nil allocator return go byte slice (#217) +- fix(buffer): Make padding internal to z.buffer (#216) +- chore(buffer): add a parent directory field in z.Buffer (#215) +- Make Allocator concurrent +- Fix infinite loop in allocator (#214) +- Add trim func +- Use allocator pool. Turn off freelist. +- Add freelists to Allocator to reuse. +- make DeleteBelow delete values that are less than lo (#211) +- Avoid an unnecessary Load procedure in IncrementOffset. +- Add Stats method in Btree. +- chore(script): fix local test script (#210) +- fix(btree): Increase buffer size if needed. (#209) +- chore(btree): add occupancy ratio, search benchmark and compact bug fix (#208) +- Add licenses, remove prints, and fix a bug in compact +- Add IncrementOffset API for z.buffers (#206) +- Show count when printing histogram (#201) +- Zbuffer: Add LenNoPadding and make padding 8 bytes (#204) +- Allocate Go memory in case allocator is nil. +- Add leak detection via leak build flag and fix a leak during cache.Close. +- Add some APIs for allocator and buffer +- Sync before truncation or close. +- Handle nil MmapFile for Sync. +- Public methods must not panic after Close() (#202) +- Check for RD_ONLY correctly. +- Modify MmapFile APIs +- Add a bunch of APIs around MmapFile +- Move APIs for mmapfile creation over to z package. +- Add ZeroOut func +- Add SliceOffsets +- z: Add TotalSize method on bloom filter (#197) +- Add Msync func +- Buffer: Use 256 GB mmap size instead of MaxInt64 (#198) +- Add a simple test to check next2Pow +- Improve memory performance (#195) +- Have a way to automatically mmap a growing buffer (#196) +- Introduce Mmapped buffers and Merge Sort (#194) +- Add a way to access an allocator via reference. +- Use jemalloc.a to ensure compilation with the Go binary +- Fix up a build issue with ReadMemStats +- Add ReadMemStats function (#193) +- Allocator helps allocate memory to be used by unsafe structs (#192) +- Improve histogram output +- Move Closer from y to z (#191) +- Add histogram.Mean() method (#188) +- Introduce Calloc: Manual Memory Management via jemalloc (#186) + +## [0.0.3] - 2020-07-06 + +[0.0.3]: https://github.com/dgraph-io/ristretto/compare/v0.0.2..v0.0.3 + +### Changed + +### Added + +### Fixed + +- z: use MemHashString and xxhash.Sum64String ([#153][]) +- Check conflict key before updating expiration map. ([#154][]) +- Fix race condition in Cache.Clear ([#133][]) +- Improve handling of updated items ([#168][]) +- Fix droppedSets count while updating the item ([#171][]) + +## [0.0.2] - 2020-02-24 + +[0.0.2]: https://github.com/dgraph-io/ristretto/compare/v0.0.1..v0.0.2 + +### Added + +- Sets with TTL. ([#122][]) + +### Fixed + +- Fix the way metrics are handled for deletions. ([#111][]) +- Support nil `*Cache` values in `Clear` and `Close`. ([#119][]) +- Delete item immediately. ([#113][]) +- Remove key from policy after TTL eviction. ([#130][]) + +[#111]: https://github.com/dgraph-io/ristretto/issues/111 +[#113]: https://github.com/dgraph-io/ristretto/issues/113 +[#119]: https://github.com/dgraph-io/ristretto/issues/119 +[#122]: https://github.com/dgraph-io/ristretto/issues/122 +[#130]: https://github.com/dgraph-io/ristretto/issues/130 + +## 0.0.1 + +First release. Basic cache functionality based on a LFU policy. diff --git a/vendor/github.com/dgraph-io/ristretto/LICENSE b/vendor/github.com/dgraph-io/ristretto/LICENSE new file mode 100644 index 00000000000..d9a10c0d8e8 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/LICENSE @@ -0,0 +1,176 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS diff --git a/vendor/github.com/dgraph-io/ristretto/README.md b/vendor/github.com/dgraph-io/ristretto/README.md new file mode 100644 index 00000000000..e71ae3df9e9 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/README.md @@ -0,0 +1,220 @@ +# Ristretto +[![Go Doc](https://img.shields.io/badge/godoc-reference-blue.svg)](http://godoc.org/github.com/dgraph-io/ristretto) +[![ci-ristretto-tests](https://github.com/dgraph-io/ristretto/actions/workflows/ci-ristretto-tests.yml/badge.svg)](https://github.com/dgraph-io/ristretto/actions/workflows/ci-ristretto-tests.yml) +[![ci-ristretto-lint](https://github.com/dgraph-io/ristretto/actions/workflows/ci-ristretto-lint.yml/badge.svg)](https://github.com/dgraph-io/ristretto/actions/workflows/ci-ristretto-lint.yml) +[![Coverage Status](https://coveralls.io/repos/github/dgraph-io/ristretto/badge.svg?branch=main)](https://coveralls.io/github/dgraph-io/ristretto?branch=main) +[![Go Report Card](https://img.shields.io/badge/go%20report-A%2B-brightgreen)](https://goreportcard.com/report/github.com/dgraph-io/ristretto) + +Ristretto is a fast, concurrent cache library built with a focus on performance and correctness. + +The motivation to build Ristretto comes from the need for a contention-free +cache in [Dgraph][]. + +[Dgraph]: https://github.com/dgraph-io/dgraph + +## Features + +* **High Hit Ratios** - with our unique admission/eviction policy pairing, Ristretto's performance is best in class. + * **Eviction: SampledLFU** - on par with exact LRU and better performance on Search and Database traces. + * **Admission: TinyLFU** - extra performance with little memory overhead (12 bits per counter). +* **Fast Throughput** - we use a variety of techniques for managing contention and the result is excellent throughput. +* **Cost-Based Eviction** - any large new item deemed valuable can evict multiple smaller items (cost could be anything). +* **Fully Concurrent** - you can use as many goroutines as you want with little throughput degradation. +* **Metrics** - optional performance metrics for throughput, hit ratios, and other stats. +* **Simple API** - just figure out your ideal `Config` values and you're off and running. + +## Status + +Ristretto is production-ready. See [Projects using Ristretto](#projects-using-ristretto). + +## Table of Contents + +* [Usage](#Usage) + * [Example](#Example) + * [Config](#Config) + * [NumCounters](#Config) + * [MaxCost](#Config) + * [BufferItems](#Config) + * [Metrics](#Config) + * [OnEvict](#Config) + * [KeyToHash](#Config) + * [Cost](#Config) +* [Benchmarks](#Benchmarks) + * [Hit Ratios](#Hit-Ratios) + * [Search](#Search) + * [Database](#Database) + * [Looping](#Looping) + * [CODASYL](#CODASYL) + * [Throughput](#Throughput) + * [Mixed](#Mixed) + * [Read](#Read) + * [Write](#Write) +* [Projects using Ristretto](#projects-using-ristretto) +* [FAQ](#FAQ) + +## Usage + +### Example + +```go +func main() { + cache, err := ristretto.NewCache(&ristretto.Config{ + NumCounters: 1e7, // number of keys to track frequency of (10M). + MaxCost: 1 << 30, // maximum cost of cache (1GB). + BufferItems: 64, // number of keys per Get buffer. + }) + if err != nil { + panic(err) + } + + // set a value with a cost of 1 + cache.Set("key", "value", 1) + + // wait for value to pass through buffers + time.Sleep(10 * time.Millisecond) + + value, found := cache.Get("key") + if !found { + panic("missing value") + } + fmt.Println(value) + cache.Del("key") +} +``` + +### Config + +The `Config` struct is passed to `NewCache` when creating Ristretto instances (see the example above). + +**NumCounters** `int64` + +NumCounters is the number of 4-bit access counters to keep for admission and eviction. We've seen good performance in setting this to 10x the number of items you expect to keep in the cache when full. + +For example, if you expect each item to have a cost of 1 and MaxCost is 100, set NumCounters to 1,000. Or, if you use variable cost values but expect the cache to hold around 10,000 items when full, set NumCounters to 100,000. The important thing is the *number of unique items* in the full cache, not necessarily the MaxCost value. + +**MaxCost** `int64` + +MaxCost is how eviction decisions are made. For example, if MaxCost is 100 and a new item with a cost of 1 increases total cache cost to 101, 1 item will be evicted. + +MaxCost can also be used to denote the max size in bytes. For example, if MaxCost is 1,000,000 (1MB) and the cache is full with 1,000 1KB items, a new item (that's accepted) would cause 5 1KB items to be evicted. + +MaxCost could be anything as long as it matches how you're using the cost values when calling Set. + +**BufferItems** `int64` + +BufferItems is the size of the Get buffers. The best value we've found for this is 64. + +If for some reason you see Get performance decreasing with lots of contention (you shouldn't), try increasing this value in increments of 64. This is a fine-tuning mechanism and you probably won't have to touch this. + +**Metrics** `bool` + +Metrics is true when you want real-time logging of a variety of stats. The reason this is a Config flag is because there's a 10% throughput performance overhead. + +**OnEvict** `func(hashes [2]uint64, value interface{}, cost int64)` + +OnEvict is called for every eviction. + +**KeyToHash** `func(key interface{}) [2]uint64` + +KeyToHash is the hashing algorithm used for every key. If this is nil, Ristretto has a variety of [defaults depending on the underlying interface type](https://github.com/dgraph-io/ristretto/blob/master/z/z.go#L19-L41). + +Note that if you want 128bit hashes you should use the full `[2]uint64`, +otherwise just fill the `uint64` at the `0` position and it will behave like +any 64bit hash. + +**Cost** `func(value interface{}) int64` + +Cost is an optional function you can pass to the Config in order to evaluate +item cost at runtime, and only for the Set calls that aren't dropped (this is +useful if calculating item cost is particularly expensive and you don't want to +waste time on items that will be dropped anyways). + +To signal to Ristretto that you'd like to use this Cost function: + +1. Set the Cost field to a non-nil function. +2. When calling Set for new items or item updates, use a `cost` of 0. + +## Benchmarks + +The benchmarks can be found in https://github.com/dgraph-io/benchmarks/tree/master/cachebench/ristretto. + +### Hit Ratios + +#### Search + +This trace is described as "disk read accesses initiated by a large commercial +search engine in response to various web search requests." + +

+ +

+ +#### Database + +This trace is described as "a database server running at a commercial site +running an ERP application on top of a commercial database." + +

+ +

+ +#### Looping + +This trace demonstrates a looping access pattern. + +

+ +

+ +#### CODASYL + +This trace is described as "references to a CODASYL database for a one hour +period." + +

+ +

+ +### Throughput + +All throughput benchmarks were ran on an Intel Core i7-8700K (3.7GHz) with 16gb +of RAM. + +#### Mixed + +

+ +

+ +#### Read + +

+ +

+ +#### Write + +

+ +

+ +## Projects Using Ristretto + +Below is a list of known projects that use Ristretto: + +- [Badger](https://github.com/dgraph-io/badger) - Embeddable key-value DB in Go +- [Dgraph](https://github.com/dgraph-io/dgraph) - Horizontally scalable and distributed GraphQL database with a graph backend +- [Vitess](https://github.com/vitessio/vitess) - Database clustering system for horizontal scaling of MySQL +- [SpiceDB](https://github.com/authzed/spicedb) - Horizontally scalable permissions database + +## FAQ + +### How are you achieving this performance? What shortcuts are you taking? + +We go into detail in the [Ristretto blog post](https://blog.dgraph.io/post/introducing-ristretto-high-perf-go-cache/), but in short: our throughput performance can be attributed to a mix of batching and eventual consistency. Our hit ratio performance is mostly due to an excellent [admission policy](https://arxiv.org/abs/1512.00727) and SampledLFU eviction policy. + +As for "shortcuts," the only thing Ristretto does that could be construed as one is dropping some Set calls. That means a Set call for a new item (updates are guaranteed) isn't guaranteed to make it into the cache. The new item could be dropped at two points: when passing through the Set buffer or when passing through the admission policy. However, this doesn't affect hit ratios much at all as we expect the most popular items to be Set multiple times and eventually make it in the cache. + +### Is Ristretto distributed? + +No, it's just like any other Go library that you can import into your project and use in a single process. diff --git a/vendor/github.com/dgraph-io/ristretto/cache.go b/vendor/github.com/dgraph-io/ristretto/cache.go new file mode 100644 index 00000000000..7226245bcc4 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/cache.go @@ -0,0 +1,719 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +// Ristretto is a fast, fixed size, in-memory cache with a dual focus on +// throughput and hit ratio performance. You can easily add Ristretto to an +// existing system and keep the most valuable data where you need it. +package ristretto + +import ( + "bytes" + "errors" + "fmt" + "sync" + "sync/atomic" + "time" + "unsafe" + + "github.com/dgraph-io/ristretto/z" +) + +var ( + // TODO: find the optimal value for this or make it configurable + setBufSize = 32 * 1024 +) + +type itemCallback func(*Item) + +const itemSize = int64(unsafe.Sizeof(storeItem{})) + +// Cache is a thread-safe implementation of a hashmap with a TinyLFU admission +// policy and a Sampled LFU eviction policy. You can use the same Cache instance +// from as many goroutines as you want. +type Cache struct { + // store is the central concurrent hashmap where key-value items are stored. + store store + // policy determines what gets let in to the cache and what gets kicked out. + policy policy + // getBuf is a custom ring buffer implementation that gets pushed to when + // keys are read. + getBuf *ringBuffer + // setBuf is a buffer allowing us to batch/drop Sets during times of high + // contention. + setBuf chan *Item + // onEvict is called for item evictions. + onEvict itemCallback + // onReject is called when an item is rejected via admission policy. + onReject itemCallback + // onExit is called whenever a value goes out of scope from the cache. + onExit (func(interface{})) + // KeyToHash function is used to customize the key hashing algorithm. + // Each key will be hashed using the provided function. If keyToHash value + // is not set, the default keyToHash function is used. + keyToHash func(interface{}) (uint64, uint64) + // stop is used to stop the processItems goroutine. + stop chan struct{} + // indicates whether cache is closed. + isClosed bool + // cost calculates cost from a value. + cost func(value interface{}) int64 + // ignoreInternalCost dictates whether to ignore the cost of internally storing + // the item in the cost calculation. + ignoreInternalCost bool + // cleanupTicker is used to periodically check for entries whose TTL has passed. + cleanupTicker *time.Ticker + // Metrics contains a running log of important statistics like hits, misses, + // and dropped items. + Metrics *Metrics +} + +// Config is passed to NewCache for creating new Cache instances. +type Config struct { + // NumCounters determines the number of counters (keys) to keep that hold + // access frequency information. It's generally a good idea to have more + // counters than the max cache capacity, as this will improve eviction + // accuracy and subsequent hit ratios. + // + // For example, if you expect your cache to hold 1,000,000 items when full, + // NumCounters should be 10,000,000 (10x). Each counter takes up roughly + // 3 bytes (4 bits for each counter * 4 copies plus about a byte per + // counter for the bloom filter). Note that the number of counters is + // internally rounded up to the nearest power of 2, so the space usage + // may be a little larger than 3 bytes * NumCounters. + NumCounters int64 + // MaxCost can be considered as the cache capacity, in whatever units you + // choose to use. + // + // For example, if you want the cache to have a max capacity of 100MB, you + // would set MaxCost to 100,000,000 and pass an item's number of bytes as + // the `cost` parameter for calls to Set. If new items are accepted, the + // eviction process will take care of making room for the new item and not + // overflowing the MaxCost value. + MaxCost int64 + // BufferItems determines the size of Get buffers. + // + // Unless you have a rare use case, using `64` as the BufferItems value + // results in good performance. + BufferItems int64 + // Metrics determines whether cache statistics are kept during the cache's + // lifetime. There *is* some overhead to keeping statistics, so you should + // only set this flag to true when testing or throughput performance isn't a + // major factor. + Metrics bool + // OnEvict is called for every eviction and passes the hashed key, value, + // and cost to the function. + OnEvict func(item *Item) + // OnReject is called for every rejection done via the policy. + OnReject func(item *Item) + // OnExit is called whenever a value is removed from cache. This can be + // used to do manual memory deallocation. Would also be called on eviction + // and rejection of the value. + OnExit func(val interface{}) + // KeyToHash function is used to customize the key hashing algorithm. + // Each key will be hashed using the provided function. If keyToHash value + // is not set, the default keyToHash function is used. + KeyToHash func(key interface{}) (uint64, uint64) + // Cost evaluates a value and outputs a corresponding cost. This function + // is ran after Set is called for a new item or an item update with a cost + // param of 0. + Cost func(value interface{}) int64 + // IgnoreInternalCost set to true indicates to the cache that the cost of + // internally storing the value should be ignored. This is useful when the + // cost passed to set is not using bytes as units. Keep in mind that setting + // this to true will increase the memory usage. + IgnoreInternalCost bool +} + +type itemFlag byte + +const ( + itemNew itemFlag = iota + itemDelete + itemUpdate +) + +// Item is passed to setBuf so items can eventually be added to the cache. +type Item struct { + flag itemFlag + Key uint64 + Conflict uint64 + Value interface{} + Cost int64 + Expiration time.Time + wg *sync.WaitGroup +} + +// NewCache returns a new Cache instance and any configuration errors, if any. +func NewCache(config *Config) (*Cache, error) { + switch { + case config.NumCounters == 0: + return nil, errors.New("NumCounters can't be zero") + case config.MaxCost == 0: + return nil, errors.New("MaxCost can't be zero") + case config.BufferItems == 0: + return nil, errors.New("BufferItems can't be zero") + } + policy := newPolicy(config.NumCounters, config.MaxCost) + cache := &Cache{ + store: newStore(), + policy: policy, + getBuf: newRingBuffer(policy, config.BufferItems), + setBuf: make(chan *Item, setBufSize), + keyToHash: config.KeyToHash, + stop: make(chan struct{}), + cost: config.Cost, + ignoreInternalCost: config.IgnoreInternalCost, + cleanupTicker: time.NewTicker(time.Duration(bucketDurationSecs) * time.Second / 2), + } + cache.onExit = func(val interface{}) { + if config.OnExit != nil && val != nil { + config.OnExit(val) + } + } + cache.onEvict = func(item *Item) { + if config.OnEvict != nil { + config.OnEvict(item) + } + cache.onExit(item.Value) + } + cache.onReject = func(item *Item) { + if config.OnReject != nil { + config.OnReject(item) + } + cache.onExit(item.Value) + } + if cache.keyToHash == nil { + cache.keyToHash = z.KeyToHash + } + if config.Metrics { + cache.collectMetrics() + } + // NOTE: benchmarks seem to show that performance decreases the more + // goroutines we have running cache.processItems(), so 1 should + // usually be sufficient + go cache.processItems() + return cache, nil +} + +func (c *Cache) Wait() { + if c == nil || c.isClosed { + return + } + wg := &sync.WaitGroup{} + wg.Add(1) + c.setBuf <- &Item{wg: wg} + wg.Wait() +} + +// Get returns the value (if any) and a boolean representing whether the +// value was found or not. The value can be nil and the boolean can be true at +// the same time. +func (c *Cache) Get(key interface{}) (interface{}, bool) { + if c == nil || c.isClosed || key == nil { + return nil, false + } + keyHash, conflictHash := c.keyToHash(key) + c.getBuf.Push(keyHash) + value, ok := c.store.Get(keyHash, conflictHash) + if ok { + c.Metrics.add(hit, keyHash, 1) + } else { + c.Metrics.add(miss, keyHash, 1) + } + return value, ok +} + +// Set attempts to add the key-value item to the cache. If it returns false, +// then the Set was dropped and the key-value item isn't added to the cache. If +// it returns true, there's still a chance it could be dropped by the policy if +// its determined that the key-value item isn't worth keeping, but otherwise the +// item will be added and other items will be evicted in order to make room. +// +// To dynamically evaluate the items cost using the Config.Coster function, set +// the cost parameter to 0 and Coster will be ran when needed in order to find +// the items true cost. +func (c *Cache) Set(key, value interface{}, cost int64) bool { + return c.SetWithTTL(key, value, cost, 0*time.Second) +} + +// SetWithTTL works like Set but adds a key-value pair to the cache that will expire +// after the specified TTL (time to live) has passed. A zero value means the value never +// expires, which is identical to calling Set. A negative value is a no-op and the value +// is discarded. +func (c *Cache) SetWithTTL(key, value interface{}, cost int64, ttl time.Duration) bool { + if c == nil || c.isClosed || key == nil { + return false + } + + var expiration time.Time + switch { + case ttl == 0: + // No expiration. + break + case ttl < 0: + // Treat this a a no-op. + return false + default: + expiration = time.Now().Add(ttl) + } + + keyHash, conflictHash := c.keyToHash(key) + i := &Item{ + flag: itemNew, + Key: keyHash, + Conflict: conflictHash, + Value: value, + Cost: cost, + Expiration: expiration, + } + // cost is eventually updated. The expiration must also be immediately updated + // to prevent items from being prematurely removed from the map. + if prev, ok := c.store.Update(i); ok { + c.onExit(prev) + i.flag = itemUpdate + } + // Attempt to send item to policy. + select { + case c.setBuf <- i: + return true + default: + if i.flag == itemUpdate { + // Return true if this was an update operation since we've already + // updated the store. For all the other operations (set/delete), we + // return false which means the item was not inserted. + return true + } + c.Metrics.add(dropSets, keyHash, 1) + return false + } +} + +// Del deletes the key-value item from the cache if it exists. +func (c *Cache) Del(key interface{}) { + if c == nil || c.isClosed || key == nil { + return + } + keyHash, conflictHash := c.keyToHash(key) + // Delete immediately. + _, prev := c.store.Del(keyHash, conflictHash) + c.onExit(prev) + // If we've set an item, it would be applied slightly later. + // So we must push the same item to `setBuf` with the deletion flag. + // This ensures that if a set is followed by a delete, it will be + // applied in the correct order. + c.setBuf <- &Item{ + flag: itemDelete, + Key: keyHash, + Conflict: conflictHash, + } +} + +// GetTTL returns the TTL for the specified key and a bool that is true if the +// item was found and is not expired. +func (c *Cache) GetTTL(key interface{}) (time.Duration, bool) { + if c == nil || key == nil { + return 0, false + } + + keyHash, conflictHash := c.keyToHash(key) + if _, ok := c.store.Get(keyHash, conflictHash); !ok { + // not found + return 0, false + } + + expiration := c.store.Expiration(keyHash) + if expiration.IsZero() { + // found but no expiration + return 0, true + } + + if time.Now().After(expiration) { + // found but expired + return 0, false + } + + return time.Until(expiration), true +} + +// Close stops all goroutines and closes all channels. +func (c *Cache) Close() { + if c == nil || c.isClosed { + return + } + c.Clear() + + // Block until processItems goroutine is returned. + c.stop <- struct{}{} + close(c.stop) + close(c.setBuf) + c.policy.Close() + c.isClosed = true +} + +// Clear empties the hashmap and zeroes all policy counters. Note that this is +// not an atomic operation (but that shouldn't be a problem as it's assumed that +// Set/Get calls won't be occurring until after this). +func (c *Cache) Clear() { + if c == nil || c.isClosed { + return + } + // Block until processItems goroutine is returned. + c.stop <- struct{}{} + + // Clear out the setBuf channel. +loop: + for { + select { + case i := <-c.setBuf: + if i.wg != nil { + i.wg.Done() + continue + } + if i.flag != itemUpdate { + // In itemUpdate, the value is already set in the store. So, no need to call + // onEvict here. + c.onEvict(i) + } + default: + break loop + } + } + + // Clear value hashmap and policy data. + c.policy.Clear() + c.store.Clear(c.onEvict) + // Only reset metrics if they're enabled. + if c.Metrics != nil { + c.Metrics.Clear() + } + // Restart processItems goroutine. + go c.processItems() +} + +// MaxCost returns the max cost of the cache. +func (c *Cache) MaxCost() int64 { + if c == nil { + return 0 + } + return c.policy.MaxCost() +} + +// UpdateMaxCost updates the maxCost of an existing cache. +func (c *Cache) UpdateMaxCost(maxCost int64) { + if c == nil { + return + } + c.policy.UpdateMaxCost(maxCost) +} + +// processItems is ran by goroutines processing the Set buffer. +func (c *Cache) processItems() { + startTs := make(map[uint64]time.Time) + numToKeep := 100000 // TODO: Make this configurable via options. + + trackAdmission := func(key uint64) { + if c.Metrics == nil { + return + } + startTs[key] = time.Now() + if len(startTs) > numToKeep { + for k := range startTs { + if len(startTs) <= numToKeep { + break + } + delete(startTs, k) + } + } + } + onEvict := func(i *Item) { + if ts, has := startTs[i.Key]; has { + c.Metrics.trackEviction(int64(time.Since(ts) / time.Second)) + delete(startTs, i.Key) + } + if c.onEvict != nil { + c.onEvict(i) + } + } + + for { + select { + case i := <-c.setBuf: + if i.wg != nil { + i.wg.Done() + continue + } + // Calculate item cost value if new or update. + if i.Cost == 0 && c.cost != nil && i.flag != itemDelete { + i.Cost = c.cost(i.Value) + } + if !c.ignoreInternalCost { + // Add the cost of internally storing the object. + i.Cost += itemSize + } + + switch i.flag { + case itemNew: + victims, added := c.policy.Add(i.Key, i.Cost) + if added { + c.store.Set(i) + c.Metrics.add(keyAdd, i.Key, 1) + trackAdmission(i.Key) + } else { + c.onReject(i) + } + for _, victim := range victims { + victim.Conflict, victim.Value = c.store.Del(victim.Key, 0) + onEvict(victim) + } + + case itemUpdate: + c.policy.Update(i.Key, i.Cost) + + case itemDelete: + c.policy.Del(i.Key) // Deals with metrics updates. + _, val := c.store.Del(i.Key, i.Conflict) + c.onExit(val) + } + case <-c.cleanupTicker.C: + c.store.Cleanup(c.policy, onEvict) + case <-c.stop: + return + } + } +} + +// collectMetrics just creates a new *Metrics instance and adds the pointers +// to the cache and policy instances. +func (c *Cache) collectMetrics() { + c.Metrics = newMetrics() + c.policy.CollectMetrics(c.Metrics) +} + +type metricType int + +const ( + // The following 2 keep track of hits and misses. + hit = iota + miss + // The following 3 keep track of number of keys added, updated and evicted. + keyAdd + keyUpdate + keyEvict + // The following 2 keep track of cost of keys added and evicted. + costAdd + costEvict + // The following keep track of how many sets were dropped or rejected later. + dropSets + rejectSets + // The following 2 keep track of how many gets were kept and dropped on the + // floor. + dropGets + keepGets + // This should be the final enum. Other enums should be set before this. + doNotUse +) + +func stringFor(t metricType) string { + switch t { + case hit: + return "hit" + case miss: + return "miss" + case keyAdd: + return "keys-added" + case keyUpdate: + return "keys-updated" + case keyEvict: + return "keys-evicted" + case costAdd: + return "cost-added" + case costEvict: + return "cost-evicted" + case dropSets: + return "sets-dropped" + case rejectSets: + return "sets-rejected" // by policy. + case dropGets: + return "gets-dropped" + case keepGets: + return "gets-kept" + default: + return "unidentified" + } +} + +// Metrics is a snapshot of performance statistics for the lifetime of a cache instance. +type Metrics struct { + all [doNotUse][]*uint64 + + mu sync.RWMutex + life *z.HistogramData // Tracks the life expectancy of a key. +} + +func newMetrics() *Metrics { + s := &Metrics{ + life: z.NewHistogramData(z.HistogramBounds(1, 16)), + } + for i := 0; i < doNotUse; i++ { + s.all[i] = make([]*uint64, 256) + slice := s.all[i] + for j := range slice { + slice[j] = new(uint64) + } + } + return s +} + +func (p *Metrics) add(t metricType, hash, delta uint64) { + if p == nil { + return + } + valp := p.all[t] + // Avoid false sharing by padding at least 64 bytes of space between two + // atomic counters which would be incremented. + idx := (hash % 25) * 10 + atomic.AddUint64(valp[idx], delta) +} + +func (p *Metrics) get(t metricType) uint64 { + if p == nil { + return 0 + } + valp := p.all[t] + var total uint64 + for i := range valp { + total += atomic.LoadUint64(valp[i]) + } + return total +} + +// Hits is the number of Get calls where a value was found for the corresponding key. +func (p *Metrics) Hits() uint64 { + return p.get(hit) +} + +// Misses is the number of Get calls where a value was not found for the corresponding key. +func (p *Metrics) Misses() uint64 { + return p.get(miss) +} + +// KeysAdded is the total number of Set calls where a new key-value item was added. +func (p *Metrics) KeysAdded() uint64 { + return p.get(keyAdd) +} + +// KeysUpdated is the total number of Set calls where the value was updated. +func (p *Metrics) KeysUpdated() uint64 { + return p.get(keyUpdate) +} + +// KeysEvicted is the total number of keys evicted. +func (p *Metrics) KeysEvicted() uint64 { + return p.get(keyEvict) +} + +// CostAdded is the sum of costs that have been added (successful Set calls). +func (p *Metrics) CostAdded() uint64 { + return p.get(costAdd) +} + +// CostEvicted is the sum of all costs that have been evicted. +func (p *Metrics) CostEvicted() uint64 { + return p.get(costEvict) +} + +// SetsDropped is the number of Set calls that don't make it into internal +// buffers (due to contention or some other reason). +func (p *Metrics) SetsDropped() uint64 { + return p.get(dropSets) +} + +// SetsRejected is the number of Set calls rejected by the policy (TinyLFU). +func (p *Metrics) SetsRejected() uint64 { + return p.get(rejectSets) +} + +// GetsDropped is the number of Get counter increments that are dropped +// internally. +func (p *Metrics) GetsDropped() uint64 { + return p.get(dropGets) +} + +// GetsKept is the number of Get counter increments that are kept. +func (p *Metrics) GetsKept() uint64 { + return p.get(keepGets) +} + +// Ratio is the number of Hits over all accesses (Hits + Misses). This is the +// percentage of successful Get calls. +func (p *Metrics) Ratio() float64 { + if p == nil { + return 0.0 + } + hits, misses := p.get(hit), p.get(miss) + if hits == 0 && misses == 0 { + return 0.0 + } + return float64(hits) / float64(hits+misses) +} + +func (p *Metrics) trackEviction(numSeconds int64) { + if p == nil { + return + } + p.mu.Lock() + defer p.mu.Unlock() + p.life.Update(numSeconds) +} + +func (p *Metrics) LifeExpectancySeconds() *z.HistogramData { + if p == nil { + return nil + } + p.mu.RLock() + defer p.mu.RUnlock() + return p.life.Copy() +} + +// Clear resets all the metrics. +func (p *Metrics) Clear() { + if p == nil { + return + } + for i := 0; i < doNotUse; i++ { + for j := range p.all[i] { + atomic.StoreUint64(p.all[i][j], 0) + } + } + p.mu.Lock() + p.life = z.NewHistogramData(z.HistogramBounds(1, 16)) + p.mu.Unlock() +} + +// String returns a string representation of the metrics. +func (p *Metrics) String() string { + if p == nil { + return "" + } + var buf bytes.Buffer + for i := 0; i < doNotUse; i++ { + t := metricType(i) + fmt.Fprintf(&buf, "%s: %d ", stringFor(t), p.get(t)) + } + fmt.Fprintf(&buf, "gets-total: %d ", p.get(hit)+p.get(miss)) + fmt.Fprintf(&buf, "hit-ratio: %.2f", p.Ratio()) + return buf.String() +} diff --git a/vendor/github.com/dgraph-io/ristretto/policy.go b/vendor/github.com/dgraph-io/ristretto/policy.go new file mode 100644 index 00000000000..bf23f91fd9b --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/policy.go @@ -0,0 +1,423 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package ristretto + +import ( + "math" + "sync" + "sync/atomic" + + "github.com/dgraph-io/ristretto/z" +) + +const ( + // lfuSample is the number of items to sample when looking at eviction + // candidates. 5 seems to be the most optimal number [citation needed]. + lfuSample = 5 +) + +// policy is the interface encapsulating eviction/admission behavior. +// +// TODO: remove this interface and just rename defaultPolicy to policy, as we +// are probably only going to use/implement/maintain one policy. +type policy interface { + ringConsumer + // Add attempts to Add the key-cost pair to the Policy. It returns a slice + // of evicted keys and a bool denoting whether or not the key-cost pair + // was added. If it returns true, the key should be stored in cache. + Add(uint64, int64) ([]*Item, bool) + // Has returns true if the key exists in the Policy. + Has(uint64) bool + // Del deletes the key from the Policy. + Del(uint64) + // Cap returns the available capacity. + Cap() int64 + // Close stops all goroutines and closes all channels. + Close() + // Update updates the cost value for the key. + Update(uint64, int64) + // Cost returns the cost value of a key or -1 if missing. + Cost(uint64) int64 + // Optionally, set stats object to track how policy is performing. + CollectMetrics(*Metrics) + // Clear zeroes out all counters and clears hashmaps. + Clear() + // MaxCost returns the current max cost of the cache policy. + MaxCost() int64 + // UpdateMaxCost updates the max cost of the cache policy. + UpdateMaxCost(int64) +} + +func newPolicy(numCounters, maxCost int64) policy { + return newDefaultPolicy(numCounters, maxCost) +} + +type defaultPolicy struct { + sync.Mutex + admit *tinyLFU + evict *sampledLFU + itemsCh chan []uint64 + stop chan struct{} + isClosed bool + metrics *Metrics +} + +func newDefaultPolicy(numCounters, maxCost int64) *defaultPolicy { + p := &defaultPolicy{ + admit: newTinyLFU(numCounters), + evict: newSampledLFU(maxCost), + itemsCh: make(chan []uint64, 3), + stop: make(chan struct{}), + } + go p.processItems() + return p +} + +func (p *defaultPolicy) CollectMetrics(metrics *Metrics) { + p.metrics = metrics + p.evict.metrics = metrics +} + +type policyPair struct { + key uint64 + cost int64 +} + +func (p *defaultPolicy) processItems() { + for { + select { + case items := <-p.itemsCh: + p.Lock() + p.admit.Push(items) + p.Unlock() + case <-p.stop: + return + } + } +} + +func (p *defaultPolicy) Push(keys []uint64) bool { + if p.isClosed { + return false + } + + if len(keys) == 0 { + return true + } + + select { + case p.itemsCh <- keys: + p.metrics.add(keepGets, keys[0], uint64(len(keys))) + return true + default: + p.metrics.add(dropGets, keys[0], uint64(len(keys))) + return false + } +} + +// Add decides whether the item with the given key and cost should be accepted by +// the policy. It returns the list of victims that have been evicted and a boolean +// indicating whether the incoming item should be accepted. +func (p *defaultPolicy) Add(key uint64, cost int64) ([]*Item, bool) { + p.Lock() + defer p.Unlock() + + // Cannot add an item bigger than entire cache. + if cost > p.evict.getMaxCost() { + return nil, false + } + + // No need to go any further if the item is already in the cache. + if has := p.evict.updateIfHas(key, cost); has { + // An update does not count as an addition, so return false. + return nil, false + } + + // If the execution reaches this point, the key doesn't exist in the cache. + // Calculate the remaining room in the cache (usually bytes). + room := p.evict.roomLeft(cost) + if room >= 0 { + // There's enough room in the cache to store the new item without + // overflowing. Do that now and stop here. + p.evict.add(key, cost) + p.metrics.add(costAdd, key, uint64(cost)) + return nil, true + } + + // incHits is the hit count for the incoming item. + incHits := p.admit.Estimate(key) + // sample is the eviction candidate pool to be filled via random sampling. + // TODO: perhaps we should use a min heap here. Right now our time + // complexity is N for finding the min. Min heap should bring it down to + // O(lg N). + sample := make([]*policyPair, 0, lfuSample) + // As items are evicted they will be appended to victims. + victims := make([]*Item, 0) + + // Delete victims until there's enough space or a minKey is found that has + // more hits than incoming item. + for ; room < 0; room = p.evict.roomLeft(cost) { + // Fill up empty slots in sample. + sample = p.evict.fillSample(sample) + + // Find minimally used item in sample. + minKey, minHits, minId, minCost := uint64(0), int64(math.MaxInt64), 0, int64(0) + for i, pair := range sample { + // Look up hit count for sample key. + if hits := p.admit.Estimate(pair.key); hits < minHits { + minKey, minHits, minId, minCost = pair.key, hits, i, pair.cost + } + } + + // If the incoming item isn't worth keeping in the policy, reject. + if incHits < minHits { + p.metrics.add(rejectSets, key, 1) + return victims, false + } + + // Delete the victim from metadata. + p.evict.del(minKey) + + // Delete the victim from sample. + sample[minId] = sample[len(sample)-1] + sample = sample[:len(sample)-1] + // Store victim in evicted victims slice. + victims = append(victims, &Item{ + Key: minKey, + Conflict: 0, + Cost: minCost, + }) + } + + p.evict.add(key, cost) + p.metrics.add(costAdd, key, uint64(cost)) + return victims, true +} + +func (p *defaultPolicy) Has(key uint64) bool { + p.Lock() + _, exists := p.evict.keyCosts[key] + p.Unlock() + return exists +} + +func (p *defaultPolicy) Del(key uint64) { + p.Lock() + p.evict.del(key) + p.Unlock() +} + +func (p *defaultPolicy) Cap() int64 { + p.Lock() + capacity := int64(p.evict.getMaxCost() - p.evict.used) + p.Unlock() + return capacity +} + +func (p *defaultPolicy) Update(key uint64, cost int64) { + p.Lock() + p.evict.updateIfHas(key, cost) + p.Unlock() +} + +func (p *defaultPolicy) Cost(key uint64) int64 { + p.Lock() + if cost, found := p.evict.keyCosts[key]; found { + p.Unlock() + return cost + } + p.Unlock() + return -1 +} + +func (p *defaultPolicy) Clear() { + p.Lock() + p.admit.clear() + p.evict.clear() + p.Unlock() +} + +func (p *defaultPolicy) Close() { + if p.isClosed { + return + } + + // Block until the p.processItems goroutine returns. + p.stop <- struct{}{} + close(p.stop) + close(p.itemsCh) + p.isClosed = true +} + +func (p *defaultPolicy) MaxCost() int64 { + if p == nil || p.evict == nil { + return 0 + } + return p.evict.getMaxCost() +} + +func (p *defaultPolicy) UpdateMaxCost(maxCost int64) { + if p == nil || p.evict == nil { + return + } + p.evict.updateMaxCost(maxCost) +} + +// sampledLFU is an eviction helper storing key-cost pairs. +type sampledLFU struct { + // NOTE: align maxCost to 64-bit boundary for use with atomic. + // As per https://golang.org/pkg/sync/atomic/: "On ARM, x86-32, + // and 32-bit MIPS, it is the caller’s responsibility to arrange + // for 64-bit alignment of 64-bit words accessed atomically. + // The first word in a variable or in an allocated struct, array, + // or slice can be relied upon to be 64-bit aligned." + maxCost int64 + used int64 + metrics *Metrics + keyCosts map[uint64]int64 +} + +func newSampledLFU(maxCost int64) *sampledLFU { + return &sampledLFU{ + keyCosts: make(map[uint64]int64), + maxCost: maxCost, + } +} + +func (p *sampledLFU) getMaxCost() int64 { + return atomic.LoadInt64(&p.maxCost) +} + +func (p *sampledLFU) updateMaxCost(maxCost int64) { + atomic.StoreInt64(&p.maxCost, maxCost) +} + +func (p *sampledLFU) roomLeft(cost int64) int64 { + return p.getMaxCost() - (p.used + cost) +} + +func (p *sampledLFU) fillSample(in []*policyPair) []*policyPair { + if len(in) >= lfuSample { + return in + } + for key, cost := range p.keyCosts { + in = append(in, &policyPair{key, cost}) + if len(in) >= lfuSample { + return in + } + } + return in +} + +func (p *sampledLFU) del(key uint64) { + cost, ok := p.keyCosts[key] + if !ok { + return + } + p.used -= cost + delete(p.keyCosts, key) + p.metrics.add(costEvict, key, uint64(cost)) + p.metrics.add(keyEvict, key, 1) +} + +func (p *sampledLFU) add(key uint64, cost int64) { + p.keyCosts[key] = cost + p.used += cost +} + +func (p *sampledLFU) updateIfHas(key uint64, cost int64) bool { + if prev, found := p.keyCosts[key]; found { + // Update the cost of an existing key, but don't worry about evicting. + // Evictions will be handled the next time a new item is added. + p.metrics.add(keyUpdate, key, 1) + if prev > cost { + diff := prev - cost + p.metrics.add(costAdd, key, ^uint64(uint64(diff)-1)) + } else if cost > prev { + diff := cost - prev + p.metrics.add(costAdd, key, uint64(diff)) + } + p.used += cost - prev + p.keyCosts[key] = cost + return true + } + return false +} + +func (p *sampledLFU) clear() { + p.used = 0 + p.keyCosts = make(map[uint64]int64) +} + +// tinyLFU is an admission helper that keeps track of access frequency using +// tiny (4-bit) counters in the form of a count-min sketch. +// tinyLFU is NOT thread safe. +type tinyLFU struct { + freq *cmSketch + door *z.Bloom + incrs int64 + resetAt int64 +} + +func newTinyLFU(numCounters int64) *tinyLFU { + return &tinyLFU{ + freq: newCmSketch(numCounters), + door: z.NewBloomFilter(float64(numCounters), 0.01), + resetAt: numCounters, + } +} + +func (p *tinyLFU) Push(keys []uint64) { + for _, key := range keys { + p.Increment(key) + } +} + +func (p *tinyLFU) Estimate(key uint64) int64 { + hits := p.freq.Estimate(key) + if p.door.Has(key) { + hits++ + } + return hits +} + +func (p *tinyLFU) Increment(key uint64) { + // Flip doorkeeper bit if not already done. + if added := p.door.AddIfNotHas(key); !added { + // Increment count-min counter if doorkeeper bit is already set. + p.freq.Increment(key) + } + p.incrs++ + if p.incrs >= p.resetAt { + p.reset() + } +} + +func (p *tinyLFU) reset() { + // Zero out incrs. + p.incrs = 0 + // clears doorkeeper bits + p.door.Clear() + // halves count-min counters + p.freq.Reset() +} + +func (p *tinyLFU) clear() { + p.incrs = 0 + p.door.Clear() + p.freq.Clear() +} diff --git a/vendor/github.com/dgraph-io/ristretto/ring.go b/vendor/github.com/dgraph-io/ristretto/ring.go new file mode 100644 index 00000000000..5dbed4cc59c --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/ring.go @@ -0,0 +1,91 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package ristretto + +import ( + "sync" +) + +// ringConsumer is the user-defined object responsible for receiving and +// processing items in batches when buffers are drained. +type ringConsumer interface { + Push([]uint64) bool +} + +// ringStripe is a singular ring buffer that is not concurrent safe. +type ringStripe struct { + cons ringConsumer + data []uint64 + capa int +} + +func newRingStripe(cons ringConsumer, capa int64) *ringStripe { + return &ringStripe{ + cons: cons, + data: make([]uint64, 0, capa), + capa: int(capa), + } +} + +// Push appends an item in the ring buffer and drains (copies items and +// sends to Consumer) if full. +func (s *ringStripe) Push(item uint64) { + s.data = append(s.data, item) + // Decide if the ring buffer should be drained. + if len(s.data) >= s.capa { + // Send elements to consumer and create a new ring stripe. + if s.cons.Push(s.data) { + s.data = make([]uint64, 0, s.capa) + } else { + s.data = s.data[:0] + } + } +} + +// ringBuffer stores multiple buffers (stripes) and distributes Pushed items +// between them to lower contention. +// +// This implements the "batching" process described in the BP-Wrapper paper +// (section III part A). +type ringBuffer struct { + pool *sync.Pool +} + +// newRingBuffer returns a striped ring buffer. The Consumer in ringConfig will +// be called when individual stripes are full and need to drain their elements. +func newRingBuffer(cons ringConsumer, capa int64) *ringBuffer { + // LOSSY buffers use a very simple sync.Pool for concurrently reusing + // stripes. We do lose some stripes due to GC (unheld items in sync.Pool + // are cleared), but the performance gains generally outweigh the small + // percentage of elements lost. The performance primarily comes from + // low-level runtime functions used in the standard library that aren't + // available to us (such as runtime_procPin()). + return &ringBuffer{ + pool: &sync.Pool{ + New: func() interface{} { return newRingStripe(cons, capa) }, + }, + } +} + +// Push adds an element to one of the internal stripes and possibly drains if +// the stripe becomes full. +func (b *ringBuffer) Push(item uint64) { + // Reuse or create a new stripe. + stripe := b.pool.Get().(*ringStripe) + stripe.Push(item) + b.pool.Put(stripe) +} diff --git a/vendor/github.com/dgraph-io/ristretto/sketch.go b/vendor/github.com/dgraph-io/ristretto/sketch.go new file mode 100644 index 00000000000..6368d2bde00 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/sketch.go @@ -0,0 +1,156 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +// This package includes multiple probabalistic data structures needed for +// admission/eviction metadata. Most are Counting Bloom Filter variations, but +// a caching-specific feature that is also required is a "freshness" mechanism, +// which basically serves as a "lifetime" process. This freshness mechanism +// was described in the original TinyLFU paper [1], but other mechanisms may +// be better suited for certain data distributions. +// +// [1]: https://arxiv.org/abs/1512.00727 +package ristretto + +import ( + "fmt" + "math/rand" + "time" +) + +// cmSketch is a Count-Min sketch implementation with 4-bit counters, heavily +// based on Damian Gryski's CM4 [1]. +// +// [1]: https://github.com/dgryski/go-tinylfu/blob/master/cm4.go +type cmSketch struct { + rows [cmDepth]cmRow + seed [cmDepth]uint64 + mask uint64 +} + +const ( + // cmDepth is the number of counter copies to store (think of it as rows). + cmDepth = 4 +) + +func newCmSketch(numCounters int64) *cmSketch { + if numCounters == 0 { + panic("cmSketch: bad numCounters") + } + // Get the next power of 2 for better cache performance. + numCounters = next2Power(numCounters) + sketch := &cmSketch{mask: uint64(numCounters - 1)} + // Initialize rows of counters and seeds. + // Cryptographic precision not needed + source := rand.New(rand.NewSource(time.Now().UnixNano())) //nolint:gosec + for i := 0; i < cmDepth; i++ { + sketch.seed[i] = source.Uint64() + sketch.rows[i] = newCmRow(numCounters) + } + return sketch +} + +// Increment increments the count(ers) for the specified key. +func (s *cmSketch) Increment(hashed uint64) { + for i := range s.rows { + s.rows[i].increment((hashed ^ s.seed[i]) & s.mask) + } +} + +// Estimate returns the value of the specified key. +func (s *cmSketch) Estimate(hashed uint64) int64 { + min := byte(255) + for i := range s.rows { + val := s.rows[i].get((hashed ^ s.seed[i]) & s.mask) + if val < min { + min = val + } + } + return int64(min) +} + +// Reset halves all counter values. +func (s *cmSketch) Reset() { + for _, r := range s.rows { + r.reset() + } +} + +// Clear zeroes all counters. +func (s *cmSketch) Clear() { + for _, r := range s.rows { + r.clear() + } +} + +// cmRow is a row of bytes, with each byte holding two counters. +type cmRow []byte + +func newCmRow(numCounters int64) cmRow { + return make(cmRow, numCounters/2) +} + +func (r cmRow) get(n uint64) byte { + return byte(r[n/2]>>((n&1)*4)) & 0x0f +} + +func (r cmRow) increment(n uint64) { + // Index of the counter. + i := n / 2 + // Shift distance (even 0, odd 4). + s := (n & 1) * 4 + // Counter value. + v := (r[i] >> s) & 0x0f + // Only increment if not max value (overflow wrap is bad for LFU). + if v < 15 { + r[i] += 1 << s + } +} + +func (r cmRow) reset() { + // Halve each counter. + for i := range r { + r[i] = (r[i] >> 1) & 0x77 + } +} + +func (r cmRow) clear() { + // Zero each counter. + for i := range r { + r[i] = 0 + } +} + +func (r cmRow) string() string { + s := "" + for i := uint64(0); i < uint64(len(r)*2); i++ { + s += fmt.Sprintf("%02d ", (r[(i/2)]>>((i&1)*4))&0x0f) + } + s = s[:len(s)-1] + return s +} + +// next2Power rounds x up to the next power of 2, if it's not already one. +func next2Power(x int64) int64 { + x-- + x |= x >> 1 + x |= x >> 2 + x |= x >> 4 + x |= x >> 8 + x |= x >> 16 + x |= x >> 32 + x++ + return x +} diff --git a/vendor/github.com/dgraph-io/ristretto/store.go b/vendor/github.com/dgraph-io/ristretto/store.go new file mode 100644 index 00000000000..e42a98b787f --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/store.go @@ -0,0 +1,242 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package ristretto + +import ( + "sync" + "time" +) + +// TODO: Do we need this to be a separate struct from Item? +type storeItem struct { + key uint64 + conflict uint64 + value interface{} + expiration time.Time +} + +// store is the interface fulfilled by all hash map implementations in this +// file. Some hash map implementations are better suited for certain data +// distributions than others, so this allows us to abstract that out for use +// in Ristretto. +// +// Every store is safe for concurrent usage. +type store interface { + // Get returns the value associated with the key parameter. + Get(uint64, uint64) (interface{}, bool) + // Expiration returns the expiration time for this key. + Expiration(uint64) time.Time + // Set adds the key-value pair to the Map or updates the value if it's + // already present. The key-value pair is passed as a pointer to an + // item object. + Set(*Item) + // Del deletes the key-value pair from the Map. + Del(uint64, uint64) (uint64, interface{}) + // Update attempts to update the key with a new value and returns true if + // successful. + Update(*Item) (interface{}, bool) + // Cleanup removes items that have an expired TTL. + Cleanup(policy policy, onEvict itemCallback) + // Clear clears all contents of the store. + Clear(onEvict itemCallback) +} + +// newStore returns the default store implementation. +func newStore() store { + return newShardedMap() +} + +const numShards uint64 = 256 + +type shardedMap struct { + shards []*lockedMap + expiryMap *expirationMap +} + +func newShardedMap() *shardedMap { + sm := &shardedMap{ + shards: make([]*lockedMap, int(numShards)), + expiryMap: newExpirationMap(), + } + for i := range sm.shards { + sm.shards[i] = newLockedMap(sm.expiryMap) + } + return sm +} + +func (sm *shardedMap) Get(key, conflict uint64) (interface{}, bool) { + return sm.shards[key%numShards].get(key, conflict) +} + +func (sm *shardedMap) Expiration(key uint64) time.Time { + return sm.shards[key%numShards].Expiration(key) +} + +func (sm *shardedMap) Set(i *Item) { + if i == nil { + // If item is nil make this Set a no-op. + return + } + + sm.shards[i.Key%numShards].Set(i) +} + +func (sm *shardedMap) Del(key, conflict uint64) (uint64, interface{}) { + return sm.shards[key%numShards].Del(key, conflict) +} + +func (sm *shardedMap) Update(newItem *Item) (interface{}, bool) { + return sm.shards[newItem.Key%numShards].Update(newItem) +} + +func (sm *shardedMap) Cleanup(policy policy, onEvict itemCallback) { + sm.expiryMap.cleanup(sm, policy, onEvict) +} + +func (sm *shardedMap) Clear(onEvict itemCallback) { + for i := uint64(0); i < numShards; i++ { + sm.shards[i].Clear(onEvict) + } +} + +type lockedMap struct { + sync.RWMutex + data map[uint64]storeItem + em *expirationMap +} + +func newLockedMap(em *expirationMap) *lockedMap { + return &lockedMap{ + data: make(map[uint64]storeItem), + em: em, + } +} + +func (m *lockedMap) get(key, conflict uint64) (interface{}, bool) { + m.RLock() + item, ok := m.data[key] + m.RUnlock() + if !ok { + return nil, false + } + if conflict != 0 && (conflict != item.conflict) { + return nil, false + } + + // Handle expired items. + if !item.expiration.IsZero() && time.Now().After(item.expiration) { + return nil, false + } + return item.value, true +} + +func (m *lockedMap) Expiration(key uint64) time.Time { + m.RLock() + defer m.RUnlock() + return m.data[key].expiration +} + +func (m *lockedMap) Set(i *Item) { + if i == nil { + // If the item is nil make this Set a no-op. + return + } + + m.Lock() + defer m.Unlock() + item, ok := m.data[i.Key] + + if ok { + // The item existed already. We need to check the conflict key and reject the + // update if they do not match. Only after that the expiration map is updated. + if i.Conflict != 0 && (i.Conflict != item.conflict) { + return + } + m.em.update(i.Key, i.Conflict, item.expiration, i.Expiration) + } else { + // The value is not in the map already. There's no need to return anything. + // Simply add the expiration map. + m.em.add(i.Key, i.Conflict, i.Expiration) + } + + m.data[i.Key] = storeItem{ + key: i.Key, + conflict: i.Conflict, + value: i.Value, + expiration: i.Expiration, + } +} + +func (m *lockedMap) Del(key, conflict uint64) (uint64, interface{}) { + m.Lock() + item, ok := m.data[key] + if !ok { + m.Unlock() + return 0, nil + } + if conflict != 0 && (conflict != item.conflict) { + m.Unlock() + return 0, nil + } + + if !item.expiration.IsZero() { + m.em.del(key, item.expiration) + } + + delete(m.data, key) + m.Unlock() + return item.conflict, item.value +} + +func (m *lockedMap) Update(newItem *Item) (interface{}, bool) { + m.Lock() + item, ok := m.data[newItem.Key] + if !ok { + m.Unlock() + return nil, false + } + if newItem.Conflict != 0 && (newItem.Conflict != item.conflict) { + m.Unlock() + return nil, false + } + + m.em.update(newItem.Key, newItem.Conflict, item.expiration, newItem.Expiration) + m.data[newItem.Key] = storeItem{ + key: newItem.Key, + conflict: newItem.Conflict, + value: newItem.Value, + expiration: newItem.Expiration, + } + + m.Unlock() + return item.value, true +} + +func (m *lockedMap) Clear(onEvict itemCallback) { + m.Lock() + i := &Item{} + if onEvict != nil { + for _, si := range m.data { + i.Key = si.key + i.Conflict = si.conflict + i.Value = si.value + onEvict(i) + } + } + m.data = make(map[uint64]storeItem) + m.Unlock() +} diff --git a/vendor/github.com/dgraph-io/ristretto/ttl.go b/vendor/github.com/dgraph-io/ristretto/ttl.go new file mode 100644 index 00000000000..337976ad438 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/ttl.go @@ -0,0 +1,147 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package ristretto + +import ( + "sync" + "time" +) + +var ( + // TODO: find the optimal value or make it configurable. + bucketDurationSecs = int64(5) +) + +func storageBucket(t time.Time) int64 { + return (t.Unix() / bucketDurationSecs) + 1 +} + +func cleanupBucket(t time.Time) int64 { + // The bucket to cleanup is always behind the storage bucket by one so that + // no elements in that bucket (which might not have expired yet) are deleted. + return storageBucket(t) - 1 +} + +// bucket type is a map of key to conflict. +type bucket map[uint64]uint64 + +// expirationMap is a map of bucket number to the corresponding bucket. +type expirationMap struct { + sync.RWMutex + buckets map[int64]bucket +} + +func newExpirationMap() *expirationMap { + return &expirationMap{ + buckets: make(map[int64]bucket), + } +} + +func (m *expirationMap) add(key, conflict uint64, expiration time.Time) { + if m == nil { + return + } + + // Items that don't expire don't need to be in the expiration map. + if expiration.IsZero() { + return + } + + bucketNum := storageBucket(expiration) + m.Lock() + defer m.Unlock() + + b, ok := m.buckets[bucketNum] + if !ok { + b = make(bucket) + m.buckets[bucketNum] = b + } + b[key] = conflict +} + +func (m *expirationMap) update(key, conflict uint64, oldExpTime, newExpTime time.Time) { + if m == nil { + return + } + + m.Lock() + defer m.Unlock() + + oldBucketNum := storageBucket(oldExpTime) + oldBucket, ok := m.buckets[oldBucketNum] + if ok { + delete(oldBucket, key) + } + + newBucketNum := storageBucket(newExpTime) + newBucket, ok := m.buckets[newBucketNum] + if !ok { + newBucket = make(bucket) + m.buckets[newBucketNum] = newBucket + } + newBucket[key] = conflict +} + +func (m *expirationMap) del(key uint64, expiration time.Time) { + if m == nil { + return + } + + bucketNum := storageBucket(expiration) + m.Lock() + defer m.Unlock() + _, ok := m.buckets[bucketNum] + if !ok { + return + } + delete(m.buckets[bucketNum], key) +} + +// cleanup removes all the items in the bucket that was just completed. It deletes +// those items from the store, and calls the onEvict function on those items. +// This function is meant to be called periodically. +func (m *expirationMap) cleanup(store store, policy policy, onEvict itemCallback) { + if m == nil { + return + } + + m.Lock() + now := time.Now() + bucketNum := cleanupBucket(now) + keys := m.buckets[bucketNum] + delete(m.buckets, bucketNum) + m.Unlock() + + for key, conflict := range keys { + // Sanity check. Verify that the store agrees that this key is expired. + if store.Expiration(key).After(now) { + continue + } + + cost := policy.Cost(key) + policy.Del(key) + _, value := store.Del(key, conflict) + + if onEvict != nil { + onEvict(&Item{Key: key, + Conflict: conflict, + Value: value, + Cost: cost, + }) + } + } +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/LICENSE b/vendor/github.com/dgraph-io/ristretto/z/LICENSE new file mode 100644 index 00000000000..0860cbfe85e --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/LICENSE @@ -0,0 +1,64 @@ +bbloom.go + +// The MIT License (MIT) +// Copyright (c) 2014 Andreas Briese, eduToolbox@Bri-C GmbH, Sarstedt + +// Permission is hereby granted, free of charge, to any person obtaining a copy of +// this software and associated documentation files (the "Software"), to deal in +// the Software without restriction, including without limitation the rights to +// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of +// the Software, and to permit persons to whom the Software is furnished to do so, +// subject to the following conditions: + +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. + +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS +// FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR +// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER +// IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN +// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +rtutil.go + +// MIT License + +// Copyright (c) 2019 Ewan Chou + +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: + +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. + +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. + +Modifications: + +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + diff --git a/vendor/github.com/dgraph-io/ristretto/z/README.md b/vendor/github.com/dgraph-io/ristretto/z/README.md new file mode 100644 index 00000000000..6d77e146ebd --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/README.md @@ -0,0 +1,129 @@ +## bbloom: a bitset Bloom filter for go/golang +=== + +package implements a fast bloom filter with real 'bitset' and JSONMarshal/JSONUnmarshal to store/reload the Bloom filter. + +NOTE: the package uses unsafe.Pointer to set and read the bits from the bitset. If you're uncomfortable with using the unsafe package, please consider using my bloom filter package at github.com/AndreasBriese/bloom + +=== + +changelog 11/2015: new thread safe methods AddTS(), HasTS(), AddIfNotHasTS() following a suggestion from Srdjan Marinovic (github @a-little-srdjan), who used this to code a bloomfilter cache. + +This bloom filter was developed to strengthen a website-log database and was tested and optimized for this log-entry mask: "2014/%02i/%02i %02i:%02i:%02i /info.html". +Nonetheless bbloom should work with any other form of entries. + +~~Hash function is a modified Berkeley DB sdbm hash (to optimize for smaller strings). sdbm http://www.cse.yorku.ca/~oz/hash.html~~ + +Found sipHash (SipHash-2-4, a fast short-input PRF created by Jean-Philippe Aumasson and Daniel J. Bernstein.) to be about as fast. sipHash had been ported by Dimtry Chestnyk to Go (github.com/dchest/siphash ) + +Minimum hashset size is: 512 ([4]uint64; will be set automatically). + +###install + +```sh +go get github.com/AndreasBriese/bbloom +``` + +###test ++ change to folder ../bbloom ++ create wordlist in file "words.txt" (you might use `python permut.py`) ++ run 'go test -bench=.' within the folder + +```go +go test -bench=. +``` + +~~If you've installed the GOCONVEY TDD-framework http://goconvey.co/ you can run the tests automatically.~~ + +using go's testing framework now (have in mind that the op timing is related to 65536 operations of Add, Has, AddIfNotHas respectively) + +### usage + +after installation add + +```go +import ( + ... + "github.com/AndreasBriese/bbloom" + ... + ) +``` + +at your header. In the program use + +```go +// create a bloom filter for 65536 items and 1 % wrong-positive ratio +bf := bbloom.New(float64(1<<16), float64(0.01)) + +// or +// create a bloom filter with 650000 for 65536 items and 7 locs per hash explicitly +// bf = bbloom.New(float64(650000), float64(7)) +// or +bf = bbloom.New(650000.0, 7.0) + +// add one item +bf.Add([]byte("butter")) + +// Number of elements added is exposed now +// Note: ElemNum will not be included in JSON export (for compatability to older version) +nOfElementsInFilter := bf.ElemNum + +// check if item is in the filter +isIn := bf.Has([]byte("butter")) // should be true +isNotIn := bf.Has([]byte("Butter")) // should be false + +// 'add only if item is new' to the bloomfilter +added := bf.AddIfNotHas([]byte("butter")) // should be false because 'butter' is already in the set +added = bf.AddIfNotHas([]byte("buTTer")) // should be true because 'buTTer' is new + +// thread safe versions for concurrent use: AddTS, HasTS, AddIfNotHasTS +// add one item +bf.AddTS([]byte("peanutbutter")) +// check if item is in the filter +isIn = bf.HasTS([]byte("peanutbutter")) // should be true +isNotIn = bf.HasTS([]byte("peanutButter")) // should be false +// 'add only if item is new' to the bloomfilter +added = bf.AddIfNotHasTS([]byte("butter")) // should be false because 'peanutbutter' is already in the set +added = bf.AddIfNotHasTS([]byte("peanutbuTTer")) // should be true because 'penutbuTTer' is new + +// convert to JSON ([]byte) +Json := bf.JSONMarshal() + +// bloomfilters Mutex is exposed for external un-/locking +// i.e. mutex lock while doing JSON conversion +bf.Mtx.Lock() +Json = bf.JSONMarshal() +bf.Mtx.Unlock() + +// restore a bloom filter from storage +bfNew := bbloom.JSONUnmarshal(Json) + +isInNew := bfNew.Has([]byte("butter")) // should be true +isNotInNew := bfNew.Has([]byte("Butter")) // should be false + +``` + +to work with the bloom filter. + +### why 'fast'? + +It's about 3 times faster than William Fitzgeralds bitset bloom filter https://github.com/willf/bloom . And it is about so fast as my []bool set variant for Boom filters (see https://github.com/AndreasBriese/bloom ) but having a 8times smaller memory footprint: + + + Bloom filter (filter size 524288, 7 hashlocs) + github.com/AndreasBriese/bbloom 'Add' 65536 items (10 repetitions): 6595800 ns (100 ns/op) + github.com/AndreasBriese/bbloom 'Has' 65536 items (10 repetitions): 5986600 ns (91 ns/op) + github.com/AndreasBriese/bloom 'Add' 65536 items (10 repetitions): 6304684 ns (96 ns/op) + github.com/AndreasBriese/bloom 'Has' 65536 items (10 repetitions): 6568663 ns (100 ns/op) + + github.com/willf/bloom 'Add' 65536 items (10 repetitions): 24367224 ns (371 ns/op) + github.com/willf/bloom 'Test' 65536 items (10 repetitions): 21881142 ns (333 ns/op) + github.com/dataence/bloom/standard 'Add' 65536 items (10 repetitions): 23041644 ns (351 ns/op) + github.com/dataence/bloom/standard 'Check' 65536 items (10 repetitions): 19153133 ns (292 ns/op) + github.com/cabello/bloom 'Add' 65536 items (10 repetitions): 131921507 ns (2012 ns/op) + github.com/cabello/bloom 'Contains' 65536 items (10 repetitions): 131108962 ns (2000 ns/op) + +(on MBPro15 OSX10.8.5 i7 4Core 2.4Ghz) + + +With 32bit bloom filters (bloom32) using modified sdbm, bloom32 does hashing with only 2 bit shifts, one xor and one substraction per byte. smdb is about as fast as fnv64a but gives less collisions with the dataset (see mask above). bloom.New(float64(10 * 1<<16),float64(7)) populated with 1<<16 random items from the dataset (see above) and tested against the rest results in less than 0.05% collisions. diff --git a/vendor/github.com/dgraph-io/ristretto/z/allocator.go b/vendor/github.com/dgraph-io/ristretto/z/allocator.go new file mode 100644 index 00000000000..eae0f834490 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/allocator.go @@ -0,0 +1,403 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "bytes" + "fmt" + "math" + "math/bits" + "math/rand" + "strings" + "sync" + "sync/atomic" + "time" + "unsafe" + + "github.com/dustin/go-humanize" +) + +// Allocator amortizes the cost of small allocations by allocating memory in +// bigger chunks. Internally it uses z.Calloc to allocate memory. Once +// allocated, the memory is not moved, so it is safe to use the allocated bytes +// to unsafe cast them to Go struct pointers. Maintaining a freelist is slow. +// Instead, Allocator only allocates memory, with the idea that finally we +// would just release the entire Allocator. +type Allocator struct { + sync.Mutex + compIdx uint64 // Stores bufIdx in 32 MSBs and posIdx in 32 LSBs. + buffers [][]byte + Ref uint64 + Tag string +} + +// allocs keeps references to all Allocators, so we can safely discard them later. +var allocsMu *sync.Mutex +var allocRef uint64 +var allocs map[uint64]*Allocator +var calculatedLog2 []int + +func init() { + allocsMu = new(sync.Mutex) + allocs = make(map[uint64]*Allocator) + + // Set up a unique Ref per process. + rand.Seed(time.Now().UnixNano()) + allocRef = uint64(rand.Int63n(1<<16)) << 48 //nolint:gosec // cryptographic precision not needed + + calculatedLog2 = make([]int, 1025) + for i := 1; i <= 1024; i++ { + calculatedLog2[i] = int(math.Log2(float64(i))) + } +} + +// NewAllocator creates an allocator starting with the given size. +func NewAllocator(sz int, tag string) *Allocator { + ref := atomic.AddUint64(&allocRef, 1) + // We should not allow a zero sized page because addBufferWithMinSize + // will run into an infinite loop trying to double the pagesize. + if sz < 512 { + sz = 512 + } + a := &Allocator{ + Ref: ref, + buffers: make([][]byte, 64), + Tag: tag, + } + l2 := uint64(log2(sz)) + if bits.OnesCount64(uint64(sz)) > 1 { + l2 += 1 + } + a.buffers[0] = Calloc(1<> 32), int(pos & 0xFFFFFFFF) +} + +// Size returns the size of the allocations so far. +func (a *Allocator) Size() int { + pos := atomic.LoadUint64(&a.compIdx) + bi, pi := parse(pos) + var sz int + for i, b := range a.buffers { + if i < bi { + sz += len(b) + continue + } + sz += pi + return sz + } + panic("Size should not reach here") +} + +func log2(sz int) int { + if sz < len(calculatedLog2) { + return calculatedLog2[sz] + } + pow := 10 + sz >>= 10 + for sz > 1 { + sz >>= 1 + pow++ + } + return pow +} + +func (a *Allocator) Allocated() uint64 { + var alloc int + for _, b := range a.buffers { + alloc += cap(b) + } + return uint64(alloc) +} + +func (a *Allocator) TrimTo(max int) { + var alloc int + for i, b := range a.buffers { + if len(b) == 0 { + break + } + alloc += len(b) + if alloc < max { + continue + } + Free(b) + a.buffers[i] = nil + } +} + +// Release would release the memory back. Remember to make this call to avoid memory leaks. +func (a *Allocator) Release() { + if a == nil { + return + } + + var alloc int + for _, b := range a.buffers { + if len(b) == 0 { + break + } + alloc += len(b) + Free(b) + } + + allocsMu.Lock() + delete(allocs, a.Ref) + allocsMu.Unlock() +} + +const maxAlloc = 1 << 30 + +func (a *Allocator) MaxAlloc() int { + return maxAlloc +} + +const nodeAlign = unsafe.Sizeof(uint64(0)) - 1 + +func (a *Allocator) AllocateAligned(sz int) []byte { + tsz := sz + int(nodeAlign) + out := a.Allocate(tsz) + // We are reusing allocators. In that case, it's important to zero out the memory allocated + // here. We don't always zero it out (in Allocate), because other functions would be immediately + // overwriting the allocated slices anyway (see Copy). + ZeroOut(out, 0, len(out)) + + addr := uintptr(unsafe.Pointer(&out[0])) + aligned := (addr + nodeAlign) & ^nodeAlign + start := int(aligned - addr) + + return out[start : start+sz] +} + +func (a *Allocator) Copy(buf []byte) []byte { + if a == nil { + return append([]byte{}, buf...) + } + out := a.Allocate(len(buf)) + copy(out, buf) + return out +} + +func (a *Allocator) addBufferAt(bufIdx, minSz int) { + for { + if bufIdx >= len(a.buffers) { + panic(fmt.Sprintf("Allocator can not allocate more than %d buffers", len(a.buffers))) + } + if len(a.buffers[bufIdx]) == 0 { + break + } + if minSz <= len(a.buffers[bufIdx]) { + // No need to do anything. We already have a buffer which can satisfy minSz. + return + } + bufIdx++ + } + assert(bufIdx > 0) + // We need to allocate a new buffer. + // Make pageSize double of the last allocation. + pageSize := 2 * len(a.buffers[bufIdx-1]) + // Ensure pageSize is bigger than sz. + for pageSize < minSz { + pageSize *= 2 + } + // If bigger than maxAlloc, trim to maxAlloc. + if pageSize > maxAlloc { + pageSize = maxAlloc + } + + buf := Calloc(pageSize, a.Tag) + assert(len(a.buffers[bufIdx]) == 0) + a.buffers[bufIdx] = buf +} + +func (a *Allocator) Allocate(sz int) []byte { + if a == nil { + return make([]byte, sz) + } + if sz > maxAlloc { + panic(fmt.Sprintf("Unable to allocate more than %d\n", maxAlloc)) + } + if sz == 0 { + return nil + } + for { + pos := atomic.AddUint64(&a.compIdx, uint64(sz)) + bufIdx, posIdx := parse(pos) + buf := a.buffers[bufIdx] + if posIdx > len(buf) { + a.Lock() + newPos := atomic.LoadUint64(&a.compIdx) + newBufIdx, _ := parse(newPos) + if newBufIdx != bufIdx { + a.Unlock() + continue + } + a.addBufferAt(bufIdx+1, sz) + atomic.StoreUint64(&a.compIdx, uint64((bufIdx+1)<<32)) + a.Unlock() + // We added a new buffer. Let's acquire slice the right way by going back to the top. + continue + } + data := buf[posIdx-sz : posIdx] + return data + } +} + +type AllocatorPool struct { + numGets int64 + allocCh chan *Allocator + closer *Closer +} + +func NewAllocatorPool(sz int) *AllocatorPool { + a := &AllocatorPool{ + allocCh: make(chan *Allocator, sz), + closer: NewCloser(1), + } + go a.freeupAllocators() + return a +} + +func (p *AllocatorPool) Get(sz int, tag string) *Allocator { + if p == nil { + return NewAllocator(sz, tag) + } + atomic.AddInt64(&p.numGets, 1) + select { + case alloc := <-p.allocCh: + alloc.Reset() + alloc.Tag = tag + return alloc + default: + return NewAllocator(sz, tag) + } +} +func (p *AllocatorPool) Return(a *Allocator) { + if a == nil { + return + } + if p == nil { + a.Release() + return + } + a.TrimTo(400 << 20) + + select { + case p.allocCh <- a: + return + default: + a.Release() + } +} + +func (p *AllocatorPool) Release() { + if p == nil { + return + } + p.closer.SignalAndWait() +} + +func (p *AllocatorPool) freeupAllocators() { + defer p.closer.Done() + + ticker := time.NewTicker(2 * time.Second) + defer ticker.Stop() + + releaseOne := func() bool { + select { + case alloc := <-p.allocCh: + alloc.Release() + return true + default: + return false + } + } + + var last int64 + for { + select { + case <-p.closer.HasBeenClosed(): + close(p.allocCh) + for alloc := range p.allocCh { + alloc.Release() + } + return + + case <-ticker.C: + gets := atomic.LoadInt64(&p.numGets) + if gets != last { + // Some retrievals were made since the last time. So, let's avoid doing a release. + last = gets + continue + } + releaseOne() + } + } +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/bbloom.go b/vendor/github.com/dgraph-io/ristretto/z/bbloom.go new file mode 100644 index 00000000000..37135b012fa --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/bbloom.go @@ -0,0 +1,211 @@ +// The MIT License (MIT) +// Copyright (c) 2014 Andreas Briese, eduToolbox@Bri-C GmbH, Sarstedt + +// Permission is hereby granted, free of charge, to any person obtaining a copy of +// this software and associated documentation files (the "Software"), to deal in +// the Software without restriction, including without limitation the rights to +// use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of +// the Software, and to permit persons to whom the Software is furnished to do so, +// subject to the following conditions: + +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. + +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS +// FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR +// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER +// IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN +// CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +package z + +import ( + "bytes" + "encoding/json" + "math" + "unsafe" + + "github.com/golang/glog" +) + +// helper +var mask = []uint8{1, 2, 4, 8, 16, 32, 64, 128} + +func getSize(ui64 uint64) (size uint64, exponent uint64) { + if ui64 < uint64(512) { + ui64 = uint64(512) + } + size = uint64(1) + for size < ui64 { + size <<= 1 + exponent++ + } + return size, exponent +} + +func calcSizeByWrongPositives(numEntries, wrongs float64) (uint64, uint64) { + size := -1 * numEntries * math.Log(wrongs) / math.Pow(float64(0.69314718056), 2) + locs := math.Ceil(float64(0.69314718056) * size / numEntries) + return uint64(size), uint64(locs) +} + +// NewBloomFilter returns a new bloomfilter. +func NewBloomFilter(params ...float64) (bloomfilter *Bloom) { + var entries, locs uint64 + if len(params) == 2 { + if params[1] < 1 { + entries, locs = calcSizeByWrongPositives(params[0], params[1]) + } else { + entries, locs = uint64(params[0]), uint64(params[1]) + } + } else { + glog.Fatal("usage: New(float64(number_of_entries), float64(number_of_hashlocations))" + + " i.e. New(float64(1000), float64(3)) or New(float64(number_of_entries)," + + " float64(number_of_hashlocations)) i.e. New(float64(1000), float64(0.03))") + } + size, exponent := getSize(entries) + bloomfilter = &Bloom{ + sizeExp: exponent, + size: size - 1, + setLocs: locs, + shift: 64 - exponent, + } + bloomfilter.Size(size) + return bloomfilter +} + +// Bloom filter +type Bloom struct { + bitset []uint64 + ElemNum uint64 + sizeExp uint64 + size uint64 + setLocs uint64 + shift uint64 +} + +// <--- http://www.cse.yorku.ca/~oz/hash.html +// modified Berkeley DB Hash (32bit) +// hash is casted to l, h = 16bit fragments +// func (bl Bloom) absdbm(b *[]byte) (l, h uint64) { +// hash := uint64(len(*b)) +// for _, c := range *b { +// hash = uint64(c) + (hash << 6) + (hash << bl.sizeExp) - hash +// } +// h = hash >> bl.shift +// l = hash << bl.shift >> bl.shift +// return l, h +// } + +// Add adds hash of a key to the bloomfilter. +func (bl *Bloom) Add(hash uint64) { + h := hash >> bl.shift + l := hash << bl.shift >> bl.shift + for i := uint64(0); i < bl.setLocs; i++ { + bl.Set((h + i*l) & bl.size) + bl.ElemNum++ + } +} + +// Has checks if bit(s) for entry hash is/are set, +// returns true if the hash was added to the Bloom Filter. +func (bl Bloom) Has(hash uint64) bool { + h := hash >> bl.shift + l := hash << bl.shift >> bl.shift + for i := uint64(0); i < bl.setLocs; i++ { + if !bl.IsSet((h + i*l) & bl.size) { + return false + } + } + return true +} + +// AddIfNotHas only Adds hash, if it's not present in the bloomfilter. +// Returns true if hash was added. +// Returns false if hash was already registered in the bloomfilter. +func (bl *Bloom) AddIfNotHas(hash uint64) bool { + if bl.Has(hash) { + return false + } + bl.Add(hash) + return true +} + +// TotalSize returns the total size of the bloom filter. +func (bl *Bloom) TotalSize() int { + // The bl struct has 5 members and each one is 8 byte. The bitset is a + // uint64 byte slice. + return len(bl.bitset)*8 + 5*8 +} + +// Size makes Bloom filter with as bitset of size sz. +func (bl *Bloom) Size(sz uint64) { + bl.bitset = make([]uint64, sz>>6) +} + +// Clear resets the Bloom filter. +func (bl *Bloom) Clear() { + for i := range bl.bitset { + bl.bitset[i] = 0 + } +} + +// Set sets the bit[idx] of bitset. +func (bl *Bloom) Set(idx uint64) { + ptr := unsafe.Pointer(uintptr(unsafe.Pointer(&bl.bitset[idx>>6])) + uintptr((idx%64)>>3)) + *(*uint8)(ptr) |= mask[idx%8] +} + +// IsSet checks if bit[idx] of bitset is set, returns true/false. +func (bl *Bloom) IsSet(idx uint64) bool { + ptr := unsafe.Pointer(uintptr(unsafe.Pointer(&bl.bitset[idx>>6])) + uintptr((idx%64)>>3)) + r := ((*(*uint8)(ptr)) >> (idx % 8)) & 1 + return r == 1 +} + +// bloomJSONImExport +// Im/Export structure used by JSONMarshal / JSONUnmarshal +type bloomJSONImExport struct { + FilterSet []byte + SetLocs uint64 +} + +// NewWithBoolset takes a []byte slice and number of locs per entry, +// returns the bloomfilter with a bitset populated according to the input []byte. +func newWithBoolset(bs *[]byte, locs uint64) *Bloom { + bloomfilter := NewBloomFilter(float64(len(*bs)<<3), float64(locs)) + for i, b := range *bs { + *(*uint8)(unsafe.Pointer(uintptr(unsafe.Pointer(&bloomfilter.bitset[0])) + uintptr(i))) = b + } + return bloomfilter +} + +// JSONUnmarshal takes JSON-Object (type bloomJSONImExport) as []bytes +// returns bloom32 / bloom64 object. +func JSONUnmarshal(dbData []byte) (*Bloom, error) { + bloomImEx := bloomJSONImExport{} + if err := json.Unmarshal(dbData, &bloomImEx); err != nil { + return nil, err + } + buf := bytes.NewBuffer(bloomImEx.FilterSet) + bs := buf.Bytes() + bf := newWithBoolset(&bs, bloomImEx.SetLocs) + return bf, nil +} + +// JSONMarshal returns JSON-object (type bloomJSONImExport) as []byte. +func (bl Bloom) JSONMarshal() []byte { + bloomImEx := bloomJSONImExport{} + bloomImEx.SetLocs = bl.setLocs + bloomImEx.FilterSet = make([]byte, len(bl.bitset)<<3) + for i := range bloomImEx.FilterSet { + bloomImEx.FilterSet[i] = *(*byte)(unsafe.Pointer(uintptr(unsafe.Pointer(&bl.bitset[0])) + + uintptr(i))) + } + data, err := json.Marshal(bloomImEx) + if err != nil { + glog.Fatal("json.Marshal failed: ", err) + } + return data +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/btree.go b/vendor/github.com/dgraph-io/ristretto/z/btree.go new file mode 100644 index 00000000000..12b735bb032 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/btree.go @@ -0,0 +1,710 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "fmt" + "math" + "os" + "reflect" + "strings" + "unsafe" + + "github.com/dgraph-io/ristretto/z/simd" +) + +var ( + pageSize = os.Getpagesize() + maxKeys = (pageSize / 16) - 1 + oneThird = int(float64(maxKeys) / 3) +) + +const ( + absoluteMax = uint64(math.MaxUint64 - 1) + minSize = 1 << 20 +) + +// Tree represents the structure for custom mmaped B+ tree. +// It supports keys in range [1, math.MaxUint64-1] and values [1, math.Uint64]. +type Tree struct { + buffer *Buffer + data []byte + nextPage uint64 + freePage uint64 + stats TreeStats +} + +func (t *Tree) initRootNode() { + // This is the root node. + t.newNode(0) + // This acts as the rightmost pointer (all the keys are <= this key). + t.Set(absoluteMax, 0) +} + +// NewTree returns an in-memory B+ tree. +func NewTree(tag string) *Tree { + const defaultTag = "tree" + if tag == "" { + tag = defaultTag + } + t := &Tree{buffer: NewBuffer(minSize, tag)} + t.Reset() + return t +} + +// NewTree returns a persistent on-disk B+ tree. +func NewTreePersistent(path string) (*Tree, error) { + t := &Tree{} + var err error + + // Open the buffer from disk and set it to the maximum allocated size. + t.buffer, err = NewBufferPersistent(path, minSize) + if err != nil { + return nil, err + } + t.buffer.offset = uint64(len(t.buffer.buf)) + t.data = t.buffer.Bytes() + + // pageID can never be 0 if the tree has been initialized. + root := t.node(1) + isInitialized := root.pageID() != 0 + + if !isInitialized { + t.nextPage = 1 + t.freePage = 0 + t.initRootNode() + } else { + t.reinit() + } + + return t, nil +} + +// reinit sets the internal variables of a Tree, which are normally stored +// in-memory, but are lost when loading from disk. +func (t *Tree) reinit() { + // Calculate t.nextPage by finding the first node whose pageID is not set. + t.nextPage = 1 + for int(t.nextPage)*pageSize < len(t.data) { + n := t.node(t.nextPage) + if n.pageID() == 0 { + break + } + t.nextPage++ + } + maxPageId := t.nextPage - 1 + + // Calculate t.freePage by finding the page to which no other page points. + // This would be the head of the page linked list. + // tailPages[i] is true if pageId i+1 is not the head of the list. + tailPages := make([]bool, maxPageId) + // Mark all pages containing nodes as tail pages. + t.Iterate(func(n node) { + i := n.pageID() - 1 + tailPages[i] = true + // If this is a leaf node, increment the stats. + if n.isLeaf() { + t.stats.NumLeafKeys += n.numKeys() + } + }) + // pointedPages is a list of page IDs that the tail pages point to. + pointedPages := make([]uint64, 0) + for i, isTail := range tailPages { + if !isTail { + pageId := uint64(i) + 1 + // Skip if nextPageId = 0, as that is equivalent to null page. + if nextPageId := t.node(pageId).uint64(0); nextPageId != 0 { + pointedPages = append(pointedPages, nextPageId) + } + t.stats.NumPagesFree++ + } + } + + // Mark all pages being pointed to as tail pages. + for _, pageId := range pointedPages { + i := pageId - 1 + tailPages[i] = true + } + // There should only be one head page left. + for i, isTail := range tailPages { + if !isTail { + pageId := uint64(i) + 1 + t.freePage = pageId + break + } + } +} + +// Reset resets the tree and truncates it to maxSz. +func (t *Tree) Reset() { + // Tree relies on uninitialized data being zeroed out, so we need to Memclr + // the data before using it again. + Memclr(t.buffer.buf) + t.buffer.Reset() + t.buffer.AllocateOffset(minSize) + t.data = t.buffer.Bytes() + t.stats = TreeStats{} + t.nextPage = 1 + t.freePage = 0 + t.initRootNode() +} + +// Close releases the memory used by the tree. +func (t *Tree) Close() error { + if t == nil { + return nil + } + return t.buffer.Release() +} + +type TreeStats struct { + Allocated int // Derived. + Bytes int // Derived. + NumLeafKeys int // Calculated. + NumPages int // Derived. + NumPagesFree int // Calculated. + Occupancy float64 // Derived. + PageSize int // Derived. +} + +// Stats returns stats about the tree. +func (t *Tree) Stats() TreeStats { + numPages := int(t.nextPage - 1) + out := TreeStats{ + Bytes: numPages * pageSize, + Allocated: len(t.data), + NumLeafKeys: t.stats.NumLeafKeys, + NumPages: numPages, + NumPagesFree: t.stats.NumPagesFree, + PageSize: pageSize, + } + out.Occupancy = 100.0 * float64(out.NumLeafKeys) / float64(maxKeys*numPages) + return out +} + +// BytesToUint64Slice converts a byte slice to a uint64 slice. +func BytesToUint64Slice(b []byte) []uint64 { + if len(b) == 0 { + return nil + } + var u64s []uint64 + hdr := (*reflect.SliceHeader)(unsafe.Pointer(&u64s)) + hdr.Len = len(b) / 8 + hdr.Cap = hdr.Len + hdr.Data = uintptr(unsafe.Pointer(&b[0])) + return u64s +} + +func (t *Tree) newNode(bit uint64) node { + var pageId uint64 + if t.freePage > 0 { + pageId = t.freePage + t.stats.NumPagesFree-- + } else { + pageId = t.nextPage + t.nextPage++ + offset := int(pageId) * pageSize + reqSize := offset + pageSize + if reqSize > len(t.data) { + t.buffer.AllocateOffset(reqSize - len(t.data)) + t.data = t.buffer.Bytes() + } + } + n := t.node(pageId) + if t.freePage > 0 { + t.freePage = n.uint64(0) + } + zeroOut(n) + n.setBit(bit) + n.setAt(keyOffset(maxKeys), pageId) + return n +} + +func getNode(data []byte) node { + return node(BytesToUint64Slice(data)) +} + +func zeroOut(data []uint64) { + for i := 0; i < len(data); i++ { + data[i] = 0 + } +} + +func (t *Tree) node(pid uint64) node { + // page does not exist + if pid == 0 { + return nil + } + start := pageSize * int(pid) + return getNode(t.data[start : start+pageSize]) +} + +// Set sets the key-value pair in the tree. +func (t *Tree) Set(k, v uint64) { + if k == math.MaxUint64 || k == 0 { + panic("Error setting zero or MaxUint64") + } + root := t.set(1, k, v) + if root.isFull() { + right := t.split(1) + left := t.newNode(root.bits()) + // Re-read the root as the underlying buffer for tree might have changed during split. + root = t.node(1) + copy(left[:keyOffset(maxKeys)], root) + left.setNumKeys(root.numKeys()) + + // reset the root node. + zeroOut(root[:keyOffset(maxKeys)]) + root.setNumKeys(0) + + // set the pointers for left and right child in the root node. + root.set(left.maxKey(), left.pageID()) + root.set(right.maxKey(), right.pageID()) + } +} + +// For internal nodes, they contain . +// where all entries <= key are stored in the corresponding ptr. +func (t *Tree) set(pid, k, v uint64) node { + n := t.node(pid) + if n.isLeaf() { + t.stats.NumLeafKeys += n.set(k, v) + return n + } + + // This is an internal node. + idx := n.search(k) + if idx >= maxKeys { + panic("search returned index >= maxKeys") + } + // If no key at idx. + if n.key(idx) == 0 { + n.setAt(keyOffset(idx), k) + n.setNumKeys(n.numKeys() + 1) + } + child := t.node(n.val(idx)) + if child == nil { + child = t.newNode(bitLeaf) + n = t.node(pid) + n.setAt(valOffset(idx), child.pageID()) + } + child = t.set(child.pageID(), k, v) + // Re-read n as the underlying buffer for tree might have changed during set. + n = t.node(pid) + if child.isFull() { + // Just consider the left sibling for simplicity. + // if t.shareWithSibling(n, idx) { + // return n + // } + + nn := t.split(child.pageID()) + // Re-read n and child as the underlying buffer for tree might have changed during split. + n = t.node(pid) + child = t.node(n.uint64(valOffset(idx))) + // Set child pointers in the node n. + // Note that key for right node (nn) already exist in node n, but the + // pointer is updated. + n.set(child.maxKey(), child.pageID()) + n.set(nn.maxKey(), nn.pageID()) + } + return n +} + +// Get looks for key and returns the corresponding value. +// If key is not found, 0 is returned. +func (t *Tree) Get(k uint64) uint64 { + if k == math.MaxUint64 || k == 0 { + panic("Does not support getting MaxUint64/Zero") + } + root := t.node(1) + return t.get(root, k) +} + +func (t *Tree) get(n node, k uint64) uint64 { + if n.isLeaf() { + return n.get(k) + } + // This is internal node + idx := n.search(k) + if idx == n.numKeys() || n.key(idx) == 0 { + return 0 + } + child := t.node(n.uint64(valOffset(idx))) + assert(child != nil) + return t.get(child, k) +} + +// DeleteBelow deletes all keys with value under ts. +func (t *Tree) DeleteBelow(ts uint64) { + root := t.node(1) + t.stats.NumLeafKeys = 0 + t.compact(root, ts) + assert(root.numKeys() >= 1) +} + +func (t *Tree) compact(n node, ts uint64) int { + if n.isLeaf() { + numKeys := n.compact(ts) + t.stats.NumLeafKeys += n.numKeys() + return numKeys + } + // Not leaf. + N := n.numKeys() + for i := 0; i < N; i++ { + assert(n.key(i) > 0) + childID := n.uint64(valOffset(i)) + child := t.node(childID) + if rem := t.compact(child, ts); rem == 0 && i < N-1 { + // If no valid key is remaining we can drop this child. However, don't do that if this + // is the max key. + t.stats.NumLeafKeys -= child.numKeys() + child.setAt(0, t.freePage) + t.freePage = childID + n.setAt(valOffset(i), 0) + t.stats.NumPagesFree++ + } + } + // We use ts=1 here because we want to delete all the keys whose value is 0, which means they no + // longer have a valid page for that key. + return n.compact(1) +} + +func (t *Tree) iterate(n node, fn func(node)) { + fn(n) + if n.isLeaf() { + return + } + // Explore children. + for i := 0; i < maxKeys; i++ { + if n.key(i) == 0 { + return + } + childID := n.uint64(valOffset(i)) + assert(childID > 0) + + child := t.node(childID) + t.iterate(child, fn) + } +} + +// Iterate iterates over the tree and executes the fn on each node. +func (t *Tree) Iterate(fn func(node)) { + root := t.node(1) + t.iterate(root, fn) +} + +// IterateKV iterates through all keys and values in the tree. +// If newVal is non-zero, it will be set in the tree. +func (t *Tree) IterateKV(f func(key, val uint64) (newVal uint64)) { + t.Iterate(func(n node) { + // Only leaf nodes contain keys. + if !n.isLeaf() { + return + } + + for i := 0; i < n.numKeys(); i++ { + key := n.key(i) + val := n.val(i) + + // A zero value here means that this is a bogus entry. + if val == 0 { + continue + } + + newVal := f(key, val) + if newVal != 0 { + n.setAt(valOffset(i), newVal) + } + } + }) +} + +func (t *Tree) print(n node, parentID uint64) { + n.print(parentID) + if n.isLeaf() { + return + } + pid := n.pageID() + for i := 0; i < maxKeys; i++ { + if n.key(i) == 0 { + return + } + childID := n.uint64(valOffset(i)) + child := t.node(childID) + t.print(child, pid) + } +} + +// Print iterates over the tree and prints all valid KVs. +func (t *Tree) Print() { + root := t.node(1) + t.print(root, 0) +} + +// Splits the node into two. It moves right half of the keys from the original node to a newly +// created right node. It returns the right node. +func (t *Tree) split(pid uint64) node { + n := t.node(pid) + if !n.isFull() { + panic("This should be called only when n is full") + } + + // Create a new node nn, copy over half the keys from n, and set the parent to n's parent. + nn := t.newNode(n.bits()) + // Re-read n as the underlying buffer for tree might have changed during newNode. + n = t.node(pid) + rightHalf := n[keyOffset(maxKeys/2):keyOffset(maxKeys)] + copy(nn, rightHalf) + nn.setNumKeys(maxKeys - maxKeys/2) + + // Remove entries from node n. + zeroOut(rightHalf) + n.setNumKeys(maxKeys / 2) + return nn +} + +// shareWithSiblingXXX is unused for now. The idea is to move some keys to +// sibling when a node is full. But, I don't see any special benefits in our +// access pattern. It doesn't result in better occupancy ratios. +func (t *Tree) shareWithSiblingXXX(n node, idx int) bool { + if idx == 0 { + return false + } + left := t.node(n.val(idx - 1)) + ns := left.numKeys() + if ns >= maxKeys/2 { + // Sibling is already getting full. + return false + } + + right := t.node(n.val(idx)) + // Copy over keys from right child to left child. + copied := copy(left[keyOffset(ns):], right[:keyOffset(oneThird)]) + copied /= 2 // Considering that key-val constitute one key. + left.setNumKeys(ns + copied) + + // Update the max key in parent node n for the left sibling. + n.setAt(keyOffset(idx-1), left.maxKey()) + + // Now move keys to left for the right sibling. + until := copy(right, right[keyOffset(oneThird):keyOffset(maxKeys)]) + right.setNumKeys(until / 2) + zeroOut(right[until:keyOffset(maxKeys)]) + return true +} + +// Each node in the node is of size pageSize. Two kinds of nodes. Leaf nodes and internal nodes. +// Leaf nodes only contain the data. Internal nodes would contain the key and the offset to the +// child node. +// Internal node would have first entry as +// <0 offset to child>, <1000 offset>, <5000 offset>, and so on... +// Leaf nodes would just have: , , and so on... +// Last 16 bytes of the node are off limits. +// | pageID (8 bytes) | metaBits (1 byte) | 3 free bytes | numKeys (4 bytes) | +type node []uint64 + +func (n node) uint64(start int) uint64 { return n[start] } + +// func (n node) uint32(start int) uint32 { return *(*uint32)(unsafe.Pointer(&n[start])) } + +func keyOffset(i int) int { return 2 * i } +func valOffset(i int) int { return 2*i + 1 } +func (n node) numKeys() int { return int(n.uint64(valOffset(maxKeys)) & 0xFFFFFFFF) } +func (n node) pageID() uint64 { return n.uint64(keyOffset(maxKeys)) } +func (n node) key(i int) uint64 { return n.uint64(keyOffset(i)) } +func (n node) val(i int) uint64 { return n.uint64(valOffset(i)) } +func (n node) data(i int) []uint64 { return n[keyOffset(i):keyOffset(i+1)] } + +func (n node) setAt(start int, k uint64) { + n[start] = k +} + +func (n node) setNumKeys(num int) { + idx := valOffset(maxKeys) + val := n[idx] + val &= 0xFFFFFFFF00000000 + val |= uint64(num) + n[idx] = val +} + +func (n node) moveRight(lo int) { + hi := n.numKeys() + assert(hi != maxKeys) + // copy works despite of overlap in src and dst. + // See https://golang.org/pkg/builtin/#copy + copy(n[keyOffset(lo+1):keyOffset(hi+1)], n[keyOffset(lo):keyOffset(hi)]) +} + +const ( + bitLeaf = uint64(1 << 63) +) + +func (n node) setBit(b uint64) { + vo := valOffset(maxKeys) + val := n[vo] + val &= 0xFFFFFFFF + val |= b + n[vo] = val +} +func (n node) bits() uint64 { + return n.val(maxKeys) & 0xFF00000000000000 +} +func (n node) isLeaf() bool { + return n.bits()&bitLeaf > 0 +} + +// isFull checks that the node is already full. +func (n node) isFull() bool { + return n.numKeys() == maxKeys +} + +// Search returns the index of a smallest key >= k in a node. +func (n node) search(k uint64) int { + N := n.numKeys() + if N < 4 { + for i := 0; i < N; i++ { + if ki := n.key(i); ki >= k { + return i + } + } + return N + } + return int(simd.Search(n[:2*N], k)) + // lo, hi := 0, N + // // Reduce the search space using binary seach and then do linear search. + // for hi-lo > 32 { + // mid := (hi + lo) / 2 + // km := n.key(mid) + // if k == km { + // return mid + // } + // if k > km { + // // key is greater than the key at mid, so move right. + // lo = mid + 1 + // } else { + // // else move left. + // hi = mid + // } + // } + // for i := lo; i <= hi; i++ { + // if ki := n.key(i); ki >= k { + // return i + // } + // } + // return N +} +func (n node) maxKey() uint64 { + idx := n.numKeys() + // idx points to the first key which is zero. + if idx > 0 { + idx-- + } + return n.key(idx) +} + +// compacts the node i.e., remove all the kvs with value < lo. It returns the remaining number of +// keys. +func (n node) compact(lo uint64) int { + N := n.numKeys() + mk := n.maxKey() + var left, right int + for right = 0; right < N; right++ { + if n.val(right) < lo && n.key(right) < mk { + // Skip over this key. Don't copy it. + continue + } + // Valid data. Copy it from right to left. Advance left. + if left != right { + copy(n.data(left), n.data(right)) + } + left++ + } + // zero out rest of the kv pairs. + zeroOut(n[keyOffset(left):keyOffset(right)]) + n.setNumKeys(left) + + // If the only key we have is the max key, and its value is less than lo, then we can indicate + // to the caller by returning a zero that it's OK to drop the node. + if left == 1 && n.key(0) == mk && n.val(0) < lo { + return 0 + } + return left +} + +func (n node) get(k uint64) uint64 { + idx := n.search(k) + // key is not found + if idx == n.numKeys() { + return 0 + } + if ki := n.key(idx); ki == k { + return n.val(idx) + } + return 0 +} + +// set returns true if it added a new key. +func (n node) set(k, v uint64) (numAdded int) { + idx := n.search(k) + ki := n.key(idx) + if n.numKeys() == maxKeys { + // This happens during split of non-root node, when we are updating the child pointer of + // right node. Hence, the key should already exist. + assert(ki == k) + } + if ki > k { + // Found the first entry which is greater than k. So, we need to fit k + // just before it. For that, we should move the rest of the data in the + // node to the right to make space for k. + n.moveRight(idx) + } + // If the k does not exist already, increment the number of keys. + if ki != k { + n.setNumKeys(n.numKeys() + 1) + numAdded = 1 + } + if ki == 0 || ki >= k { + n.setAt(keyOffset(idx), k) + n.setAt(valOffset(idx), v) + return + } + panic("shouldn't reach here") +} + +func (n node) iterate(fn func(node, int)) { + for i := 0; i < maxKeys; i++ { + if k := n.key(i); k > 0 { + fn(n, i) + } else { + break + } + } +} + +func (n node) print(parentID uint64) { + var keys []string + n.iterate(func(n node, i int) { + keys = append(keys, fmt.Sprintf("%d", n.key(i))) + }) + if len(keys) > 8 { + copy(keys[4:], keys[len(keys)-4:]) + keys[3] = "..." + keys = keys[:8] + } + fmt.Printf("%d Child of: %d num keys: %d keys: %s\n", + n.pageID(), parentID, n.numKeys(), strings.Join(keys, " ")) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/buffer.go b/vendor/github.com/dgraph-io/ristretto/z/buffer.go new file mode 100644 index 00000000000..5a22de8c7f4 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/buffer.go @@ -0,0 +1,544 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "encoding/binary" + "fmt" + "io/ioutil" + "os" + "sort" + "sync/atomic" + + "github.com/golang/glog" + "github.com/pkg/errors" +) + +const ( + defaultCapacity = 64 + defaultTag = "buffer" +) + +// Buffer is equivalent of bytes.Buffer without the ability to read. It is NOT thread-safe. +// +// In UseCalloc mode, z.Calloc is used to allocate memory, which depending upon how the code is +// compiled could use jemalloc for allocations. +// +// In UseMmap mode, Buffer uses file mmap to allocate memory. This allows us to store big data +// structures without using physical memory. +// +// MaxSize can be set to limit the memory usage. +type Buffer struct { + padding uint64 // number of starting bytes used for padding + offset uint64 // used length of the buffer + buf []byte // backing slice for the buffer + bufType BufferType // type of the underlying buffer + curSz int // capacity of the buffer + maxSz int // causes a panic if the buffer grows beyond this size + mmapFile *MmapFile // optional mmap backing for the buffer + autoMmapAfter int // Calloc falls back to an mmaped tmpfile after crossing this size + autoMmapDir string // directory for autoMmap to create a tempfile in + persistent bool // when enabled, Release will not delete the underlying mmap file + tag string // used for jemalloc stats +} + +func NewBuffer(capacity int, tag string) *Buffer { + if capacity < defaultCapacity { + capacity = defaultCapacity + } + if tag == "" { + tag = defaultTag + } + return &Buffer{ + buf: Calloc(capacity, tag), + bufType: UseCalloc, + curSz: capacity, + offset: 8, + padding: 8, + tag: tag, + } +} + +// It is the caller's responsibility to set offset after this, because Buffer +// doesn't remember what it was. +func NewBufferPersistent(path string, capacity int) (*Buffer, error) { + file, err := os.OpenFile(path, os.O_RDWR|os.O_CREATE, 0666) + if err != nil { + return nil, err + } + buffer, err := newBufferFile(file, capacity) + if err != nil { + return nil, err + } + buffer.persistent = true + return buffer, nil +} + +func NewBufferTmp(dir string, capacity int) (*Buffer, error) { + if dir == "" { + dir = tmpDir + } + file, err := ioutil.TempFile(dir, "buffer") + if err != nil { + return nil, err + } + return newBufferFile(file, capacity) +} + +func newBufferFile(file *os.File, capacity int) (*Buffer, error) { + if capacity < defaultCapacity { + capacity = defaultCapacity + } + mmapFile, err := OpenMmapFileUsing(file, capacity, true) + if err != nil && err != NewFile { + return nil, err + } + buf := &Buffer{ + buf: mmapFile.Data, + bufType: UseMmap, + curSz: len(mmapFile.Data), + mmapFile: mmapFile, + offset: 8, + padding: 8, + } + return buf, nil +} + +func NewBufferSlice(slice []byte) *Buffer { + return &Buffer{ + offset: uint64(len(slice)), + buf: slice, + bufType: UseInvalid, + } +} + +func (b *Buffer) WithAutoMmap(threshold int, path string) *Buffer { + if b.bufType != UseCalloc { + panic("can only autoMmap with UseCalloc") + } + b.autoMmapAfter = threshold + if path == "" { + b.autoMmapDir = tmpDir + } else { + b.autoMmapDir = path + } + return b +} + +func (b *Buffer) WithMaxSize(size int) *Buffer { + b.maxSz = size + return b +} + +func (b *Buffer) IsEmpty() bool { + return int(b.offset) == b.StartOffset() +} + +// LenWithPadding would return the number of bytes written to the buffer so far +// plus the padding at the start of the buffer. +func (b *Buffer) LenWithPadding() int { + return int(atomic.LoadUint64(&b.offset)) +} + +// LenNoPadding would return the number of bytes written to the buffer so far +// (without the padding). +func (b *Buffer) LenNoPadding() int { + return int(atomic.LoadUint64(&b.offset) - b.padding) +} + +// Bytes would return all the written bytes as a slice. +func (b *Buffer) Bytes() []byte { + off := atomic.LoadUint64(&b.offset) + return b.buf[b.padding:off] +} + +// Grow would grow the buffer to have at least n more bytes. In case the buffer is at capacity, it +// would reallocate twice the size of current capacity + n, to ensure n bytes can be written to the +// buffer without further allocation. In UseMmap mode, this might result in underlying file +// expansion. +func (b *Buffer) Grow(n int) { + if b.buf == nil { + panic("z.Buffer needs to be initialized before using") + } + if b.maxSz > 0 && int(b.offset)+n > b.maxSz { + err := fmt.Errorf( + "z.Buffer max size exceeded: %d offset: %d grow: %d", b.maxSz, b.offset, n) + panic(err) + } + if int(b.offset)+n < b.curSz { + return + } + + // Calculate new capacity. + growBy := b.curSz + n + // Don't allocate more than 1GB at a time. + if growBy > 1<<30 { + growBy = 1 << 30 + } + // Allocate at least n, even if it exceeds the 1GB limit above. + if n > growBy { + growBy = n + } + b.curSz += growBy + + switch b.bufType { + case UseCalloc: + // If autoMmap gets triggered, copy the slice over to an mmaped file. + if b.autoMmapAfter > 0 && b.curSz > b.autoMmapAfter { + b.bufType = UseMmap + file, err := ioutil.TempFile(b.autoMmapDir, "") + if err != nil { + panic(err) + } + mmapFile, err := OpenMmapFileUsing(file, b.curSz, true) + if err != nil && err != NewFile { + panic(err) + } + assert(int(b.offset) == copy(mmapFile.Data, b.buf[:b.offset])) + Free(b.buf) + b.mmapFile = mmapFile + b.buf = mmapFile.Data + break + } + + // Else, reallocate the slice. + newBuf := Calloc(b.curSz, b.tag) + assert(int(b.offset) == copy(newBuf, b.buf[:b.offset])) + Free(b.buf) + b.buf = newBuf + + case UseMmap: + // Truncate and remap the underlying file. + if err := b.mmapFile.Truncate(int64(b.curSz)); err != nil { + err = errors.Wrapf(err, + "while trying to truncate file: %s to size: %d", b.mmapFile.Fd.Name(), b.curSz) + panic(err) + } + b.buf = b.mmapFile.Data + + default: + panic("can only use Grow on UseCalloc and UseMmap buffers") + } +} + +// Allocate is a way to get a slice of size n back from the buffer. This slice can be directly +// written to. Warning: Allocate is not thread-safe. The byte slice returned MUST be used before +// further calls to Buffer. +func (b *Buffer) Allocate(n int) []byte { + b.Grow(n) + off := b.offset + b.offset += uint64(n) + return b.buf[off:int(b.offset)] +} + +// AllocateOffset works the same way as allocate, but instead of returning a byte slice, it returns +// the offset of the allocation. +func (b *Buffer) AllocateOffset(n int) int { + b.Grow(n) + b.offset += uint64(n) + return int(b.offset) - n +} + +func (b *Buffer) writeLen(sz int) { + buf := b.Allocate(4) + binary.BigEndian.PutUint32(buf, uint32(sz)) +} + +// SliceAllocate would encode the size provided into the buffer, followed by a call to Allocate, +// hence returning the slice of size sz. This can be used to allocate a lot of small buffers into +// this big buffer. +// Note that SliceAllocate should NOT be mixed with normal calls to Write. +func (b *Buffer) SliceAllocate(sz int) []byte { + b.Grow(4 + sz) + b.writeLen(sz) + return b.Allocate(sz) +} + +func (b *Buffer) StartOffset() int { + return int(b.padding) +} + +func (b *Buffer) WriteSlice(slice []byte) { + dst := b.SliceAllocate(len(slice)) + assert(len(slice) == copy(dst, slice)) +} + +func (b *Buffer) SliceIterate(f func(slice []byte) error) error { + if b.IsEmpty() { + return nil + } + slice, next := []byte{}, b.StartOffset() + for next >= 0 { + slice, next = b.Slice(next) + if len(slice) == 0 { + continue + } + if err := f(slice); err != nil { + return err + } + } + return nil +} + +const ( + UseCalloc BufferType = iota + UseMmap + UseInvalid +) + +type BufferType int + +func (t BufferType) String() string { + switch t { + case UseCalloc: + return "UseCalloc" + case UseMmap: + return "UseMmap" + default: + return "UseInvalid" + } +} + +type LessFunc func(a, b []byte) bool +type sortHelper struct { + offsets []int + b *Buffer + tmp *Buffer + less LessFunc + small []int +} + +func (s *sortHelper) sortSmall(start, end int) { + s.tmp.Reset() + s.small = s.small[:0] + next := start + for next >= 0 && next < end { + s.small = append(s.small, next) + _, next = s.b.Slice(next) + } + + // We are sorting the slices pointed to by s.small offsets, but only moving the offsets around. + sort.Slice(s.small, func(i, j int) bool { + left, _ := s.b.Slice(s.small[i]) + right, _ := s.b.Slice(s.small[j]) + return s.less(left, right) + }) + // Now we iterate over the s.small offsets and copy over the slices. The result is now in order. + for _, off := range s.small { + s.tmp.Write(rawSlice(s.b.buf[off:])) + } + assert(end-start == copy(s.b.buf[start:end], s.tmp.Bytes())) +} + +func assert(b bool) { + if !b { + glog.Fatalf("%+v", errors.Errorf("Assertion failure")) + } +} +func check(err error) { + if err != nil { + glog.Fatalf("%+v", err) + } +} +func check2(_ interface{}, err error) { + check(err) +} + +func (s *sortHelper) merge(left, right []byte, start, end int) { + if len(left) == 0 || len(right) == 0 { + return + } + s.tmp.Reset() + check2(s.tmp.Write(left)) + left = s.tmp.Bytes() + + var ls, rs []byte + + copyLeft := func() { + assert(len(ls) == copy(s.b.buf[start:], ls)) + left = left[len(ls):] + start += len(ls) + } + copyRight := func() { + assert(len(rs) == copy(s.b.buf[start:], rs)) + right = right[len(rs):] + start += len(rs) + } + + for start < end { + if len(left) == 0 { + assert(len(right) == copy(s.b.buf[start:end], right)) + return + } + if len(right) == 0 { + assert(len(left) == copy(s.b.buf[start:end], left)) + return + } + ls = rawSlice(left) + rs = rawSlice(right) + + // We skip the first 4 bytes in the rawSlice, because that stores the length. + if s.less(ls[4:], rs[4:]) { + copyLeft() + } else { + copyRight() + } + } +} + +func (s *sortHelper) sort(lo, hi int) []byte { + assert(lo <= hi) + + mid := lo + (hi-lo)/2 + loff, hoff := s.offsets[lo], s.offsets[hi] + if lo == mid { + // No need to sort, just return the buffer. + return s.b.buf[loff:hoff] + } + + // lo, mid would sort from [offset[lo], offset[mid]) . + left := s.sort(lo, mid) + // Typically we'd use mid+1, but here mid represents an offset in the buffer. Each offset + // contains a thousand entries. So, if we do mid+1, we'd skip over those entries. + right := s.sort(mid, hi) + + s.merge(left, right, loff, hoff) + return s.b.buf[loff:hoff] +} + +// SortSlice is like SortSliceBetween but sorting over the entire buffer. +func (b *Buffer) SortSlice(less func(left, right []byte) bool) { + b.SortSliceBetween(b.StartOffset(), int(b.offset), less) +} +func (b *Buffer) SortSliceBetween(start, end int, less LessFunc) { + if start >= end { + return + } + if start == 0 { + panic("start can never be zero") + } + + var offsets []int + next, count := start, 0 + for next >= 0 && next < end { + if count%1024 == 0 { + offsets = append(offsets, next) + } + _, next = b.Slice(next) + count++ + } + assert(len(offsets) > 0) + if offsets[len(offsets)-1] != end { + offsets = append(offsets, end) + } + + szTmp := int(float64((end-start)/2) * 1.1) + s := &sortHelper{ + offsets: offsets, + b: b, + less: less, + small: make([]int, 0, 1024), + tmp: NewBuffer(szTmp, b.tag), + } + defer s.tmp.Release() + + left := offsets[0] + for _, off := range offsets[1:] { + s.sortSmall(left, off) + left = off + } + s.sort(0, len(offsets)-1) +} + +func rawSlice(buf []byte) []byte { + sz := binary.BigEndian.Uint32(buf) + return buf[:4+int(sz)] +} + +// Slice would return the slice written at offset. +func (b *Buffer) Slice(offset int) ([]byte, int) { + if offset >= int(b.offset) { + return nil, -1 + } + + sz := binary.BigEndian.Uint32(b.buf[offset:]) + start := offset + 4 + next := start + int(sz) + res := b.buf[start:next] + if next >= int(b.offset) { + next = -1 + } + return res, next +} + +// SliceOffsets is an expensive function. Use sparingly. +func (b *Buffer) SliceOffsets() []int { + next := b.StartOffset() + var offsets []int + for next >= 0 { + offsets = append(offsets, next) + _, next = b.Slice(next) + } + return offsets +} + +func (b *Buffer) Data(offset int) []byte { + if offset > b.curSz { + panic("offset beyond current size") + } + return b.buf[offset:b.curSz] +} + +// Write would write p bytes to the buffer. +func (b *Buffer) Write(p []byte) (n int, err error) { + n = len(p) + b.Grow(n) + assert(n == copy(b.buf[b.offset:], p)) + b.offset += uint64(n) + return n, nil +} + +// Reset would reset the buffer to be reused. +func (b *Buffer) Reset() { + b.offset = uint64(b.StartOffset()) +} + +// Release would free up the memory allocated by the buffer. Once the usage of buffer is done, it is +// important to call Release, otherwise a memory leak can happen. +func (b *Buffer) Release() error { + if b == nil { + return nil + } + switch b.bufType { + case UseCalloc: + Free(b.buf) + case UseMmap: + if b.mmapFile == nil { + return nil + } + path := b.mmapFile.Fd.Name() + if err := b.mmapFile.Close(-1); err != nil { + return errors.Wrapf(err, "while closing file: %s", path) + } + if !b.persistent { + if err := os.Remove(path); err != nil { + return errors.Wrapf(err, "while deleting file %s", path) + } + } + } + return nil +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/calloc.go b/vendor/github.com/dgraph-io/ristretto/z/calloc.go new file mode 100644 index 00000000000..2e5d6138137 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/calloc.go @@ -0,0 +1,42 @@ +package z + +import "sync/atomic" + +var numBytes int64 + +// NumAllocBytes returns the number of bytes allocated using calls to z.Calloc. The allocations +// could be happening via either Go or jemalloc, depending upon the build flags. +func NumAllocBytes() int64 { + return atomic.LoadInt64(&numBytes) +} + +// MemStats is used to fetch JE Malloc Stats. The stats are fetched from +// the mallctl namespace http://jemalloc.net/jemalloc.3.html#mallctl_namespace. +type MemStats struct { + // Total number of bytes allocated by the application. + // http://jemalloc.net/jemalloc.3.html#stats.allocated + Allocated uint64 + // Total number of bytes in active pages allocated by the application. This + // is a multiple of the page size, and greater than or equal to + // Allocated. + // http://jemalloc.net/jemalloc.3.html#stats.active + Active uint64 + // Maximum number of bytes in physically resident data pages mapped by the + // allocator, comprising all pages dedicated to allocator metadata, pages + // backing active allocations, and unused dirty pages. This is a maximum + // rather than precise because pages may not actually be physically + // resident if they correspond to demand-zeroed virtual memory that has not + // yet been touched. This is a multiple of the page size, and is larger + // than stats.active. + // http://jemalloc.net/jemalloc.3.html#stats.resident + Resident uint64 + // Total number of bytes in virtual memory mappings that were retained + // rather than being returned to the operating system via e.g. munmap(2) or + // similar. Retained virtual memory is typically untouched, decommitted, or + // purged, so it has no strongly associated physical memory (see extent + // hooks http://jemalloc.net/jemalloc.3.html#arena.i.extent_hooks for + // details). Retained memory is excluded from mapped memory statistics, + // e.g. stats.mapped (http://jemalloc.net/jemalloc.3.html#stats.mapped). + // http://jemalloc.net/jemalloc.3.html#stats.retained + Retained uint64 +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/calloc_32bit.go b/vendor/github.com/dgraph-io/ristretto/z/calloc_32bit.go new file mode 100644 index 00000000000..3a0442614f6 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/calloc_32bit.go @@ -0,0 +1,14 @@ +// Copyright 2020 The LevelDB-Go and Pebble Authors. All rights reserved. Use +// of this source code is governed by a BSD-style license that can be found in +// the LICENSE file. + +// +build 386 amd64p32 arm armbe mips mipsle mips64p32 mips64p32le ppc sparc + +package z + +const ( + // MaxArrayLen is a safe maximum length for slices on this architecture. + MaxArrayLen = 1<<31 - 1 + // MaxBufferSize is the size of virtually unlimited buffer on this architecture. + MaxBufferSize = 1 << 30 +) diff --git a/vendor/github.com/dgraph-io/ristretto/z/calloc_64bit.go b/vendor/github.com/dgraph-io/ristretto/z/calloc_64bit.go new file mode 100644 index 00000000000..b898248bbab --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/calloc_64bit.go @@ -0,0 +1,14 @@ +// Copyright 2020 The LevelDB-Go and Pebble Authors. All rights reserved. Use +// of this source code is governed by a BSD-style license that can be found in +// the LICENSE file. + +// +build amd64 arm64 arm64be ppc64 ppc64le mips64 mips64le riscv64 s390x sparc64 + +package z + +const ( + // MaxArrayLen is a safe maximum length for slices on this architecture. + MaxArrayLen = 1<<50 - 1 + // MaxBufferSize is the size of virtually unlimited buffer on this architecture. + MaxBufferSize = 256 << 30 +) diff --git a/vendor/github.com/dgraph-io/ristretto/z/calloc_jemalloc.go b/vendor/github.com/dgraph-io/ristretto/z/calloc_jemalloc.go new file mode 100644 index 00000000000..904d73ac578 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/calloc_jemalloc.go @@ -0,0 +1,172 @@ +// Copyright 2020 The LevelDB-Go and Pebble Authors. All rights reserved. Use +// of this source code is governed by a BSD-style license that can be found in +// the LICENSE file. + +// +build jemalloc + +package z + +/* +#cgo LDFLAGS: /usr/local/lib/libjemalloc.a -L/usr/local/lib -Wl,-rpath,/usr/local/lib -ljemalloc -lm -lstdc++ -pthread -ldl +#include +#include +*/ +import "C" +import ( + "bytes" + "fmt" + "sync" + "sync/atomic" + "unsafe" + + "github.com/dustin/go-humanize" +) + +// The go:linkname directives provides backdoor access to private functions in +// the runtime. Below we're accessing the throw function. + +//go:linkname throw runtime.throw +func throw(s string) + +// New allocates a slice of size n. The returned slice is from manually managed +// memory and MUST be released by calling Free. Failure to do so will result in +// a memory leak. +// +// Compile jemalloc with ./configure --with-jemalloc-prefix="je_" +// https://android.googlesource.com/platform/external/jemalloc_new/+/6840b22e8e11cb68b493297a5cd757d6eaa0b406/TUNING.md +// These two config options seems useful for frequent allocations and deallocations in +// multi-threaded programs (like we have). +// JE_MALLOC_CONF="background_thread:true,metadata_thp:auto" +// +// Compile Go program with `go build -tags=jemalloc` to enable this. + +type dalloc struct { + t string + sz int +} + +var dallocsMu sync.Mutex +var dallocs map[unsafe.Pointer]*dalloc + +func init() { + // By initializing dallocs, we can start tracking allocations and deallocations via z.Calloc. + dallocs = make(map[unsafe.Pointer]*dalloc) +} + +func Calloc(n int, tag string) []byte { + if n == 0 { + return make([]byte, 0) + } + // We need to be conscious of the Cgo pointer passing rules: + // + // https://golang.org/cmd/cgo/#hdr-Passing_pointers + // + // ... + // Note: the current implementation has a bug. While Go code is permitted + // to write nil or a C pointer (but not a Go pointer) to C memory, the + // current implementation may sometimes cause a runtime error if the + // contents of the C memory appear to be a Go pointer. Therefore, avoid + // passing uninitialized C memory to Go code if the Go code is going to + // store pointer values in it. Zero out the memory in C before passing it + // to Go. + + ptr := C.je_calloc(C.size_t(n), 1) + if ptr == nil { + // NB: throw is like panic, except it guarantees the process will be + // terminated. The call below is exactly what the Go runtime invokes when + // it cannot allocate memory. + throw("out of memory") + } + + uptr := unsafe.Pointer(ptr) + dallocsMu.Lock() + dallocs[uptr] = &dalloc{ + t: tag, + sz: n, + } + dallocsMu.Unlock() + atomic.AddInt64(&numBytes, int64(n)) + // Interpret the C pointer as a pointer to a Go array, then slice. + return (*[MaxArrayLen]byte)(uptr)[:n:n] +} + +// CallocNoRef does the exact same thing as Calloc with jemalloc enabled. +func CallocNoRef(n int, tag string) []byte { + return Calloc(n, tag) +} + +// Free frees the specified slice. +func Free(b []byte) { + if sz := cap(b); sz != 0 { + b = b[:cap(b)] + ptr := unsafe.Pointer(&b[0]) + C.je_free(ptr) + atomic.AddInt64(&numBytes, -int64(sz)) + dallocsMu.Lock() + delete(dallocs, ptr) + dallocsMu.Unlock() + } +} + +func Leaks() string { + if dallocs == nil { + return "Leak detection disabled. Enable with 'leak' build flag." + } + dallocsMu.Lock() + defer dallocsMu.Unlock() + if len(dallocs) == 0 { + return "NO leaks found." + } + m := make(map[string]int) + for _, da := range dallocs { + m[da.t] += da.sz + } + var buf bytes.Buffer + fmt.Fprintf(&buf, "Allocations:\n") + for f, sz := range m { + fmt.Fprintf(&buf, "%s at file: %s\n", humanize.IBytes(uint64(sz)), f) + } + return buf.String() +} + +// ReadMemStats populates stats with JE Malloc statistics. +func ReadMemStats(stats *MemStats) { + if stats == nil { + return + } + // Call an epoch mallclt to refresh the stats data as mentioned in the docs. + // http://jemalloc.net/jemalloc.3.html#epoch + // Note: This epoch mallctl is as expensive as a malloc call. It takes up the + // malloc_mutex_lock. + epoch := 1 + sz := unsafe.Sizeof(&epoch) + C.je_mallctl( + (C.CString)("epoch"), + unsafe.Pointer(&epoch), + (*C.size_t)(unsafe.Pointer(&sz)), + unsafe.Pointer(&epoch), + (C.size_t)(unsafe.Sizeof(epoch))) + stats.Allocated = fetchStat("stats.allocated") + stats.Active = fetchStat("stats.active") + stats.Resident = fetchStat("stats.resident") + stats.Retained = fetchStat("stats.retained") +} + +// fetchStat is used to read a specific attribute from je malloc stats using mallctl. +func fetchStat(s string) uint64 { + var out uint64 + sz := unsafe.Sizeof(&out) + C.je_mallctl( + (C.CString)(s), // Query: eg: stats.allocated, stats.resident, etc. + unsafe.Pointer(&out), // Variable to store the output. + (*C.size_t)(unsafe.Pointer(&sz)), // Size of the output variable. + nil, // Input variable used to set a value. + 0) // Size of the input variable. + return out +} + +func StatsPrint() { + opts := C.CString("mdablxe") + C.je_malloc_stats_print(nil, nil, opts) + C.free(unsafe.Pointer(opts)) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/calloc_nojemalloc.go b/vendor/github.com/dgraph-io/ristretto/z/calloc_nojemalloc.go new file mode 100644 index 00000000000..93ceedf9069 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/calloc_nojemalloc.go @@ -0,0 +1,37 @@ +// Copyright 2020 The LevelDB-Go and Pebble Authors. All rights reserved. Use +// of this source code is governed by a BSD-style license that can be found in +// the LICENSE file. + +// +build !jemalloc !cgo + +package z + +import ( + "fmt" +) + +// Provides versions of Calloc, CallocNoRef, etc when jemalloc is not available +// (eg: build without jemalloc tag). + +// Calloc allocates a slice of size n. +func Calloc(n int, tag string) []byte { + return make([]byte, n) +} + +// CallocNoRef will not give you memory back without jemalloc. +func CallocNoRef(n int, tag string) []byte { + // We do the add here just to stay compatible with a corresponding Free call. + return nil +} + +// Free does not do anything in this mode. +func Free(b []byte) {} + +func Leaks() string { return "Leaks: Using Go memory" } +func StatsPrint() { + fmt.Println("Using Go memory") +} + +// ReadMemStats doesn't do anything since all the memory is being managed +// by the Go runtime. +func ReadMemStats(_ *MemStats) { return } diff --git a/vendor/github.com/dgraph-io/ristretto/z/file.go b/vendor/github.com/dgraph-io/ristretto/z/file.go new file mode 100644 index 00000000000..880caf0ad9c --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/file.go @@ -0,0 +1,217 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "encoding/binary" + "fmt" + "io" + "os" + "path/filepath" + + "github.com/pkg/errors" +) + +// MmapFile represents an mmapd file and includes both the buffer to the data +// and the file descriptor. +type MmapFile struct { + Data []byte + Fd *os.File +} + +var NewFile = errors.New("Create a new file") + +func OpenMmapFileUsing(fd *os.File, sz int, writable bool) (*MmapFile, error) { + filename := fd.Name() + fi, err := fd.Stat() + if err != nil { + return nil, errors.Wrapf(err, "cannot stat file: %s", filename) + } + + var rerr error + fileSize := fi.Size() + if sz > 0 && fileSize == 0 { + // If file is empty, truncate it to sz. + if err := fd.Truncate(int64(sz)); err != nil { + return nil, errors.Wrapf(err, "error while truncation") + } + fileSize = int64(sz) + rerr = NewFile + } + + // fmt.Printf("Mmaping file: %s with writable: %v filesize: %d\n", fd.Name(), writable, fileSize) + buf, err := Mmap(fd, writable, fileSize) // Mmap up to file size. + if err != nil { + return nil, errors.Wrapf(err, "while mmapping %s with size: %d", fd.Name(), fileSize) + } + + if fileSize == 0 { + dir, _ := filepath.Split(filename) + go SyncDir(dir) + } + return &MmapFile{ + Data: buf, + Fd: fd, + }, rerr +} + +// OpenMmapFile opens an existing file or creates a new file. If the file is +// created, it would truncate the file to maxSz. In both cases, it would mmap +// the file to maxSz and returned it. In case the file is created, z.NewFile is +// returned. +func OpenMmapFile(filename string, flag int, maxSz int) (*MmapFile, error) { + // fmt.Printf("opening file %s with flag: %v\n", filename, flag) + fd, err := os.OpenFile(filename, flag, 0666) + if err != nil { + return nil, errors.Wrapf(err, "unable to open: %s", filename) + } + writable := true + if flag == os.O_RDONLY { + writable = false + } + return OpenMmapFileUsing(fd, maxSz, writable) +} + +type mmapReader struct { + Data []byte + offset int +} + +func (mr *mmapReader) Read(buf []byte) (int, error) { + if mr.offset > len(mr.Data) { + return 0, io.EOF + } + n := copy(buf, mr.Data[mr.offset:]) + mr.offset += n + if n < len(buf) { + return n, io.EOF + } + return n, nil +} + +func (m *MmapFile) NewReader(offset int) io.Reader { + return &mmapReader{ + Data: m.Data, + offset: offset, + } +} + +// Bytes returns data starting from offset off of size sz. If there's not enough data, it would +// return nil slice and io.EOF. +func (m *MmapFile) Bytes(off, sz int) ([]byte, error) { + if len(m.Data[off:]) < sz { + return nil, io.EOF + } + return m.Data[off : off+sz], nil +} + +// Slice returns the slice at the given offset. +func (m *MmapFile) Slice(offset int) []byte { + sz := binary.BigEndian.Uint32(m.Data[offset:]) + start := offset + 4 + next := start + int(sz) + if next > len(m.Data) { + return []byte{} + } + res := m.Data[start:next] + return res +} + +// AllocateSlice allocates a slice of the given size at the given offset. +func (m *MmapFile) AllocateSlice(sz, offset int) ([]byte, int, error) { + start := offset + 4 + + // If the file is too small, double its size or increase it by 1GB, whichever is smaller. + if start+sz > len(m.Data) { + const oneGB = 1 << 30 + growBy := len(m.Data) + if growBy > oneGB { + growBy = oneGB + } + if growBy < sz+4 { + growBy = sz + 4 + } + if err := m.Truncate(int64(len(m.Data) + growBy)); err != nil { + return nil, 0, err + } + } + + binary.BigEndian.PutUint32(m.Data[offset:], uint32(sz)) + return m.Data[start : start+sz], start + sz, nil +} + +func (m *MmapFile) Sync() error { + if m == nil { + return nil + } + return Msync(m.Data) +} + +func (m *MmapFile) Delete() error { + // Badger can set the m.Data directly, without setting any Fd. In that case, this should be a + // NOOP. + if m.Fd == nil { + return nil + } + + if err := Munmap(m.Data); err != nil { + return fmt.Errorf("while munmap file: %s, error: %v\n", m.Fd.Name(), err) + } + m.Data = nil + if err := m.Fd.Truncate(0); err != nil { + return fmt.Errorf("while truncate file: %s, error: %v\n", m.Fd.Name(), err) + } + if err := m.Fd.Close(); err != nil { + return fmt.Errorf("while close file: %s, error: %v\n", m.Fd.Name(), err) + } + return os.Remove(m.Fd.Name()) +} + +// Close would close the file. It would also truncate the file if maxSz >= 0. +func (m *MmapFile) Close(maxSz int64) error { + // Badger can set the m.Data directly, without setting any Fd. In that case, this should be a + // NOOP. + if m.Fd == nil { + return nil + } + if err := m.Sync(); err != nil { + return fmt.Errorf("while sync file: %s, error: %v\n", m.Fd.Name(), err) + } + if err := Munmap(m.Data); err != nil { + return fmt.Errorf("while munmap file: %s, error: %v\n", m.Fd.Name(), err) + } + if maxSz >= 0 { + if err := m.Fd.Truncate(maxSz); err != nil { + return fmt.Errorf("while truncate file: %s, error: %v\n", m.Fd.Name(), err) + } + } + return m.Fd.Close() +} + +func SyncDir(dir string) error { + df, err := os.Open(dir) + if err != nil { + return errors.Wrapf(err, "while opening %s", dir) + } + if err := df.Sync(); err != nil { + return errors.Wrapf(err, "while syncing %s", dir) + } + if err := df.Close(); err != nil { + return errors.Wrapf(err, "while closing %s", dir) + } + return nil +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/file_default.go b/vendor/github.com/dgraph-io/ristretto/z/file_default.go new file mode 100644 index 00000000000..d9c0db43e7f --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/file_default.go @@ -0,0 +1,39 @@ +// +build !linux + +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import "fmt" + +// Truncate would truncate the mmapped file to the given size. On Linux, we truncate +// the underlying file and then call mremap, but on other systems, we unmap first, +// then truncate, then re-map. +func (m *MmapFile) Truncate(maxSz int64) error { + if err := m.Sync(); err != nil { + return fmt.Errorf("while sync file: %s, error: %v\n", m.Fd.Name(), err) + } + if err := Munmap(m.Data); err != nil { + return fmt.Errorf("while munmap file: %s, error: %v\n", m.Fd.Name(), err) + } + if err := m.Fd.Truncate(maxSz); err != nil { + return fmt.Errorf("while truncate file: %s, error: %v\n", m.Fd.Name(), err) + } + var err error + m.Data, err = Mmap(m.Fd, true, maxSz) // Mmap up to max size. + return err +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/file_linux.go b/vendor/github.com/dgraph-io/ristretto/z/file_linux.go new file mode 100644 index 00000000000..7f670bd2cc0 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/file_linux.go @@ -0,0 +1,37 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "fmt" +) + +// Truncate would truncate the mmapped file to the given size. On Linux, we truncate +// the underlying file and then call mremap, but on other systems, we unmap first, +// then truncate, then re-map. +func (m *MmapFile) Truncate(maxSz int64) error { + if err := m.Sync(); err != nil { + return fmt.Errorf("while sync file: %s, error: %v\n", m.Fd.Name(), err) + } + if err := m.Fd.Truncate(maxSz); err != nil { + return fmt.Errorf("while truncate file: %s, error: %v\n", m.Fd.Name(), err) + } + + var err error + m.Data, err = mremap(m.Data, int(maxSz)) // Mmap up to max size. + return err +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/flags.go b/vendor/github.com/dgraph-io/ristretto/z/flags.go new file mode 100644 index 00000000000..a55c474ab2b --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/flags.go @@ -0,0 +1,311 @@ +package z + +import ( + "fmt" + "os" + "os/user" + "path/filepath" + "sort" + "strconv" + "strings" + "time" + + "github.com/golang/glog" + "github.com/pkg/errors" +) + +// SuperFlagHelp makes it really easy to generate command line `--help` output for a SuperFlag. For +// example: +// +// const flagDefaults = `enabled=true; path=some/path;` +// +// var help string = z.NewSuperFlagHelp(flagDefaults). +// Flag("enabled", "Turns on ."). +// Flag("path", "The path to ."). +// Flag("another", "Not present in defaults, but still included."). +// String() +// +// The `help` string would then contain: +// +// enabled=true; Turns on . +// path=some/path; The path to . +// another=; Not present in defaults, but still included. +// +// All flags are sorted alphabetically for consistent `--help` output. Flags with default values are +// placed at the top, and everything else goes under. +type SuperFlagHelp struct { + head string + defaults *SuperFlag + flags map[string]string +} + +func NewSuperFlagHelp(defaults string) *SuperFlagHelp { + return &SuperFlagHelp{ + defaults: NewSuperFlag(defaults), + flags: make(map[string]string, 0), + } +} + +func (h *SuperFlagHelp) Head(head string) *SuperFlagHelp { + h.head = head + return h +} + +func (h *SuperFlagHelp) Flag(name, description string) *SuperFlagHelp { + h.flags[name] = description + return h +} + +func (h *SuperFlagHelp) String() string { + defaultLines := make([]string, 0) + otherLines := make([]string, 0) + for name, help := range h.flags { + val, found := h.defaults.m[name] + line := fmt.Sprintf(" %s=%s; %s\n", name, val, help) + if found { + defaultLines = append(defaultLines, line) + } else { + otherLines = append(otherLines, line) + } + } + sort.Strings(defaultLines) + sort.Strings(otherLines) + dls := strings.Join(defaultLines, "") + ols := strings.Join(otherLines, "") + if len(h.defaults.m) == 0 && len(ols) == 0 { + // remove last newline + dls = dls[:len(dls)-1] + } + // remove last newline + if len(h.defaults.m) == 0 && len(ols) > 1 { + ols = ols[:len(ols)-1] + } + return h.head + "\n" + dls + ols +} + +func parseFlag(flag string) (map[string]string, error) { + kvm := make(map[string]string) + for _, kv := range strings.Split(flag, ";") { + if strings.TrimSpace(kv) == "" { + continue + } + // For a non-empty separator, 0 < len(splits) ≤ 2. + splits := strings.SplitN(kv, "=", 2) + k := strings.TrimSpace(splits[0]) + if len(splits) < 2 { + return nil, fmt.Errorf("superflag: missing value for '%s' in flag: %s", k, flag) + } + k = strings.ToLower(k) + k = strings.ReplaceAll(k, "_", "-") + kvm[k] = strings.TrimSpace(splits[1]) + } + return kvm, nil +} + +type SuperFlag struct { + m map[string]string +} + +func NewSuperFlag(flag string) *SuperFlag { + sf, err := newSuperFlagImpl(flag) + if err != nil { + glog.Fatal(err) + } + return sf +} + +func newSuperFlagImpl(flag string) (*SuperFlag, error) { + m, err := parseFlag(flag) + if err != nil { + return nil, err + } + return &SuperFlag{m}, nil +} + +func (sf *SuperFlag) String() string { + if sf == nil { + return "" + } + kvs := make([]string, 0, len(sf.m)) + for k, v := range sf.m { + kvs = append(kvs, fmt.Sprintf("%s=%s", k, v)) + } + return strings.Join(kvs, "; ") +} + +func (sf *SuperFlag) MergeAndCheckDefault(flag string) *SuperFlag { + sf, err := sf.mergeAndCheckDefaultImpl(flag) + if err != nil { + glog.Fatal(err) + } + return sf +} + +func (sf *SuperFlag) mergeAndCheckDefaultImpl(flag string) (*SuperFlag, error) { + if sf == nil { + m, err := parseFlag(flag) + if err != nil { + return nil, err + } + return &SuperFlag{m}, nil + } + + src, err := parseFlag(flag) + if err != nil { + return nil, err + } + + numKeys := len(sf.m) + for k := range src { + if _, ok := sf.m[k]; ok { + numKeys-- + } + } + if numKeys != 0 { + return nil, fmt.Errorf("superflag: found invalid options in flag: %s.\nvalid options: %v", sf, flag) + } + for k, v := range src { + if _, ok := sf.m[k]; !ok { + sf.m[k] = v + } + } + return sf, nil +} + +func (sf *SuperFlag) Has(opt string) bool { + val := sf.GetString(opt) + return val != "" +} + +func (sf *SuperFlag) GetDuration(opt string) time.Duration { + val := sf.GetString(opt) + if val == "" { + return time.Duration(0) + } + if strings.Contains(val, "d") { + val = strings.Replace(val, "d", "", 1) + days, err := strconv.ParseUint(val, 0, 64) + if err != nil { + return time.Duration(0) + } + return time.Hour * 24 * time.Duration(days) + } + d, err := time.ParseDuration(val) + if err != nil { + return time.Duration(0) + } + return d +} + +func (sf *SuperFlag) GetBool(opt string) bool { + val := sf.GetString(opt) + if val == "" { + return false + } + b, err := strconv.ParseBool(val) + if err != nil { + err = errors.Wrapf(err, + "Unable to parse %s as bool for key: %s. Options: %s\n", + val, opt, sf) + glog.Fatalf("%+v", err) + } + return b +} + +func (sf *SuperFlag) GetFloat64(opt string) float64 { + val := sf.GetString(opt) + if val == "" { + return 0 + } + f, err := strconv.ParseFloat(val, 64) + if err != nil { + err = errors.Wrapf(err, + "Unable to parse %s as float64 for key: %s. Options: %s\n", + val, opt, sf) + glog.Fatalf("%+v", err) + } + return f +} + +func (sf *SuperFlag) GetInt64(opt string) int64 { + val := sf.GetString(opt) + if val == "" { + return 0 + } + i, err := strconv.ParseInt(val, 0, 64) + if err != nil { + err = errors.Wrapf(err, + "Unable to parse %s as int64 for key: %s. Options: %s\n", + val, opt, sf) + glog.Fatalf("%+v", err) + } + return i +} + +func (sf *SuperFlag) GetUint64(opt string) uint64 { + val := sf.GetString(opt) + if val == "" { + return 0 + } + u, err := strconv.ParseUint(val, 0, 64) + if err != nil { + err = errors.Wrapf(err, + "Unable to parse %s as uint64 for key: %s. Options: %s\n", + val, opt, sf) + glog.Fatalf("%+v", err) + } + return u +} + +func (sf *SuperFlag) GetUint32(opt string) uint32 { + val := sf.GetString(opt) + if val == "" { + return 0 + } + u, err := strconv.ParseUint(val, 0, 32) + if err != nil { + err = errors.Wrapf(err, + "Unable to parse %s as uint32 for key: %s. Options: %s\n", + val, opt, sf) + glog.Fatalf("%+v", err) + } + return uint32(u) +} + +func (sf *SuperFlag) GetString(opt string) string { + if sf == nil { + return "" + } + return sf.m[opt] +} + +func (sf *SuperFlag) GetPath(opt string) string { + p := sf.GetString(opt) + path, err := expandPath(p) + if err != nil { + glog.Fatalf("Failed to get path: %+v", err) + } + return path +} + +// expandPath expands the paths containing ~ to /home/user. It also computes the absolute path +// from the relative paths. For example: ~/abc/../cef will be transformed to /home/user/cef. +func expandPath(path string) (string, error) { + if len(path) == 0 { + return "", nil + } + if path[0] == '~' && (len(path) == 1 || os.IsPathSeparator(path[1])) { + usr, err := user.Current() + if err != nil { + return "", errors.Wrap(err, "Failed to get the home directory of the user") + } + path = filepath.Join(usr.HomeDir, path[1:]) + } + + var err error + path, err = filepath.Abs(path) + if err != nil { + return "", errors.Wrap(err, "Failed to generate absolute path") + } + return path, nil +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/histogram.go b/vendor/github.com/dgraph-io/ristretto/z/histogram.go new file mode 100644 index 00000000000..4eb0c4f6c98 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/histogram.go @@ -0,0 +1,205 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "fmt" + "math" + "strings" + + "github.com/dustin/go-humanize" +) + +// Creates bounds for an histogram. The bounds are powers of two of the form +// [2^min_exponent, ..., 2^max_exponent]. +func HistogramBounds(minExponent, maxExponent uint32) []float64 { + var bounds []float64 + for i := minExponent; i <= maxExponent; i++ { + bounds = append(bounds, float64(int(1)< 4) + bounds := make([]float64, num) + bounds[0] = 1 + bounds[1] = 2 + for i := 2; i < num; i++ { + bounds[i] = bounds[i-1] + bounds[i-2] + } + return bounds +} + +// HistogramData stores the information needed to represent the sizes of the keys and values +// as a histogram. +type HistogramData struct { + Bounds []float64 + Count int64 + CountPerBucket []int64 + Min int64 + Max int64 + Sum int64 +} + +// NewHistogramData returns a new instance of HistogramData with properly initialized fields. +func NewHistogramData(bounds []float64) *HistogramData { + return &HistogramData{ + Bounds: bounds, + CountPerBucket: make([]int64, len(bounds)+1), + Max: 0, + Min: math.MaxInt64, + } +} + +func (histogram *HistogramData) Copy() *HistogramData { + if histogram == nil { + return nil + } + return &HistogramData{ + Bounds: append([]float64{}, histogram.Bounds...), + CountPerBucket: append([]int64{}, histogram.CountPerBucket...), + Count: histogram.Count, + Min: histogram.Min, + Max: histogram.Max, + Sum: histogram.Sum, + } +} + +// Update changes the Min and Max fields if value is less than or greater than the current values. +func (histogram *HistogramData) Update(value int64) { + if histogram == nil { + return + } + if value > histogram.Max { + histogram.Max = value + } + if value < histogram.Min { + histogram.Min = value + } + + histogram.Sum += value + histogram.Count++ + + for index := 0; index <= len(histogram.Bounds); index++ { + // Allocate value in the last buckets if we reached the end of the Bounds array. + if index == len(histogram.Bounds) { + histogram.CountPerBucket[index]++ + break + } + + if value < int64(histogram.Bounds[index]) { + histogram.CountPerBucket[index]++ + break + } + } +} + +// Mean returns the mean value for the histogram. +func (histogram *HistogramData) Mean() float64 { + if histogram.Count == 0 { + return 0 + } + return float64(histogram.Sum) / float64(histogram.Count) +} + +// String converts the histogram data into human-readable string. +func (histogram *HistogramData) String() string { + if histogram == nil { + return "" + } + var b strings.Builder + + b.WriteString("\n -- Histogram: \n") + b.WriteString(fmt.Sprintf("Min value: %d \n", histogram.Min)) + b.WriteString(fmt.Sprintf("Max value: %d \n", histogram.Max)) + b.WriteString(fmt.Sprintf("Count: %d \n", histogram.Count)) + b.WriteString(fmt.Sprintf("50p: %.2f \n", histogram.Percentile(0.5))) + b.WriteString(fmt.Sprintf("75p: %.2f \n", histogram.Percentile(0.75))) + b.WriteString(fmt.Sprintf("90p: %.2f \n", histogram.Percentile(0.90))) + + numBounds := len(histogram.Bounds) + var cum float64 + for index, count := range histogram.CountPerBucket { + if count == 0 { + continue + } + + // The last bucket represents the bucket that contains the range from + // the last bound up to infinity so it's processed differently than the + // other buckets. + if index == len(histogram.CountPerBucket)-1 { + lowerBound := uint64(histogram.Bounds[numBounds-1]) + page := float64(count*100) / float64(histogram.Count) + cum += page + b.WriteString(fmt.Sprintf("[%s, %s) %d %.2f%% %.2f%%\n", + humanize.IBytes(lowerBound), "infinity", count, page, cum)) + continue + } + + upperBound := uint64(histogram.Bounds[index]) + lowerBound := uint64(0) + if index > 0 { + lowerBound = uint64(histogram.Bounds[index-1]) + } + + page := float64(count*100) / float64(histogram.Count) + cum += page + b.WriteString(fmt.Sprintf("[%d, %d) %d %.2f%% %.2f%%\n", + lowerBound, upperBound, count, page, cum)) + } + b.WriteString(" --\n") + return b.String() +} + +// Percentile returns the percentile value for the histogram. +// value of p should be between [0.0-1.0] +func (histogram *HistogramData) Percentile(p float64) float64 { + if histogram == nil { + return 0 + } + + if histogram.Count == 0 { + // if no data return the minimum range + return histogram.Bounds[0] + } + pval := int64(float64(histogram.Count) * p) + for i, v := range histogram.CountPerBucket { + pval = pval - v + if pval <= 0 { + if i == len(histogram.Bounds) { + break + } + return histogram.Bounds[i] + } + } + // default return should be the max range + return histogram.Bounds[len(histogram.Bounds)-1] +} + +// Clear reset the histogram. Helpful in situations where we need to reset the metrics +func (histogram *HistogramData) Clear() { + if histogram == nil { + return + } + + histogram.Count = 0 + histogram.CountPerBucket = make([]int64, len(histogram.Bounds)+1) + histogram.Sum = 0 + histogram.Max = 0 + histogram.Min = math.MaxInt64 +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mmap.go b/vendor/github.com/dgraph-io/ristretto/z/mmap.go new file mode 100644 index 00000000000..9b02510003c --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mmap.go @@ -0,0 +1,44 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "os" +) + +// Mmap uses the mmap system call to memory-map a file. If writable is true, +// memory protection of the pages is set so that they may be written to as well. +func Mmap(fd *os.File, writable bool, size int64) ([]byte, error) { + return mmap(fd, writable, size) +} + +// Munmap unmaps a previously mapped slice. +func Munmap(b []byte) error { + return munmap(b) +} + +// Madvise uses the madvise system call to give advise about the use of memory +// when using a slice that is memory-mapped to a file. Set the readahead flag to +// false if page references are expected in random order. +func Madvise(b []byte, readahead bool) error { + return madvise(b, readahead) +} + +// Msync would call sync on the mmapped data. +func Msync(b []byte) error { + return msync(b) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mmap_darwin.go b/vendor/github.com/dgraph-io/ristretto/z/mmap_darwin.go new file mode 100644 index 00000000000..4d6d74f1930 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mmap_darwin.go @@ -0,0 +1,59 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "os" + "syscall" + "unsafe" + + "golang.org/x/sys/unix" +) + +// Mmap uses the mmap system call to memory-map a file. If writable is true, +// memory protection of the pages is set so that they may be written to as well. +func mmap(fd *os.File, writable bool, size int64) ([]byte, error) { + mtype := unix.PROT_READ + if writable { + mtype |= unix.PROT_WRITE + } + return unix.Mmap(int(fd.Fd()), 0, int(size), mtype, unix.MAP_SHARED) +} + +// Munmap unmaps a previously mapped slice. +func munmap(b []byte) error { + return unix.Munmap(b) +} + +// This is required because the unix package does not support the madvise system call on OS X. +func madvise(b []byte, readahead bool) error { + advice := unix.MADV_NORMAL + if !readahead { + advice = unix.MADV_RANDOM + } + + _, _, e1 := syscall.Syscall(syscall.SYS_MADVISE, uintptr(unsafe.Pointer(&b[0])), + uintptr(len(b)), uintptr(advice)) + if e1 != 0 { + return e1 + } + return nil +} + +func msync(b []byte) error { + return unix.Msync(b, unix.MS_SYNC) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mmap_linux.go b/vendor/github.com/dgraph-io/ristretto/z/mmap_linux.go new file mode 100644 index 00000000000..331330cff91 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mmap_linux.go @@ -0,0 +1,71 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "os" + "unsafe" + + "golang.org/x/sys/unix" +) + +// mmap uses the mmap system call to memory-map a file. If writable is true, +// memory protection of the pages is set so that they may be written to as well. +func mmap(fd *os.File, writable bool, size int64) ([]byte, error) { + mtype := unix.PROT_READ + if writable { + mtype |= unix.PROT_WRITE + } + return unix.Mmap(int(fd.Fd()), 0, int(size), mtype, unix.MAP_SHARED) +} + +// munmap unmaps a previously mapped slice. +// +// unix.Munmap maintains an internal list of mmapped addresses, and only calls munmap +// if the address is present in that list. If we use mremap, this list is not updated. +// To bypass this, we call munmap ourselves. +func munmap(data []byte) error { + if len(data) == 0 || len(data) != cap(data) { + return unix.EINVAL + } + _, _, errno := unix.Syscall( + unix.SYS_MUNMAP, + uintptr(unsafe.Pointer(&data[0])), + uintptr(len(data)), + 0, + ) + if errno != 0 { + return errno + } + return nil +} + +// madvise uses the madvise system call to give advise about the use of memory +// when using a slice that is memory-mapped to a file. Set the readahead flag to +// false if page references are expected in random order. +func madvise(b []byte, readahead bool) error { + flags := unix.MADV_NORMAL + if !readahead { + flags = unix.MADV_RANDOM + } + return unix.Madvise(b, flags) +} + +// msync writes any modified data to persistent storage. +func msync(b []byte) error { + return unix.Msync(b, unix.MS_SYNC) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mmap_plan9.go b/vendor/github.com/dgraph-io/ristretto/z/mmap_plan9.go new file mode 100644 index 00000000000..f30729654f5 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mmap_plan9.go @@ -0,0 +1,44 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "os" + "syscall" +) + +// Mmap uses the mmap system call to memory-map a file. If writable is true, +// memory protection of the pages is set so that they may be written to as well. +func mmap(fd *os.File, writable bool, size int64) ([]byte, error) { + return nil, syscall.EPLAN9 +} + +// Munmap unmaps a previously mapped slice. +func munmap(b []byte) error { + return syscall.EPLAN9 +} + +// Madvise uses the madvise system call to give advise about the use of memory +// when using a slice that is memory-mapped to a file. Set the readahead flag to +// false if page references are expected in random order. +func madvise(b []byte, readahead bool) error { + return syscall.EPLAN9 +} + +func msync(b []byte) error { + return syscall.EPLAN9 +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mmap_unix.go b/vendor/github.com/dgraph-io/ristretto/z/mmap_unix.go new file mode 100644 index 00000000000..e8b2699cf9e --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mmap_unix.go @@ -0,0 +1,55 @@ +// +build !windows,!darwin,!plan9,!linux + +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "os" + + "golang.org/x/sys/unix" +) + +// Mmap uses the mmap system call to memory-map a file. If writable is true, +// memory protection of the pages is set so that they may be written to as well. +func mmap(fd *os.File, writable bool, size int64) ([]byte, error) { + mtype := unix.PROT_READ + if writable { + mtype |= unix.PROT_WRITE + } + return unix.Mmap(int(fd.Fd()), 0, int(size), mtype, unix.MAP_SHARED) +} + +// Munmap unmaps a previously mapped slice. +func munmap(b []byte) error { + return unix.Munmap(b) +} + +// Madvise uses the madvise system call to give advise about the use of memory +// when using a slice that is memory-mapped to a file. Set the readahead flag to +// false if page references are expected in random order. +func madvise(b []byte, readahead bool) error { + flags := unix.MADV_NORMAL + if !readahead { + flags = unix.MADV_RANDOM + } + return unix.Madvise(b, flags) +} + +func msync(b []byte) error { + return unix.Msync(b, unix.MS_SYNC) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mmap_windows.go b/vendor/github.com/dgraph-io/ristretto/z/mmap_windows.go new file mode 100644 index 00000000000..171176e9feb --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mmap_windows.go @@ -0,0 +1,96 @@ +// +build windows + +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "fmt" + "os" + "syscall" + "unsafe" +) + +func mmap(fd *os.File, write bool, size int64) ([]byte, error) { + protect := syscall.PAGE_READONLY + access := syscall.FILE_MAP_READ + + if write { + protect = syscall.PAGE_READWRITE + access = syscall.FILE_MAP_WRITE + } + fi, err := fd.Stat() + if err != nil { + return nil, err + } + + // In windows, we cannot mmap a file more than it's actual size. + // So truncate the file to the size of the mmap. + if fi.Size() < size { + if err := fd.Truncate(size); err != nil { + return nil, fmt.Errorf("truncate: %s", err) + } + } + + // Open a file mapping handle. + sizelo := uint32(size >> 32) + sizehi := uint32(size) & 0xffffffff + + handler, err := syscall.CreateFileMapping(syscall.Handle(fd.Fd()), nil, + uint32(protect), sizelo, sizehi, nil) + if err != nil { + return nil, os.NewSyscallError("CreateFileMapping", err) + } + + // Create the memory map. + addr, err := syscall.MapViewOfFile(handler, uint32(access), 0, 0, uintptr(size)) + if addr == 0 { + return nil, os.NewSyscallError("MapViewOfFile", err) + } + + // Close mapping handle. + if err := syscall.CloseHandle(syscall.Handle(handler)); err != nil { + return nil, os.NewSyscallError("CloseHandle", err) + } + + // Slice memory layout + // Copied this snippet from golang/sys package + var sl = struct { + addr uintptr + len int + cap int + }{addr, int(size), int(size)} + + // Use unsafe to turn sl into a []byte. + data := *(*[]byte)(unsafe.Pointer(&sl)) + + return data, nil +} + +func munmap(b []byte) error { + return syscall.UnmapViewOfFile(uintptr(unsafe.Pointer(&b[0]))) +} + +func madvise(b []byte, readahead bool) error { + // Do Nothing. We don’t care about this setting on Windows + return nil +} + +func msync(b []byte) error { + // TODO: Figure out how to do msync on Windows. + return nil +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mremap_linux.go b/vendor/github.com/dgraph-io/ristretto/z/mremap_linux.go new file mode 100644 index 00000000000..225678658d5 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mremap_linux.go @@ -0,0 +1,56 @@ +// +build !arm64 + +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "fmt" + "reflect" + "unsafe" + + "golang.org/x/sys/unix" +) + +// mremap is a Linux-specific system call to remap pages in memory. This can be used in place of munmap + mmap. +func mremap(data []byte, size int) ([]byte, error) { + //nolint:lll + // taken from + const MREMAP_MAYMOVE = 0x1 + + header := (*reflect.SliceHeader)(unsafe.Pointer(&data)) + mmapAddr, mmapSize, errno := unix.Syscall6( + unix.SYS_MREMAP, + header.Data, + uintptr(header.Len), + uintptr(size), + uintptr(MREMAP_MAYMOVE), + 0, + 0, + ) + if errno != 0 { + return nil, errno + } + if mmapSize != uintptr(size) { + return nil, fmt.Errorf("mremap size mismatch: requested: %d got: %d", size, mmapSize) + } + + header.Data = mmapAddr + header.Cap = size + header.Len = size + return data, nil +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/mremap_linux_arm64.go b/vendor/github.com/dgraph-io/ristretto/z/mremap_linux_arm64.go new file mode 100644 index 00000000000..09683cdfeb5 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/mremap_linux_arm64.go @@ -0,0 +1,52 @@ +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "reflect" + "unsafe" + + "golang.org/x/sys/unix" +) + +// mremap is a Linux-specific system call to remap pages in memory. This can be used in place of munmap + mmap. +func mremap(data []byte, size int) ([]byte, error) { + //nolint:lll + // taken from + const MREMAP_MAYMOVE = 0x1 + + header := (*reflect.SliceHeader)(unsafe.Pointer(&data)) + // For ARM64, the second return argument for SYS_MREMAP is inconsistent (prior allocated size) with + // other architectures, which return the size allocated + mmapAddr, _, errno := unix.Syscall6( + unix.SYS_MREMAP, + header.Data, + uintptr(header.Len), + uintptr(size), + uintptr(MREMAP_MAYMOVE), + 0, + 0, + ) + if errno != 0 { + return nil, errno + } + + header.Data = mmapAddr + header.Cap = size + header.Len = size + return data, nil +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/rtutil.go b/vendor/github.com/dgraph-io/ristretto/z/rtutil.go new file mode 100644 index 00000000000..8f317c80d30 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/rtutil.go @@ -0,0 +1,75 @@ +// MIT License + +// Copyright (c) 2019 Ewan Chou + +// Permission is hereby granted, free of charge, to any person obtaining a copy +// of this software and associated documentation files (the "Software"), to deal +// in the Software without restriction, including without limitation the rights +// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell +// copies of the Software, and to permit persons to whom the Software is +// furnished to do so, subject to the following conditions: + +// The above copyright notice and this permission notice shall be included in all +// copies or substantial portions of the Software. + +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, +// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. + +package z + +import ( + "unsafe" +) + +// NanoTime returns the current time in nanoseconds from a monotonic clock. +//go:linkname NanoTime runtime.nanotime +func NanoTime() int64 + +// CPUTicks is a faster alternative to NanoTime to measure time duration. +//go:linkname CPUTicks runtime.cputicks +func CPUTicks() int64 + +type stringStruct struct { + str unsafe.Pointer + len int +} + +//go:noescape +//go:linkname memhash runtime.memhash +func memhash(p unsafe.Pointer, h, s uintptr) uintptr + +// MemHash is the hash function used by go map, it utilizes available hardware instructions(behaves +// as aeshash if aes instruction is available). +// NOTE: The hash seed changes for every process. So, this cannot be used as a persistent hash. +func MemHash(data []byte) uint64 { + ss := (*stringStruct)(unsafe.Pointer(&data)) + return uint64(memhash(ss.str, 0, uintptr(ss.len))) +} + +// MemHashString is the hash function used by go map, it utilizes available hardware instructions +// (behaves as aeshash if aes instruction is available). +// NOTE: The hash seed changes for every process. So, this cannot be used as a persistent hash. +func MemHashString(str string) uint64 { + ss := (*stringStruct)(unsafe.Pointer(&str)) + return uint64(memhash(ss.str, 0, uintptr(ss.len))) +} + +// FastRand is a fast thread local random function. +//go:linkname FastRand runtime.fastrand +func FastRand() uint32 + +//go:linkname memclrNoHeapPointers runtime.memclrNoHeapPointers +func memclrNoHeapPointers(p unsafe.Pointer, n uintptr) + +func Memclr(b []byte) { + if len(b) == 0 { + return + } + p := unsafe.Pointer(&b[0]) + memclrNoHeapPointers(p, uintptr(len(b))) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/rtutil.s b/vendor/github.com/dgraph-io/ristretto/z/rtutil.s new file mode 100644 index 00000000000..e69de29bb2d diff --git a/vendor/github.com/dgraph-io/ristretto/z/simd/baseline.go b/vendor/github.com/dgraph-io/ristretto/z/simd/baseline.go new file mode 100644 index 00000000000..967e3a307e9 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/simd/baseline.go @@ -0,0 +1,127 @@ +package simd + +import ( + "fmt" + "runtime" + "sort" + "sync" +) + +// Search finds the key using the naive way +func Naive(xs []uint64, k uint64) int16 { + var i int + for i = 0; i < len(xs); i += 2 { + x := xs[i] + if x >= k { + return int16(i / 2) + } + } + return int16(i / 2) +} + +func Clever(xs []uint64, k uint64) int16 { + if len(xs) < 8 { + return Naive(xs, k) + } + var twos, pk [4]uint64 + pk[0] = k + pk[1] = k + pk[2] = k + pk[3] = k + for i := 0; i < len(xs); i += 8 { + twos[0] = xs[i] + twos[1] = xs[i+2] + twos[2] = xs[i+4] + twos[3] = xs[i+6] + if twos[0] >= pk[0] { + return int16(i / 2) + } + if twos[1] >= pk[1] { + return int16((i + 2) / 2) + } + if twos[2] >= pk[2] { + return int16((i + 4) / 2) + } + if twos[3] >= pk[3] { + return int16((i + 6) / 2) + } + + } + return int16(len(xs) / 2) +} + +func Parallel(xs []uint64, k uint64) int16 { + cpus := runtime.NumCPU() + if cpus%2 != 0 { + panic(fmt.Sprintf("odd number of CPUs %v", cpus)) + } + sz := len(xs)/cpus + 1 + var wg sync.WaitGroup + retChan := make(chan int16, cpus) + for i := 0; i < len(xs); i += sz { + end := i + sz + if end >= len(xs) { + end = len(xs) + } + chunk := xs[i:end] + wg.Add(1) + go func(hd int16, xs []uint64, k uint64, wg *sync.WaitGroup, ch chan int16) { + for i := 0; i < len(xs); i += 2 { + if xs[i] >= k { + ch <- (int16(i) + hd) / 2 + break + } + } + wg.Done() + }(int16(i), chunk, k, &wg, retChan) + } + wg.Wait() + close(retChan) + var min int16 = (1 << 15) - 1 + for i := range retChan { + if i < min { + min = i + } + } + if min == (1<<15)-1 { + return int16(len(xs) / 2) + } + return min +} + +func Binary(keys []uint64, key uint64) int16 { + return int16(sort.Search(len(keys), func(i int) bool { + if i*2 >= len(keys) { + return true + } + return keys[i*2] >= key + })) +} + +func cmp2_native(twos, pk [2]uint64) int16 { + if twos[0] == pk[0] { + return 0 + } + if twos[1] == pk[1] { + return 1 + } + return 2 +} + +func cmp4_native(fours, pk [4]uint64) int16 { + for i := range fours { + if fours[i] >= pk[i] { + return int16(i) + } + } + return 4 +} + +func cmp8_native(a [8]uint64, pk [4]uint64) int16 { + for i := range a { + if a[i] >= pk[0] { + return int16(i) + } + } + return 8 +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/simd/search.go b/vendor/github.com/dgraph-io/ristretto/z/simd/search.go new file mode 100644 index 00000000000..b1e639225ac --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/simd/search.go @@ -0,0 +1,51 @@ +// +build !amd64 + +/* + * Copyright 2020 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package simd + +// Search uses the Clever search to find the correct key. +func Search(xs []uint64, k uint64) int16 { + if len(xs) < 8 || (len(xs) % 8 != 0) { + return Naive(xs, k) + } + var twos, pk [4]uint64 + pk[0] = k + pk[1] = k + pk[2] = k + pk[3] = k + for i := 0; i < len(xs); i += 8 { + twos[0] = xs[i] + twos[1] = xs[i+2] + twos[2] = xs[i+4] + twos[3] = xs[i+6] + if twos[0] >= pk[0] { + return int16(i / 2) + } + if twos[1] >= pk[1] { + return int16((i + 2) / 2) + } + if twos[2] >= pk[2] { + return int16((i + 4) / 2) + } + if twos[3] >= pk[3] { + return int16((i + 6) / 2) + } + + } + return int16(len(xs) / 2) +} diff --git a/vendor/github.com/dgraph-io/ristretto/z/simd/search_amd64.s b/vendor/github.com/dgraph-io/ristretto/z/simd/search_amd64.s new file mode 100644 index 00000000000..150c846647e --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/simd/search_amd64.s @@ -0,0 +1,60 @@ +// Code generated by command: go run asm2.go -out search_amd64.s -stubs stub_search_amd64.go. DO NOT EDIT. + +#include "textflag.h" + +// func Search(xs []uint64, k uint64) int16 +TEXT ·Search(SB), NOSPLIT, $0-34 + MOVQ xs_base+0(FP), AX + MOVQ xs_len+8(FP), CX + MOVQ k+24(FP), DX + + // Save n + MOVQ CX, BX + + // Initialize idx register to zero. + XORL BP, BP + +loop: + // Unroll1 + CMPQ (AX)(BP*8), DX + JAE Found + + // Unroll2 + CMPQ 16(AX)(BP*8), DX + JAE Found2 + + // Unroll3 + CMPQ 32(AX)(BP*8), DX + JAE Found3 + + // Unroll4 + CMPQ 48(AX)(BP*8), DX + JAE Found4 + + // plus8 + ADDQ $0x08, BP + CMPQ BP, CX + JB loop + JMP NotFound + +Found2: + ADDL $0x02, BP + JMP Found + +Found3: + ADDL $0x04, BP + JMP Found + +Found4: + ADDL $0x06, BP + +Found: + MOVL BP, BX + +NotFound: + MOVL BX, BP + SHRL $0x1f, BP + ADDL BX, BP + SHRL $0x01, BP + MOVL BP, ret+32(FP) + RET diff --git a/vendor/github.com/dgraph-io/ristretto/z/simd/stub_search_amd64.go b/vendor/github.com/dgraph-io/ristretto/z/simd/stub_search_amd64.go new file mode 100644 index 00000000000..0821d38a772 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/simd/stub_search_amd64.go @@ -0,0 +1,6 @@ +// Code generated by command: go run asm2.go -out search_amd64.s -stubs stub_search_amd64.go. DO NOT EDIT. + +package simd + +// Search finds the first idx for which xs[idx] >= k in xs. +func Search(xs []uint64, k uint64) int16 diff --git a/vendor/github.com/dgraph-io/ristretto/z/z.go b/vendor/github.com/dgraph-io/ristretto/z/z.go new file mode 100644 index 00000000000..97455586a19 --- /dev/null +++ b/vendor/github.com/dgraph-io/ristretto/z/z.go @@ -0,0 +1,151 @@ +/* + * Copyright 2019 Dgraph Labs, Inc. and Contributors + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package z + +import ( + "context" + "sync" + + "github.com/cespare/xxhash/v2" +) + +// TODO: Figure out a way to re-use memhash for the second uint64 hash, we +// already know that appending bytes isn't reliable for generating a +// second hash (see Ristretto PR #88). +// +// We also know that while the Go runtime has a runtime memhash128 +// function, it's not possible to use it to generate [2]uint64 or +// anything resembling a 128bit hash, even though that's exactly what +// we need in this situation. +func KeyToHash(key interface{}) (uint64, uint64) { + if key == nil { + return 0, 0 + } + switch k := key.(type) { + case uint64: + return k, 0 + case string: + return MemHashString(k), xxhash.Sum64String(k) + case []byte: + return MemHash(k), xxhash.Sum64(k) + case byte: + return uint64(k), 0 + case int: + return uint64(k), 0 + case int32: + return uint64(k), 0 + case uint32: + return uint64(k), 0 + case int64: + return uint64(k), 0 + default: + panic("Key type not supported") + } +} + +var ( + dummyCloserChan <-chan struct{} + tmpDir string +) + +// Closer holds the two things we need to close a goroutine and wait for it to +// finish: a chan to tell the goroutine to shut down, and a WaitGroup with +// which to wait for it to finish shutting down. +type Closer struct { + waiting sync.WaitGroup + + ctx context.Context + cancel context.CancelFunc +} + +// SetTmpDir sets the temporary directory for the temporary buffers. +func SetTmpDir(dir string) { + tmpDir = dir +} + +// NewCloser constructs a new Closer, with an initial count on the WaitGroup. +func NewCloser(initial int) *Closer { + ret := &Closer{} + ret.ctx, ret.cancel = context.WithCancel(context.Background()) + ret.waiting.Add(initial) + return ret +} + +// AddRunning Add()'s delta to the WaitGroup. +func (lc *Closer) AddRunning(delta int) { + lc.waiting.Add(delta) +} + +// Ctx can be used to get a context, which would automatically get cancelled when Signal is called. +func (lc *Closer) Ctx() context.Context { + if lc == nil { + return context.Background() + } + return lc.ctx +} + +// Signal signals the HasBeenClosed signal. +func (lc *Closer) Signal() { + // Todo(ibrahim): Change Signal to return error on next badger breaking change. + lc.cancel() +} + +// HasBeenClosed gets signaled when Signal() is called. +func (lc *Closer) HasBeenClosed() <-chan struct{} { + if lc == nil { + return dummyCloserChan + } + return lc.ctx.Done() +} + +// Done calls Done() on the WaitGroup. +func (lc *Closer) Done() { + if lc == nil { + return + } + lc.waiting.Done() +} + +// Wait waits on the WaitGroup. (It waits for NewCloser's initial value, AddRunning, and Done +// calls to balance out.) +func (lc *Closer) Wait() { + lc.waiting.Wait() +} + +// SignalAndWait calls Signal(), then Wait(). +func (lc *Closer) SignalAndWait() { + lc.Signal() + lc.Wait() +} + +// ZeroOut zeroes out all the bytes in the range [start, end). +func ZeroOut(dst []byte, start, end int) { + if start < 0 || start >= len(dst) { + return // BAD + } + if end >= len(dst) { + end = len(dst) + } + if end-start <= 0 { + return + } + Memclr(dst[start:end]) + // b := dst[start:end] + // for i := range b { + // b[i] = 0x0 + // } +} diff --git a/vendor/github.com/golang/glog/LICENSE b/vendor/github.com/golang/glog/LICENSE new file mode 100644 index 00000000000..37ec93a14fd --- /dev/null +++ b/vendor/github.com/golang/glog/LICENSE @@ -0,0 +1,191 @@ +Apache License +Version 2.0, January 2004 +http://www.apache.org/licenses/ + +TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + +1. Definitions. + +"License" shall mean the terms and conditions for use, reproduction, and +distribution as defined by Sections 1 through 9 of this document. + +"Licensor" shall mean the copyright owner or entity authorized by the copyright +owner that is granting the License. + +"Legal Entity" shall mean the union of the acting entity and all other entities +that control, are controlled by, or are under common control with that entity. +For the purposes of this definition, "control" means (i) the power, direct or +indirect, to cause the direction or management of such entity, whether by +contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the +outstanding shares, or (iii) beneficial ownership of such entity. + +"You" (or "Your") shall mean an individual or Legal Entity exercising +permissions granted by this License. + +"Source" form shall mean the preferred form for making modifications, including +but not limited to software source code, documentation source, and configuration +files. + +"Object" form shall mean any form resulting from mechanical transformation or +translation of a Source form, including but not limited to compiled object code, +generated documentation, and conversions to other media types. + +"Work" shall mean the work of authorship, whether in Source or Object form, made +available under the License, as indicated by a copyright notice that is included +in or attached to the work (an example is provided in the Appendix below). + +"Derivative Works" shall mean any work, whether in Source or Object form, that +is based on (or derived from) the Work and for which the editorial revisions, +annotations, elaborations, or other modifications represent, as a whole, an +original work of authorship. For the purposes of this License, Derivative Works +shall not include works that remain separable from, or merely link (or bind by +name) to the interfaces of, the Work and Derivative Works thereof. + +"Contribution" shall mean any work of authorship, including the original version +of the Work and any modifications or additions to that Work or Derivative Works +thereof, that is intentionally submitted to Licensor for inclusion in the Work +by the copyright owner or by an individual or Legal Entity authorized to submit +on behalf of the copyright owner. For the purposes of this definition, +"submitted" means any form of electronic, verbal, or written communication sent +to the Licensor or its representatives, including but not limited to +communication on electronic mailing lists, source code control systems, and +issue tracking systems that are managed by, or on behalf of, the Licensor for +the purpose of discussing and improving the Work, but excluding communication +that is conspicuously marked or otherwise designated in writing by the copyright +owner as "Not a Contribution." + +"Contributor" shall mean Licensor and any individual or Legal Entity on behalf +of whom a Contribution has been received by Licensor and subsequently +incorporated within the Work. + +2. Grant of Copyright License. + +Subject to the terms and conditions of this License, each Contributor hereby +grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, +irrevocable copyright license to reproduce, prepare Derivative Works of, +publicly display, publicly perform, sublicense, and distribute the Work and such +Derivative Works in Source or Object form. + +3. Grant of Patent License. + +Subject to the terms and conditions of this License, each Contributor hereby +grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, +irrevocable (except as stated in this section) patent license to make, have +made, use, offer to sell, sell, import, and otherwise transfer the Work, where +such license applies only to those patent claims licensable by such Contributor +that are necessarily infringed by their Contribution(s) alone or by combination +of their Contribution(s) with the Work to which such Contribution(s) was +submitted. If You institute patent litigation against any entity (including a +cross-claim or counterclaim in a lawsuit) alleging that the Work or a +Contribution incorporated within the Work constitutes direct or contributory +patent infringement, then any patent licenses granted to You under this License +for that Work shall terminate as of the date such litigation is filed. + +4. Redistribution. + +You may reproduce and distribute copies of the Work or Derivative Works thereof +in any medium, with or without modifications, and in Source or Object form, +provided that You meet the following conditions: + +You must give any other recipients of the Work or Derivative Works a copy of +this License; and +You must cause any modified files to carry prominent notices stating that You +changed the files; and +You must retain, in the Source form of any Derivative Works that You distribute, +all copyright, patent, trademark, and attribution notices from the Source form +of the Work, excluding those notices that do not pertain to any part of the +Derivative Works; and +If the Work includes a "NOTICE" text file as part of its distribution, then any +Derivative Works that You distribute must include a readable copy of the +attribution notices contained within such NOTICE file, excluding those notices +that do not pertain to any part of the Derivative Works, in at least one of the +following places: within a NOTICE text file distributed as part of the +Derivative Works; within the Source form or documentation, if provided along +with the Derivative Works; or, within a display generated by the Derivative +Works, if and wherever such third-party notices normally appear. The contents of +the NOTICE file are for informational purposes only and do not modify the +License. You may add Your own attribution notices within Derivative Works that +You distribute, alongside or as an addendum to the NOTICE text from the Work, +provided that such additional attribution notices cannot be construed as +modifying the License. +You may add Your own copyright statement to Your modifications and may provide +additional or different license terms and conditions for use, reproduction, or +distribution of Your modifications, or for any such Derivative Works as a whole, +provided Your use, reproduction, and distribution of the Work otherwise complies +with the conditions stated in this License. + +5. Submission of Contributions. + +Unless You explicitly state otherwise, any Contribution intentionally submitted +for inclusion in the Work by You to the Licensor shall be under the terms and +conditions of this License, without any additional terms or conditions. +Notwithstanding the above, nothing herein shall supersede or modify the terms of +any separate license agreement you may have executed with Licensor regarding +such Contributions. + +6. Trademarks. + +This License does not grant permission to use the trade names, trademarks, +service marks, or product names of the Licensor, except as required for +reasonable and customary use in describing the origin of the Work and +reproducing the content of the NOTICE file. + +7. Disclaimer of Warranty. + +Unless required by applicable law or agreed to in writing, Licensor provides the +Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, +including, without limitation, any warranties or conditions of TITLE, +NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are +solely responsible for determining the appropriateness of using or +redistributing the Work and assume any risks associated with Your exercise of +permissions under this License. + +8. Limitation of Liability. + +In no event and under no legal theory, whether in tort (including negligence), +contract, or otherwise, unless required by applicable law (such as deliberate +and grossly negligent acts) or agreed to in writing, shall any Contributor be +liable to You for damages, including any direct, indirect, special, incidental, +or consequential damages of any character arising as a result of this License or +out of the use or inability to use the Work (including but not limited to +damages for loss of goodwill, work stoppage, computer failure or malfunction, or +any and all other commercial damages or losses), even if such Contributor has +been advised of the possibility of such damages. + +9. Accepting Warranty or Additional Liability. + +While redistributing the Work or Derivative Works thereof, You may choose to +offer, and charge a fee for, acceptance of support, warranty, indemnity, or +other liability obligations and/or rights consistent with this License. However, +in accepting such obligations, You may act only on Your own behalf and on Your +sole responsibility, not on behalf of any other Contributor, and only if You +agree to indemnify, defend, and hold each Contributor harmless for any liability +incurred by, or claims asserted against, such Contributor by reason of your +accepting any such warranty or additional liability. + +END OF TERMS AND CONDITIONS + +APPENDIX: How to apply the Apache License to your work + +To apply the Apache License to your work, attach the following boilerplate +notice, with the fields enclosed by brackets "[]" replaced with your own +identifying information. (Don't include the brackets!) The text should be +enclosed in the appropriate comment syntax for the file format. We also +recommend that a file or class name and description of purpose be included on +the same "printed page" as the copyright notice for easier identification within +third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/vendor/github.com/golang/glog/README.md b/vendor/github.com/golang/glog/README.md new file mode 100644 index 00000000000..a4f73883b2d --- /dev/null +++ b/vendor/github.com/golang/glog/README.md @@ -0,0 +1,36 @@ +# glog + +[![PkgGoDev](https://pkg.go.dev/badge/github.com/golang/glog)](https://pkg.go.dev/github.com/golang/glog) + +Leveled execution logs for Go. + +This is an efficient pure Go implementation of leveled logs in the +manner of the open source C++ package [_glog_](https://github.com/google/glog). + +By binding methods to booleans it is possible to use the log package without paying the expense of evaluating the arguments to the log. Through the `-vmodule` flag, the package also provides fine-grained +control over logging at the file level. + +The comment from `glog.go` introduces the ideas: + +Package _glog_ implements logging analogous to the Google-internal C++ INFO/ERROR/V setup. It provides the functions Info, Warning, Error, Fatal, plus formatting variants such as Infof. It also provides V-style loggingcontrolled by the `-v` and `-vmodule=file=2` flags. + +Basic examples: + +```go +glog.Info("Prepare to repel boarders") + +glog.Fatalf("Initialization failed: %s", err) +``` + +See the documentation for the V function for an explanation of these examples: + +```go +if glog.V(2) { + glog.Info("Starting transaction...") +} +glog.V(2).Infoln("Processed", nItems, "elements") +``` + +The repository contains an open source version of the log package used inside Google. The master copy of the source lives inside Google, not here. The code in this repo is for export only and is not itself under development. Feature requests will be ignored. + +Send bug reports to golang-nuts@googlegroups.com. diff --git a/vendor/github.com/golang/glog/glog.go b/vendor/github.com/golang/glog/glog.go new file mode 100644 index 00000000000..dd0ed101563 --- /dev/null +++ b/vendor/github.com/golang/glog/glog.go @@ -0,0 +1,592 @@ +// Go support for leveled logs, analogous to https://github.com/google/glog. +// +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package glog implements logging analogous to the Google-internal C++ INFO/ERROR/V setup. +// It provides functions Info, Warning, Error, Fatal, plus formatting variants such as +// Infof. It also provides V-style logging controlled by the -v and -vmodule=file=2 flags. +// +// Basic examples: +// +// glog.Info("Prepare to repel boarders") +// +// glog.Fatalf("Initialization failed: %s", err) +// +// See the documentation for the V function for an explanation of these examples: +// +// if glog.V(2) { +// glog.Info("Starting transaction...") +// } +// +// glog.V(2).Infoln("Processed", nItems, "elements") +// +// Log output is buffered and written periodically using Flush. Programs +// should call Flush before exiting to guarantee all log output is written. +// +// By default, all log statements write to files in a temporary directory. +// This package provides several flags that modify this behavior. +// As a result, flag.Parse must be called before any logging is done. +// +// -logtostderr=false +// Logs are written to standard error instead of to files. +// -alsologtostderr=false +// Logs are written to standard error as well as to files. +// -stderrthreshold=ERROR +// Log events at or above this severity are logged to standard +// error as well as to files. +// -log_dir="" +// Log files will be written to this directory instead of the +// default temporary directory. +// +// Other flags provide aids to debugging. +// +// -log_backtrace_at="" +// A comma-separated list of file and line numbers holding a logging +// statement, such as +// -log_backtrace_at=gopherflakes.go:234 +// A stack trace will be written to the Info log whenever execution +// hits one of these statements. (Unlike with -vmodule, the ".go" +// must bepresent.) +// -v=0 +// Enable V-leveled logging at the specified level. +// -vmodule="" +// The syntax of the argument is a comma-separated list of pattern=N, +// where pattern is a literal file name (minus the ".go" suffix) or +// "glob" pattern and N is a V level. For instance, +// -vmodule=gopher*=3 +// sets the V level to 3 in all Go files whose names begin with "gopher", +// and +// -vmodule=/path/to/glog/glog_test=1 +// sets the V level to 1 in the Go file /path/to/glog/glog_test.go. +// If a glob pattern contains a slash, it is matched against the full path, +// and the file name. Otherwise, the pattern is +// matched only against the file's basename. When both -vmodule and -v +// are specified, the -vmodule values take precedence for the specified +// modules. +package glog + +// This file contains the parts of the log package that are shared among all +// implementations (file, envelope, and appengine). + +import ( + "bytes" + "errors" + "fmt" + stdLog "log" + "os" + "reflect" + "runtime" + "runtime/pprof" + "strconv" + "sync" + "sync/atomic" + "time" + + "github.com/golang/glog/internal/logsink" + "github.com/golang/glog/internal/stackdump" +) + +var timeNow = time.Now // Stubbed out for testing. + +// MaxSize is the maximum size of a log file in bytes. +var MaxSize uint64 = 1024 * 1024 * 1800 + +// ErrNoLog is the error we return if no log file has yet been created +// for the specified log type. +var ErrNoLog = errors.New("log file not yet created") + +// OutputStats tracks the number of output lines and bytes written. +type OutputStats struct { + lines int64 + bytes int64 +} + +// Lines returns the number of lines written. +func (s *OutputStats) Lines() int64 { + return atomic.LoadInt64(&s.lines) +} + +// Bytes returns the number of bytes written. +func (s *OutputStats) Bytes() int64 { + return atomic.LoadInt64(&s.bytes) +} + +// Stats tracks the number of lines of output and number of bytes +// per severity level. Values must be read with atomic.LoadInt64. +var Stats struct { + Info, Warning, Error OutputStats +} + +var severityStats = [...]*OutputStats{ + logsink.Info: &Stats.Info, + logsink.Warning: &Stats.Warning, + logsink.Error: &Stats.Error, + logsink.Fatal: nil, +} + +// Level specifies a level of verbosity for V logs. The -v flag is of type +// Level and should be modified only through the flag.Value interface. +type Level int32 + +var metaPool sync.Pool // Pool of *logsink.Meta. + +// metaPoolGet returns a *logsink.Meta from metaPool as both an interface and a +// pointer, allocating a new one if necessary. (Returning the interface value +// directly avoids an allocation if there was an existing pointer in the pool.) +func metaPoolGet() (any, *logsink.Meta) { + if metai := metaPool.Get(); metai != nil { + return metai, metai.(*logsink.Meta) + } + meta := new(logsink.Meta) + return meta, meta +} + +type stack bool + +const ( + noStack = stack(false) + withStack = stack(true) +) + +func appendBacktrace(depth int, format string, args []any) (string, []any) { + // Capture a backtrace as a stackdump.Stack (both text and PC slice). + // Structured log sinks can extract the backtrace in whichever format they + // prefer (PCs or text), and Text sinks will include it as just another part + // of the log message. + // + // Use depth instead of depth+1 so that the backtrace always includes the + // log function itself - otherwise the reason for the trace appearing in the + // log may not be obvious to the reader. + dump := stackdump.Caller(depth) + + // Add an arg and an entry in the format string for the stack dump. + // + // Copy the "args" slice to avoid a rare but serious aliasing bug + // (corrupting the caller's slice if they passed it to a non-Fatal call + // using "..."). + format = format + "\n\n%v\n" + args = append(append([]any(nil), args...), dump) + + return format, args +} + +// logf writes a log message for a log function call (or log function wrapper) +// at the given depth in the current goroutine's stack. +func logf(depth int, severity logsink.Severity, verbose bool, stack stack, format string, args ...any) { + now := timeNow() + _, file, line, ok := runtime.Caller(depth + 1) + if !ok { + file = "???" + line = 1 + } + + if stack == withStack || backtraceAt(file, line) { + format, args = appendBacktrace(depth+1, format, args) + } + + metai, meta := metaPoolGet() + *meta = logsink.Meta{ + Time: now, + File: file, + Line: line, + Depth: depth + 1, + Severity: severity, + Verbose: verbose, + Thread: int64(pid), + } + sinkf(meta, format, args...) + metaPool.Put(metai) +} + +func sinkf(meta *logsink.Meta, format string, args ...any) { + meta.Depth++ + n, err := logsink.Printf(meta, format, args...) + if stats := severityStats[meta.Severity]; stats != nil { + atomic.AddInt64(&stats.lines, 1) + atomic.AddInt64(&stats.bytes, int64(n)) + } + + if err != nil { + logsink.Printf(meta, "glog: exiting because of error: %s", err) + sinks.file.Flush() + os.Exit(2) + } +} + +// CopyStandardLogTo arranges for messages written to the Go "log" package's +// default logs to also appear in the Google logs for the named and lower +// severities. Subsequent changes to the standard log's default output location +// or format may break this behavior. +// +// Valid names are "INFO", "WARNING", "ERROR", and "FATAL". If the name is not +// recognized, CopyStandardLogTo panics. +func CopyStandardLogTo(name string) { + sev, err := logsink.ParseSeverity(name) + if err != nil { + panic(fmt.Sprintf("log.CopyStandardLogTo(%q): %v", name, err)) + } + // Set a log format that captures the user's file and line: + // d.go:23: message + stdLog.SetFlags(stdLog.Lshortfile) + stdLog.SetOutput(logBridge(sev)) +} + +// NewStandardLogger returns a Logger that writes to the Google logs for the +// named and lower severities. +// +// Valid names are "INFO", "WARNING", "ERROR", and "FATAL". If the name is not +// recognized, NewStandardLogger panics. +func NewStandardLogger(name string) *stdLog.Logger { + sev, err := logsink.ParseSeverity(name) + if err != nil { + panic(fmt.Sprintf("log.NewStandardLogger(%q): %v", name, err)) + } + return stdLog.New(logBridge(sev), "", stdLog.Lshortfile) +} + +// logBridge provides the Write method that enables CopyStandardLogTo to connect +// Go's standard logs to the logs provided by this package. +type logBridge logsink.Severity + +// Write parses the standard logging line and passes its components to the +// logger for severity(lb). +func (lb logBridge) Write(b []byte) (n int, err error) { + var ( + file = "???" + line = 1 + text string + ) + // Split "d.go:23: message" into "d.go", "23", and "message". + if parts := bytes.SplitN(b, []byte{':'}, 3); len(parts) != 3 || len(parts[0]) < 1 || len(parts[2]) < 1 { + text = fmt.Sprintf("bad log format: %s", b) + } else { + file = string(parts[0]) + text = string(parts[2][1:]) // skip leading space + line, err = strconv.Atoi(string(parts[1])) + if err != nil { + text = fmt.Sprintf("bad line number: %s", b) + line = 1 + } + } + + // The depth below hard-codes details of how stdlog gets here. The alternative would be to walk + // up the stack looking for src/log/log.go but that seems like it would be + // unfortunately slow. + const stdLogDepth = 4 + + metai, meta := metaPoolGet() + *meta = logsink.Meta{ + Time: timeNow(), + File: file, + Line: line, + Depth: stdLogDepth, + Severity: logsink.Severity(lb), + Thread: int64(pid), + } + + format := "%s" + args := []any{text} + if backtraceAt(file, line) { + format, args = appendBacktrace(meta.Depth, format, args) + } + + sinkf(meta, format, args...) + metaPool.Put(metai) + + return len(b), nil +} + +// defaultFormat returns a fmt.Printf format specifier that formats its +// arguments as if they were passed to fmt.Print. +func defaultFormat(args []any) string { + n := len(args) + switch n { + case 0: + return "" + case 1: + return "%v" + } + + b := make([]byte, 0, n*3-1) + wasString := true // Suppress leading space. + for _, arg := range args { + isString := arg != nil && reflect.TypeOf(arg).Kind() == reflect.String + if wasString || isString { + b = append(b, "%v"...) + } else { + b = append(b, " %v"...) + } + wasString = isString + } + return string(b) +} + +// lnFormat returns a fmt.Printf format specifier that formats its arguments +// as if they were passed to fmt.Println. +func lnFormat(args []any) string { + if len(args) == 0 { + return "\n" + } + + b := make([]byte, 0, len(args)*3) + for range args { + b = append(b, "%v "...) + } + b[len(b)-1] = '\n' // Replace the last space with a newline. + return string(b) +} + +// Verbose is a boolean type that implements Infof (like Printf) etc. +// See the documentation of V for more information. +type Verbose bool + +// V reports whether verbosity at the call site is at least the requested level. +// The returned value is a boolean of type Verbose, which implements Info, Infoln +// and Infof. These methods will write to the Info log if called. +// Thus, one may write either +// +// if glog.V(2) { glog.Info("log this") } +// +// or +// +// glog.V(2).Info("log this") +// +// The second form is shorter but the first is cheaper if logging is off because it does +// not evaluate its arguments. +// +// Whether an individual call to V generates a log record depends on the setting of +// the -v and --vmodule flags; both are off by default. If the level in the call to +// V is at most the value of -v, or of -vmodule for the source file containing the +// call, the V call will log. +func V(level Level) Verbose { + return VDepth(1, level) +} + +// VDepth acts as V but uses depth to determine which call frame to check vmodule for. +// VDepth(0, level) is the same as V(level). +func VDepth(depth int, level Level) Verbose { + return Verbose(verboseEnabled(depth+1, level)) +} + +// Info is equivalent to the global Info function, guarded by the value of v. +// See the documentation of V for usage. +func (v Verbose) Info(args ...any) { + v.InfoDepth(1, args...) +} + +// InfoDepth is equivalent to the global InfoDepth function, guarded by the value of v. +// See the documentation of V for usage. +func (v Verbose) InfoDepth(depth int, args ...any) { + if v { + logf(depth+1, logsink.Info, true, noStack, defaultFormat(args), args...) + } +} + +// InfoDepthf is equivalent to the global InfoDepthf function, guarded by the value of v. +// See the documentation of V for usage. +func (v Verbose) InfoDepthf(depth int, format string, args ...any) { + if v { + logf(depth+1, logsink.Info, true, noStack, format, args...) + } +} + +// Infoln is equivalent to the global Infoln function, guarded by the value of v. +// See the documentation of V for usage. +func (v Verbose) Infoln(args ...any) { + if v { + logf(1, logsink.Info, true, noStack, lnFormat(args), args...) + } +} + +// Infof is equivalent to the global Infof function, guarded by the value of v. +// See the documentation of V for usage. +func (v Verbose) Infof(format string, args ...any) { + if v { + logf(1, logsink.Info, true, noStack, format, args...) + } +} + +// Info logs to the INFO log. +// Arguments are handled in the manner of fmt.Print; a newline is appended if missing. +func Info(args ...any) { + InfoDepth(1, args...) +} + +// InfoDepth calls Info from a different depth in the call stack. +// This enables a callee to emit logs that use the callsite information of its caller +// or any other callers in the stack. When depth == 0, the original callee's line +// information is emitted. When depth > 0, depth frames are skipped in the call stack +// and the final frame is treated like the original callee to Info. +func InfoDepth(depth int, args ...any) { + logf(depth+1, logsink.Info, false, noStack, defaultFormat(args), args...) +} + +// InfoDepthf acts as InfoDepth but with format string. +func InfoDepthf(depth int, format string, args ...any) { + logf(depth+1, logsink.Info, false, noStack, format, args...) +} + +// Infoln logs to the INFO log. +// Arguments are handled in the manner of fmt.Println; a newline is appended if missing. +func Infoln(args ...any) { + logf(1, logsink.Info, false, noStack, lnFormat(args), args...) +} + +// Infof logs to the INFO log. +// Arguments are handled in the manner of fmt.Printf; a newline is appended if missing. +func Infof(format string, args ...any) { + logf(1, logsink.Info, false, noStack, format, args...) +} + +// Warning logs to the WARNING and INFO logs. +// Arguments are handled in the manner of fmt.Print; a newline is appended if missing. +func Warning(args ...any) { + WarningDepth(1, args...) +} + +// WarningDepth acts as Warning but uses depth to determine which call frame to log. +// WarningDepth(0, "msg") is the same as Warning("msg"). +func WarningDepth(depth int, args ...any) { + logf(depth+1, logsink.Warning, false, noStack, defaultFormat(args), args...) +} + +// WarningDepthf acts as Warningf but uses depth to determine which call frame to log. +// WarningDepthf(0, "msg") is the same as Warningf("msg"). +func WarningDepthf(depth int, format string, args ...any) { + logf(depth+1, logsink.Warning, false, noStack, format, args...) +} + +// Warningln logs to the WARNING and INFO logs. +// Arguments are handled in the manner of fmt.Println; a newline is appended if missing. +func Warningln(args ...any) { + logf(1, logsink.Warning, false, noStack, lnFormat(args), args...) +} + +// Warningf logs to the WARNING and INFO logs. +// Arguments are handled in the manner of fmt.Printf; a newline is appended if missing. +func Warningf(format string, args ...any) { + logf(1, logsink.Warning, false, noStack, format, args...) +} + +// Error logs to the ERROR, WARNING, and INFO logs. +// Arguments are handled in the manner of fmt.Print; a newline is appended if missing. +func Error(args ...any) { + ErrorDepth(1, args...) +} + +// ErrorDepth acts as Error but uses depth to determine which call frame to log. +// ErrorDepth(0, "msg") is the same as Error("msg"). +func ErrorDepth(depth int, args ...any) { + logf(depth+1, logsink.Error, false, noStack, defaultFormat(args), args...) +} + +// ErrorDepthf acts as Errorf but uses depth to determine which call frame to log. +// ErrorDepthf(0, "msg") is the same as Errorf("msg"). +func ErrorDepthf(depth int, format string, args ...any) { + logf(depth+1, logsink.Error, false, noStack, format, args...) +} + +// Errorln logs to the ERROR, WARNING, and INFO logs. +// Arguments are handled in the manner of fmt.Println; a newline is appended if missing. +func Errorln(args ...any) { + logf(1, logsink.Error, false, noStack, lnFormat(args), args...) +} + +// Errorf logs to the ERROR, WARNING, and INFO logs. +// Arguments are handled in the manner of fmt.Printf; a newline is appended if missing. +func Errorf(format string, args ...any) { + logf(1, logsink.Error, false, noStack, format, args...) +} + +func fatalf(depth int, format string, args ...any) { + logf(depth+1, logsink.Fatal, false, withStack, format, args...) + sinks.file.Flush() + + err := abortProcess() // Should not return. + + // Failed to abort the process using signals. Dump a stack trace and exit. + Errorf("abortProcess returned unexpectedly: %v", err) + sinks.file.Flush() + pprof.Lookup("goroutine").WriteTo(os.Stderr, 1) + os.Exit(2) // Exit with the same code as the default SIGABRT handler. +} + +// Fatal logs to the FATAL, ERROR, WARNING, and INFO logs, +// including a stack trace of all running goroutines, then calls os.Exit(2). +// Arguments are handled in the manner of fmt.Print; a newline is appended if missing. +func Fatal(args ...any) { + FatalDepth(1, args...) +} + +// FatalDepth acts as Fatal but uses depth to determine which call frame to log. +// FatalDepth(0, "msg") is the same as Fatal("msg"). +func FatalDepth(depth int, args ...any) { + fatalf(depth+1, defaultFormat(args), args...) +} + +// FatalDepthf acts as Fatalf but uses depth to determine which call frame to log. +// FatalDepthf(0, "msg") is the same as Fatalf("msg"). +func FatalDepthf(depth int, format string, args ...any) { + fatalf(depth+1, format, args...) +} + +// Fatalln logs to the FATAL, ERROR, WARNING, and INFO logs, +// including a stack trace of all running goroutines, then calls os.Exit(2). +// Arguments are handled in the manner of fmt.Println; a newline is appended if missing. +func Fatalln(args ...any) { + fatalf(1, lnFormat(args), args...) +} + +// Fatalf logs to the FATAL, ERROR, WARNING, and INFO logs, +// including a stack trace of all running goroutines, then calls os.Exit(2). +// Arguments are handled in the manner of fmt.Printf; a newline is appended if missing. +func Fatalf(format string, args ...any) { + fatalf(1, format, args...) +} + +func exitf(depth int, format string, args ...any) { + logf(depth+1, logsink.Fatal, false, noStack, format, args...) + sinks.file.Flush() + os.Exit(1) +} + +// Exit logs to the FATAL, ERROR, WARNING, and INFO logs, then calls os.Exit(1). +// Arguments are handled in the manner of fmt.Print; a newline is appended if missing. +func Exit(args ...any) { + ExitDepth(1, args...) +} + +// ExitDepth acts as Exit but uses depth to determine which call frame to log. +// ExitDepth(0, "msg") is the same as Exit("msg"). +func ExitDepth(depth int, args ...any) { + exitf(depth+1, defaultFormat(args), args...) +} + +// ExitDepthf acts as Exitf but uses depth to determine which call frame to log. +// ExitDepthf(0, "msg") is the same as Exitf("msg"). +func ExitDepthf(depth int, format string, args ...any) { + exitf(depth+1, format, args...) +} + +// Exitln logs to the FATAL, ERROR, WARNING, and INFO logs, then calls os.Exit(1). +func Exitln(args ...any) { + exitf(1, lnFormat(args), args...) +} + +// Exitf logs to the FATAL, ERROR, WARNING, and INFO logs, then calls os.Exit(1). +// Arguments are handled in the manner of fmt.Printf; a newline is appended if missing. +func Exitf(format string, args ...any) { + exitf(1, format, args...) +} diff --git a/vendor/github.com/golang/glog/glog_file.go b/vendor/github.com/golang/glog/glog_file.go new file mode 100644 index 00000000000..e7d125c5ae4 --- /dev/null +++ b/vendor/github.com/golang/glog/glog_file.go @@ -0,0 +1,413 @@ +// Go support for leveled logs, analogous to https://github.com/google/glog. +// +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// File I/O for logs. + +package glog + +import ( + "bufio" + "bytes" + "errors" + "flag" + "fmt" + "io" + "os" + "os/user" + "path/filepath" + "runtime" + "strings" + "sync" + "time" + + "github.com/golang/glog/internal/logsink" +) + +// logDirs lists the candidate directories for new log files. +var logDirs []string + +var ( + // If non-empty, overrides the choice of directory in which to write logs. + // See createLogDirs for the full list of possible destinations. + logDir = flag.String("log_dir", "", "If non-empty, write log files in this directory") + logLink = flag.String("log_link", "", "If non-empty, add symbolic links in this directory to the log files") + logBufLevel = flag.Int("logbuflevel", int(logsink.Info), "Buffer log messages logged at this level or lower"+ + " (-1 means don't buffer; 0 means buffer INFO only; ...). Has limited applicability on non-prod platforms.") +) + +func createLogDirs() { + if *logDir != "" { + logDirs = append(logDirs, *logDir) + } + logDirs = append(logDirs, os.TempDir()) +} + +var ( + pid = os.Getpid() + program = filepath.Base(os.Args[0]) + host = "unknownhost" + userName = "unknownuser" +) + +func init() { + h, err := os.Hostname() + if err == nil { + host = shortHostname(h) + } + + current, err := user.Current() + if err == nil { + userName = current.Username + } + // Sanitize userName since it is used to construct file paths. + userName = strings.Map(func(r rune) rune { + switch { + case r >= 'a' && r <= 'z': + case r >= 'A' && r <= 'Z': + case r >= '0' && r <= '9': + default: + return '_' + } + return r + }, userName) +} + +// shortHostname returns its argument, truncating at the first period. +// For instance, given "www.google.com" it returns "www". +func shortHostname(hostname string) string { + if i := strings.Index(hostname, "."); i >= 0 { + return hostname[:i] + } + return hostname +} + +// logName returns a new log file name containing tag, with start time t, and +// the name for the symlink for tag. +func logName(tag string, t time.Time) (name, link string) { + name = fmt.Sprintf("%s.%s.%s.log.%s.%04d%02d%02d-%02d%02d%02d.%d", + program, + host, + userName, + tag, + t.Year(), + t.Month(), + t.Day(), + t.Hour(), + t.Minute(), + t.Second(), + pid) + return name, program + "." + tag +} + +var onceLogDirs sync.Once + +// create creates a new log file and returns the file and its filename, which +// contains tag ("INFO", "FATAL", etc.) and t. If the file is created +// successfully, create also attempts to update the symlink for that tag, ignoring +// errors. +func create(tag string, t time.Time) (f *os.File, filename string, err error) { + onceLogDirs.Do(createLogDirs) + if len(logDirs) == 0 { + return nil, "", errors.New("log: no log dirs") + } + name, link := logName(tag, t) + var lastErr error + for _, dir := range logDirs { + fname := filepath.Join(dir, name) + f, err := os.Create(fname) + if err == nil { + symlink := filepath.Join(dir, link) + os.Remove(symlink) // ignore err + os.Symlink(name, symlink) // ignore err + if *logLink != "" { + lsymlink := filepath.Join(*logLink, link) + os.Remove(lsymlink) // ignore err + os.Symlink(fname, lsymlink) // ignore err + } + return f, fname, nil + } + lastErr = err + } + return nil, "", fmt.Errorf("log: cannot create log: %v", lastErr) +} + +// flushSyncWriter is the interface satisfied by logging destinations. +type flushSyncWriter interface { + Flush() error + Sync() error + io.Writer + filenames() []string +} + +var sinks struct { + stderr stderrSink + file fileSink +} + +func init() { + // Register stderr first: that way if we crash during file-writing at least + // the log will have gone somewhere. + logsink.TextSinks = append(logsink.TextSinks, &sinks.stderr, &sinks.file) + + sinks.file.flushChan = make(chan logsink.Severity, 1) + go sinks.file.flushDaemon() +} + +// stderrSink is a logsink.Text that writes log entries to stderr +// if they meet certain conditions. +type stderrSink struct { + mu sync.Mutex + w io.Writer // if nil Emit uses os.Stderr directly +} + +// Enabled implements logsink.Text.Enabled. It returns true if any of the +// various stderr flags are enabled for logs of the given severity, if the log +// message is from the standard "log" package, or if google.Init has not yet run +// (and hence file logging is not yet initialized). +func (s *stderrSink) Enabled(m *logsink.Meta) bool { + return toStderr || alsoToStderr || m.Severity >= stderrThreshold.get() +} + +// Emit implements logsink.Text.Emit. +func (s *stderrSink) Emit(m *logsink.Meta, data []byte) (n int, err error) { + s.mu.Lock() + defer s.mu.Unlock() + w := s.w + if w == nil { + w = os.Stderr + } + dn, err := w.Write(data) + n += dn + return n, err +} + +// severityWriters is an array of flushSyncWriter with a value for each +// logsink.Severity. +type severityWriters [4]flushSyncWriter + +// fileSink is a logsink.Text that prints to a set of Google log files. +type fileSink struct { + mu sync.Mutex + // file holds writer for each of the log types. + file severityWriters + flushChan chan logsink.Severity +} + +// Enabled implements logsink.Text.Enabled. It returns true if google.Init +// has run and both --disable_log_to_disk and --logtostderr are false. +func (s *fileSink) Enabled(m *logsink.Meta) bool { + return !toStderr +} + +// Emit implements logsink.Text.Emit +func (s *fileSink) Emit(m *logsink.Meta, data []byte) (n int, err error) { + s.mu.Lock() + defer s.mu.Unlock() + + if err = s.createMissingFiles(m.Severity); err != nil { + return 0, err + } + for sev := m.Severity; sev >= logsink.Info; sev-- { + if _, fErr := s.file[sev].Write(data); fErr != nil && err == nil { + err = fErr // Take the first error. + } + } + n = len(data) + if int(m.Severity) > *logBufLevel { + select { + case s.flushChan <- m.Severity: + default: + } + } + + return n, err +} + +// syncBuffer joins a bufio.Writer to its underlying file, providing access to the +// file's Sync method and providing a wrapper for the Write method that provides log +// file rotation. There are conflicting methods, so the file cannot be embedded. +// s.mu is held for all its methods. +type syncBuffer struct { + sink *fileSink + *bufio.Writer + file *os.File + names []string + sev logsink.Severity + nbytes uint64 // The number of bytes written to this file +} + +func (sb *syncBuffer) Sync() error { + return sb.file.Sync() +} + +func (sb *syncBuffer) Write(p []byte) (n int, err error) { + if sb.nbytes+uint64(len(p)) >= MaxSize { + if err := sb.rotateFile(time.Now()); err != nil { + return 0, err + } + } + n, err = sb.Writer.Write(p) + sb.nbytes += uint64(n) + return n, err +} + +func (sb *syncBuffer) filenames() []string { + return sb.names +} + +const footer = "\nCONTINUED IN NEXT FILE\n" + +// rotateFile closes the syncBuffer's file and starts a new one. +func (sb *syncBuffer) rotateFile(now time.Time) error { + var err error + pn := "" + file, name, err := create(sb.sev.String(), now) + + if sb.file != nil { + // The current log file becomes the previous log at the end of + // this block, so save its name for use in the header of the next + // file. + pn = sb.file.Name() + sb.Flush() + // If there's an existing file, write a footer with the name of + // the next file in the chain, followed by the constant string + // \nCONTINUED IN NEXT FILE\n to make continuation detection simple. + sb.file.Write([]byte("Next log: ")) + sb.file.Write([]byte(name)) + sb.file.Write([]byte(footer)) + sb.file.Close() + } + + sb.file = file + sb.names = append(sb.names, name) + sb.nbytes = 0 + if err != nil { + return err + } + + sb.Writer = bufio.NewWriterSize(sb.file, bufferSize) + + // Write header. + var buf bytes.Buffer + fmt.Fprintf(&buf, "Log file created at: %s\n", now.Format("2006/01/02 15:04:05")) + fmt.Fprintf(&buf, "Running on machine: %s\n", host) + fmt.Fprintf(&buf, "Binary: Built with %s %s for %s/%s\n", runtime.Compiler, runtime.Version(), runtime.GOOS, runtime.GOARCH) + fmt.Fprintf(&buf, "Previous log: %s\n", pn) + fmt.Fprintf(&buf, "Log line format: [IWEF]mmdd hh:mm:ss.uuuuuu threadid file:line] msg\n") + n, err := sb.file.Write(buf.Bytes()) + sb.nbytes += uint64(n) + return err +} + +// bufferSize sizes the buffer associated with each log file. It's large +// so that log records can accumulate without the logging thread blocking +// on disk I/O. The flushDaemon will block instead. +const bufferSize = 256 * 1024 + +// createMissingFiles creates all the log files for severity from infoLog up to +// upTo that have not already been created. +// s.mu is held. +func (s *fileSink) createMissingFiles(upTo logsink.Severity) error { + if s.file[upTo] != nil { + return nil + } + now := time.Now() + // Files are created in increasing severity order, so we can be assured that + // if a high severity logfile exists, then so do all of lower severity. + for sev := logsink.Info; sev <= upTo; sev++ { + if s.file[sev] != nil { + continue + } + sb := &syncBuffer{ + sink: s, + sev: sev, + } + if err := sb.rotateFile(now); err != nil { + return err + } + s.file[sev] = sb + } + return nil +} + +// flushDaemon periodically flushes the log file buffers. +func (s *fileSink) flushDaemon() { + tick := time.NewTicker(30 * time.Second) + defer tick.Stop() + for { + select { + case <-tick.C: + s.Flush() + case sev := <-s.flushChan: + s.flush(sev) + } + } +} + +// Flush flushes all pending log I/O. +func Flush() { + sinks.file.Flush() +} + +// Flush flushes all the logs and attempts to "sync" their data to disk. +func (s *fileSink) Flush() error { + return s.flush(logsink.Info) +} + +// flush flushes all logs of severity threshold or greater. +func (s *fileSink) flush(threshold logsink.Severity) error { + s.mu.Lock() + defer s.mu.Unlock() + + var firstErr error + updateErr := func(err error) { + if err != nil && firstErr == nil { + firstErr = err + } + } + + // Flush from fatal down, in case there's trouble flushing. + for sev := logsink.Fatal; sev >= threshold; sev-- { + file := s.file[sev] + if file != nil { + updateErr(file.Flush()) + updateErr(file.Sync()) + } + } + + return firstErr +} + +// Names returns the names of the log files holding the FATAL, ERROR, +// WARNING, or INFO logs. Returns ErrNoLog if the log for the given +// level doesn't exist (e.g. because no messages of that level have been +// written). This may return multiple names if the log type requested +// has rolled over. +func Names(s string) ([]string, error) { + severity, err := logsink.ParseSeverity(s) + if err != nil { + return nil, err + } + + sinks.file.mu.Lock() + defer sinks.file.mu.Unlock() + f := sinks.file.file[severity] + if f == nil { + return nil, ErrNoLog + } + + return f.filenames(), nil +} diff --git a/vendor/github.com/golang/glog/glog_file_linux.go b/vendor/github.com/golang/glog/glog_file_linux.go new file mode 100644 index 00000000000..d795092d03a --- /dev/null +++ b/vendor/github.com/golang/glog/glog_file_linux.go @@ -0,0 +1,39 @@ +// Go support for leveled logs, analogous to https://github.com/google/glog. +// +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//go:build linux + +package glog + +import ( + "errors" + "runtime" + "syscall" +) + +// abortProcess attempts to kill the current process in a way that will dump the +// currently-running goroutines someplace useful (like stderr). +// +// It does this by sending SIGABRT to the current thread. +// +// If successful, abortProcess does not return. +func abortProcess() error { + runtime.LockOSThread() + if err := syscall.Tgkill(syscall.Getpid(), syscall.Gettid(), syscall.SIGABRT); err != nil { + return err + } + return errors.New("log: killed current thread with SIGABRT, but still running") +} diff --git a/vendor/github.com/golang/glog/glog_file_other.go b/vendor/github.com/golang/glog/glog_file_other.go new file mode 100644 index 00000000000..9540f14fc08 --- /dev/null +++ b/vendor/github.com/golang/glog/glog_file_other.go @@ -0,0 +1,30 @@ +// Go support for leveled logs, analogous to https://github.com/google/glog. +// +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//go:build !(unix || windows) + +package glog + +import ( + "fmt" + "runtime" +) + +// abortProcess returns an error on platforms that presumably don't support signals. +func abortProcess() error { + return fmt.Errorf("not sending SIGABRT (%s/%s does not support signals), falling back", runtime.GOOS, runtime.GOARCH) + +} diff --git a/vendor/github.com/golang/glog/glog_file_posix.go b/vendor/github.com/golang/glog/glog_file_posix.go new file mode 100644 index 00000000000..c27c7c0e4ab --- /dev/null +++ b/vendor/github.com/golang/glog/glog_file_posix.go @@ -0,0 +1,53 @@ +// Go support for leveled logs, analogous to https://github.com/google/glog. +// +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//go:build (unix || windows) && !linux + +package glog + +import ( + "os" + "syscall" + "time" +) + +// abortProcess attempts to kill the current process in a way that will dump the +// currently-running goroutines someplace useful (like stderr). +// +// It does this by sending SIGABRT to the current process. Unfortunately, the +// signal may or may not be delivered to the current thread; in order to do that +// portably, we would need to add a cgo dependency and call pthread_kill. +// +// If successful, abortProcess does not return. +func abortProcess() error { + p, err := os.FindProcess(os.Getpid()) + if err != nil { + return err + } + if err := p.Signal(syscall.SIGABRT); err != nil { + return err + } + + // Sent the signal. Now we wait for it to arrive and any SIGABRT handlers to + // run (and eventually terminate the process themselves). + // + // We could just "select{}" here, but there's an outside chance that would + // trigger the runtime's deadlock detector if there happen not to be any + // background goroutines running. So we'll sleep a while first to give + // the signal some time. + time.Sleep(10 * time.Second) + select {} +} diff --git a/vendor/github.com/golang/glog/glog_flags.go b/vendor/github.com/golang/glog/glog_flags.go new file mode 100644 index 00000000000..fa4371afd32 --- /dev/null +++ b/vendor/github.com/golang/glog/glog_flags.go @@ -0,0 +1,398 @@ +// Go support for leveled logs, analogous to https://github.com/google/glog. +// +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package glog + +import ( + "bytes" + "errors" + "flag" + "fmt" + "path/filepath" + "runtime" + "strconv" + "strings" + "sync" + "sync/atomic" + + "github.com/golang/glog/internal/logsink" +) + +// modulePat contains a filter for the -vmodule flag. +// It holds a verbosity level and a file pattern to match. +type modulePat struct { + pattern string + literal bool // The pattern is a literal string + full bool // The pattern wants to match the full path + level Level +} + +// match reports whether the file matches the pattern. It uses a string +// comparison if the pattern contains no metacharacters. +func (m *modulePat) match(full, file string) bool { + if m.literal { + if m.full { + return full == m.pattern + } + return file == m.pattern + } + if m.full { + match, _ := filepath.Match(m.pattern, full) + return match + } + match, _ := filepath.Match(m.pattern, file) + return match +} + +// isLiteral reports whether the pattern is a literal string, that is, has no metacharacters +// that require filepath.Match to be called to match the pattern. +func isLiteral(pattern string) bool { + return !strings.ContainsAny(pattern, `\*?[]`) +} + +// isFull reports whether the pattern matches the full file path, that is, +// whether it contains /. +func isFull(pattern string) bool { + return strings.ContainsRune(pattern, '/') +} + +// verboseFlags represents the setting of the -v and -vmodule flags. +type verboseFlags struct { + // moduleLevelCache is a sync.Map storing the -vmodule Level for each V() + // call site, identified by PC. If there is no matching -vmodule filter, + // the cached value is exactly v. moduleLevelCache is replaced with a new + // Map whenever the -vmodule or -v flag changes state. + moduleLevelCache atomic.Value + + // mu guards all fields below. + mu sync.Mutex + + // v stores the value of the -v flag. It may be read safely using + // sync.LoadInt32, but is only modified under mu. + v Level + + // module stores the parsed -vmodule flag. + module []modulePat + + // moduleLength caches len(module). If greater than zero, it + // means vmodule is enabled. It may be read safely using sync.LoadInt32, but + // is only modified under mu. + moduleLength int32 +} + +// NOTE: For compatibility with the open-sourced v1 version of this +// package (github.com/golang/glog) we need to retain that flag.Level +// implements the flag.Value interface. See also go/log-vs-glog. + +// String is part of the flag.Value interface. +func (l *Level) String() string { + return strconv.FormatInt(int64(l.Get().(Level)), 10) +} + +// Get is part of the flag.Value interface. +func (l *Level) Get() any { + if l == &vflags.v { + // l is the value registered for the -v flag. + return Level(atomic.LoadInt32((*int32)(l))) + } + return *l +} + +// Set is part of the flag.Value interface. +func (l *Level) Set(value string) error { + v, err := strconv.Atoi(value) + if err != nil { + return err + } + if l == &vflags.v { + // l is the value registered for the -v flag. + vflags.mu.Lock() + defer vflags.mu.Unlock() + vflags.moduleLevelCache.Store(&sync.Map{}) + atomic.StoreInt32((*int32)(l), int32(v)) + return nil + } + *l = Level(v) + return nil +} + +// vModuleFlag is the flag.Value for the --vmodule flag. +type vModuleFlag struct{ *verboseFlags } + +func (f vModuleFlag) String() string { + // Do not panic on the zero value. + // https://groups.google.com/g/golang-nuts/c/Atlr8uAjn6U/m/iId17Td5BQAJ. + if f.verboseFlags == nil { + return "" + } + f.mu.Lock() + defer f.mu.Unlock() + + var b bytes.Buffer + for i, f := range f.module { + if i > 0 { + b.WriteRune(',') + } + fmt.Fprintf(&b, "%s=%d", f.pattern, f.level) + } + return b.String() +} + +// Get returns nil for this flag type since the struct is not exported. +func (f vModuleFlag) Get() any { return nil } + +var errVmoduleSyntax = errors.New("syntax error: expect comma-separated list of filename=N") + +// Syntax: -vmodule=recordio=2,foo/bar/baz=1,gfs*=3 +func (f vModuleFlag) Set(value string) error { + var filter []modulePat + for _, pat := range strings.Split(value, ",") { + if len(pat) == 0 { + // Empty strings such as from a trailing comma can be ignored. + continue + } + patLev := strings.Split(pat, "=") + if len(patLev) != 2 || len(patLev[0]) == 0 || len(patLev[1]) == 0 { + return errVmoduleSyntax + } + pattern := patLev[0] + v, err := strconv.Atoi(patLev[1]) + if err != nil { + return errors.New("syntax error: expect comma-separated list of filename=N") + } + // TODO: check syntax of filter? + filter = append(filter, modulePat{pattern, isLiteral(pattern), isFull(pattern), Level(v)}) + } + + f.mu.Lock() + defer f.mu.Unlock() + f.module = filter + atomic.StoreInt32((*int32)(&f.moduleLength), int32(len(f.module))) + f.moduleLevelCache.Store(&sync.Map{}) + return nil +} + +func (f *verboseFlags) levelForPC(pc uintptr) Level { + if level, ok := f.moduleLevelCache.Load().(*sync.Map).Load(pc); ok { + return level.(Level) + } + + f.mu.Lock() + defer f.mu.Unlock() + level := Level(f.v) + fn := runtime.FuncForPC(pc) + file, _ := fn.FileLine(pc) + // The file is something like /a/b/c/d.go. We want just the d for + // regular matches, /a/b/c/d for full matches. + file = strings.TrimSuffix(file, ".go") + full := file + if slash := strings.LastIndex(file, "/"); slash >= 0 { + file = file[slash+1:] + } + for _, filter := range f.module { + if filter.match(full, file) { + level = filter.level + break // Use the first matching level. + } + } + f.moduleLevelCache.Load().(*sync.Map).Store(pc, level) + return level +} + +func (f *verboseFlags) enabled(callerDepth int, level Level) bool { + if atomic.LoadInt32(&f.moduleLength) == 0 { + // No vmodule values specified, so compare against v level. + return Level(atomic.LoadInt32((*int32)(&f.v))) >= level + } + + pcs := [1]uintptr{} + if runtime.Callers(callerDepth+2, pcs[:]) < 1 { + return false + } + frame, _ := runtime.CallersFrames(pcs[:]).Next() + return f.levelForPC(frame.Entry) >= level +} + +// traceLocation represents an entry in the -log_backtrace_at flag. +type traceLocation struct { + file string + line int +} + +var errTraceSyntax = errors.New("syntax error: expect file.go:234") + +func parseTraceLocation(value string) (traceLocation, error) { + fields := strings.Split(value, ":") + if len(fields) != 2 { + return traceLocation{}, errTraceSyntax + } + file, lineStr := fields[0], fields[1] + if !strings.Contains(file, ".") { + return traceLocation{}, errTraceSyntax + } + line, err := strconv.Atoi(lineStr) + if err != nil { + return traceLocation{}, errTraceSyntax + } + if line < 0 { + return traceLocation{}, errors.New("negative value for line") + } + return traceLocation{file, line}, nil +} + +// match reports whether the specified file and line matches the trace location. +// The argument file name is the full path, not the basename specified in the flag. +func (t traceLocation) match(file string, line int) bool { + if t.line != line { + return false + } + if i := strings.LastIndex(file, "/"); i >= 0 { + file = file[i+1:] + } + return t.file == file +} + +func (t traceLocation) String() string { + return fmt.Sprintf("%s:%d", t.file, t.line) +} + +// traceLocations represents the -log_backtrace_at flag. +// Syntax: -log_backtrace_at=recordio.go:234,sstable.go:456 +// Note that unlike vmodule the file extension is included here. +type traceLocations struct { + mu sync.Mutex + locsLen int32 // Safe for atomic read without mu. + locs []traceLocation +} + +func (t *traceLocations) String() string { + t.mu.Lock() + defer t.mu.Unlock() + + var buf bytes.Buffer + for i, tl := range t.locs { + if i > 0 { + buf.WriteString(",") + } + buf.WriteString(tl.String()) + } + return buf.String() +} + +// Get always returns nil for this flag type since the struct is not exported +func (t *traceLocations) Get() any { return nil } + +func (t *traceLocations) Set(value string) error { + var locs []traceLocation + for _, s := range strings.Split(value, ",") { + if s == "" { + continue + } + loc, err := parseTraceLocation(s) + if err != nil { + return err + } + locs = append(locs, loc) + } + + t.mu.Lock() + defer t.mu.Unlock() + atomic.StoreInt32(&t.locsLen, int32(len(locs))) + t.locs = locs + return nil +} + +func (t *traceLocations) match(file string, line int) bool { + if atomic.LoadInt32(&t.locsLen) == 0 { + return false + } + + t.mu.Lock() + defer t.mu.Unlock() + for _, tl := range t.locs { + if tl.match(file, line) { + return true + } + } + return false +} + +// severityFlag is an atomic flag.Value implementation for logsink.Severity. +type severityFlag int32 + +func (s *severityFlag) get() logsink.Severity { + return logsink.Severity(atomic.LoadInt32((*int32)(s))) +} +func (s *severityFlag) String() string { return strconv.FormatInt(int64(*s), 10) } +func (s *severityFlag) Get() any { return s.get() } +func (s *severityFlag) Set(value string) error { + threshold, err := logsink.ParseSeverity(value) + if err != nil { + // Not a severity name. Try a raw number. + v, err := strconv.Atoi(value) + if err != nil { + return err + } + threshold = logsink.Severity(v) + if threshold < logsink.Info || threshold > logsink.Fatal { + return fmt.Errorf("Severity %d out of range (min %d, max %d).", v, logsink.Info, logsink.Fatal) + } + } + atomic.StoreInt32((*int32)(s), int32(threshold)) + return nil +} + +var ( + vflags verboseFlags // The -v and -vmodule flags. + + logBacktraceAt traceLocations // The -log_backtrace_at flag. + + // Boolean flags. Not handled atomically because the flag.Value interface + // does not let us avoid the =true, and that shorthand is necessary for + // compatibility. TODO: does this matter enough to fix? Seems unlikely. + toStderr bool // The -logtostderr flag. + alsoToStderr bool // The -alsologtostderr flag. + + stderrThreshold severityFlag // The -stderrthreshold flag. +) + +// verboseEnabled returns whether the caller at the given depth should emit +// verbose logs at the given level, with depth 0 identifying the caller of +// verboseEnabled. +func verboseEnabled(callerDepth int, level Level) bool { + return vflags.enabled(callerDepth+1, level) +} + +// backtraceAt returns whether the logging call at the given function and line +// should also emit a backtrace of the current call stack. +func backtraceAt(file string, line int) bool { + return logBacktraceAt.match(file, line) +} + +func init() { + vflags.moduleLevelCache.Store(&sync.Map{}) + + flag.Var(&vflags.v, "v", "log level for V logs") + flag.Var(vModuleFlag{&vflags}, "vmodule", "comma-separated list of pattern=N settings for file-filtered logging") + + flag.Var(&logBacktraceAt, "log_backtrace_at", "when logging hits line file:N, emit a stack trace") + + stderrThreshold = severityFlag(logsink.Error) + + flag.BoolVar(&toStderr, "logtostderr", false, "log to standard error instead of files") + flag.BoolVar(&alsoToStderr, "alsologtostderr", false, "log to standard error as well as files") + flag.Var(&stderrThreshold, "stderrthreshold", "logs at or above this threshold go to stderr") +} diff --git a/vendor/github.com/golang/glog/internal/logsink/logsink.go b/vendor/github.com/golang/glog/internal/logsink/logsink.go new file mode 100644 index 00000000000..53758e1c9f5 --- /dev/null +++ b/vendor/github.com/golang/glog/internal/logsink/logsink.go @@ -0,0 +1,387 @@ +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package logsink + +import ( + "bytes" + "fmt" + "strconv" + "strings" + "sync" + "time" + + "github.com/golang/glog/internal/stackdump" +) + +// MaxLogMessageLen is the limit on length of a formatted log message, including +// the standard line prefix and trailing newline. +// +// Chosen to match C++ glog. +const MaxLogMessageLen = 15000 + +// A Severity is a severity at which a message can be logged. +type Severity int8 + +// These constants identify the log levels in order of increasing severity. +// A message written to a high-severity log file is also written to each +// lower-severity log file. +const ( + Info Severity = iota + Warning + Error + + // Fatal contains logs written immediately before the process terminates. + // + // Sink implementations should not terminate the process themselves: the log + // package will perform any necessary cleanup and terminate the process as + // appropriate. + Fatal +) + +func (s Severity) String() string { + switch s { + case Info: + return "INFO" + case Warning: + return "WARNING" + case Error: + return "ERROR" + case Fatal: + return "FATAL" + } + return fmt.Sprintf("%T(%d)", s, s) +} + +// ParseSeverity returns the case-insensitive Severity value for the given string. +func ParseSeverity(name string) (Severity, error) { + name = strings.ToUpper(name) + for s := Info; s <= Fatal; s++ { + if s.String() == name { + return s, nil + } + } + return -1, fmt.Errorf("logsink: invalid severity %q", name) +} + +// Meta is metadata about a logging call. +type Meta struct { + // Time is the time at which the log call was made. + Time time.Time + + // File is the source file from which the log entry originates. + File string + // Line is the line offset within the source file. + Line int + // Depth is the number of stack frames between the logsink and the log call. + Depth int + + Severity Severity + + // Verbose indicates whether the call was made via "log.V". Log entries below + // the current verbosity threshold are not sent to the sink. + Verbose bool + + // Thread ID. This can be populated with a thread ID from another source, + // such as a system we are importing logs from. In the normal case, this + // will be set to the process ID (PID), since Go doesn't have threads. + Thread int64 + + // Stack trace starting in the logging function. May be nil. + // A logsink should implement the StackWanter interface to request this. + // + // Even if WantStack returns false, this field may be set (e.g. if another + // sink wants a stack trace). + Stack *stackdump.Stack +} + +// Structured is a logging destination that accepts structured data as input. +type Structured interface { + // Printf formats according to a fmt.Printf format specifier and writes a log + // entry. The precise result of formatting depends on the sink, but should + // aim for consistency with fmt.Printf. + // + // Printf returns the number of bytes occupied by the log entry, which + // may not be equal to the total number of bytes written. + // + // Printf returns any error encountered *if* it is severe enough that the log + // package should terminate the process. + // + // The sink must not modify the *Meta parameter, nor reference it after + // Printf has returned: it may be reused in subsequent calls. + Printf(meta *Meta, format string, a ...any) (n int, err error) +} + +// StackWanter can be implemented by a logsink.Structured to indicate that it +// wants a stack trace to accompany at least some of the log messages it receives. +type StackWanter interface { + // WantStack returns true if the sink requires a stack trace for a log message + // with this metadata. + // + // NOTE: Returning true implies that meta.Stack will be non-nil. Returning + // false does NOT imply that meta.Stack will be nil. + WantStack(meta *Meta) bool +} + +// Text is a logging destination that accepts pre-formatted log lines (instead of +// structured data). +type Text interface { + // Enabled returns whether this sink should output messages for the given + // Meta. If the sink returns false for a given Meta, the Printf function will + // not call Emit on it for the corresponding log message. + Enabled(*Meta) bool + + // Emit writes a pre-formatted text log entry (including any applicable + // header) to the log. It returns the number of bytes occupied by the entry + // (which may differ from the length of the passed-in slice). + // + // Emit returns any error encountered *if* it is severe enough that the log + // package should terminate the process. + // + // The sink must not modify the *Meta parameter, nor reference it after + // Printf has returned: it may be reused in subsequent calls. + // + // NOTE: When developing a text sink, keep in mind the surface in which the + // logs will be displayed, and whether it's important that the sink be + // resistent to tampering in the style of b/211428300. Standard text sinks + // (like `stderrSink`) do not protect against this (e.g. by escaping + // characters) because the cases where they would show user-influenced bytes + // are vanishingly small. + Emit(*Meta, []byte) (n int, err error) +} + +// bufs is a pool of *bytes.Buffer used in formatting log entries. +var bufs sync.Pool // Pool of *bytes.Buffer. + +// textPrintf formats a text log entry and emits it to all specified Text sinks. +// +// The returned n is the maximum across all Emit calls. +// The returned err is the first non-nil error encountered. +// Sinks that are disabled by configuration should return (0, nil). +func textPrintf(m *Meta, textSinks []Text, format string, args ...any) (n int, err error) { + // We expect at most file, stderr, and perhaps syslog. If there are more, + // we'll end up allocating - no big deal. + const maxExpectedTextSinks = 3 + var noAllocSinks [maxExpectedTextSinks]Text + + sinks := noAllocSinks[:0] + for _, s := range textSinks { + if s.Enabled(m) { + sinks = append(sinks, s) + } + } + if len(sinks) == 0 && m.Severity != Fatal { + return 0, nil // No TextSinks specified; don't bother formatting. + } + + bufi := bufs.Get() + var buf *bytes.Buffer + if bufi == nil { + buf = bytes.NewBuffer(nil) + bufi = buf + } else { + buf = bufi.(*bytes.Buffer) + buf.Reset() + } + + // Lmmdd hh:mm:ss.uuuuuu PID/GID file:line] + // + // The "PID" entry arguably ought to be TID for consistency with other + // environments, but TID is not meaningful in a Go program due to the + // multiplexing of goroutines across threads. + // + // Avoid Fprintf, for speed. The format is so simple that we can do it quickly by hand. + // It's worth about 3X. Fprintf is hard. + const severityChar = "IWEF" + buf.WriteByte(severityChar[m.Severity]) + + _, month, day := m.Time.Date() + hour, minute, second := m.Time.Clock() + twoDigits(buf, int(month)) + twoDigits(buf, day) + buf.WriteByte(' ') + twoDigits(buf, hour) + buf.WriteByte(':') + twoDigits(buf, minute) + buf.WriteByte(':') + twoDigits(buf, second) + buf.WriteByte('.') + nDigits(buf, 6, uint64(m.Time.Nanosecond()/1000), '0') + buf.WriteByte(' ') + + nDigits(buf, 7, uint64(m.Thread), ' ') + buf.WriteByte(' ') + + { + file := m.File + if i := strings.LastIndex(file, "/"); i >= 0 { + file = file[i+1:] + } + buf.WriteString(file) + } + + buf.WriteByte(':') + { + var tmp [19]byte + buf.Write(strconv.AppendInt(tmp[:0], int64(m.Line), 10)) + } + buf.WriteString("] ") + + msgStart := buf.Len() + fmt.Fprintf(buf, format, args...) + if buf.Len() > MaxLogMessageLen-1 { + buf.Truncate(MaxLogMessageLen - 1) + } + msgEnd := buf.Len() + if b := buf.Bytes(); b[len(b)-1] != '\n' { + buf.WriteByte('\n') + } + + for _, s := range sinks { + sn, sErr := s.Emit(m, buf.Bytes()) + if sn > n { + n = sn + } + if sErr != nil && err == nil { + err = sErr + } + } + + if m.Severity == Fatal { + savedM := *m + fatalMessageStore(savedEntry{ + meta: &savedM, + msg: buf.Bytes()[msgStart:msgEnd], + }) + } else { + bufs.Put(bufi) + } + return n, err +} + +const digits = "0123456789" + +// twoDigits formats a zero-prefixed two-digit integer to buf. +func twoDigits(buf *bytes.Buffer, d int) { + buf.WriteByte(digits[(d/10)%10]) + buf.WriteByte(digits[d%10]) +} + +// nDigits formats an n-digit integer to buf, padding with pad on the left. It +// assumes d != 0. +func nDigits(buf *bytes.Buffer, n int, d uint64, pad byte) { + var tmp [20]byte + + cutoff := len(tmp) - n + j := len(tmp) - 1 + for ; d > 0; j-- { + tmp[j] = digits[d%10] + d /= 10 + } + for ; j >= cutoff; j-- { + tmp[j] = pad + } + j++ + buf.Write(tmp[j:]) +} + +// Printf writes a log entry to all registered TextSinks in this package, then +// to all registered StructuredSinks. +// +// The returned n is the maximum across all Emit and Printf calls. +// The returned err is the first non-nil error encountered. +// Sinks that are disabled by configuration should return (0, nil). +func Printf(m *Meta, format string, args ...any) (n int, err error) { + m.Depth++ + n, err = textPrintf(m, TextSinks, format, args...) + + for _, sink := range StructuredSinks { + // TODO: Support TextSinks that implement StackWanter? + if sw, ok := sink.(StackWanter); ok && sw.WantStack(m) { + if m.Stack == nil { + // First, try to find a stacktrace in args, otherwise generate one. + for _, arg := range args { + if stack, ok := arg.(stackdump.Stack); ok { + m.Stack = &stack + break + } + } + if m.Stack == nil { + stack := stackdump.Caller( /* skipDepth = */ m.Depth) + m.Stack = &stack + } + } + } + sn, sErr := sink.Printf(m, format, args...) + if sn > n { + n = sn + } + if sErr != nil && err == nil { + err = sErr + } + } + return n, err +} + +// The sets of sinks to which logs should be written. +// +// These must only be modified during package init, and are read-only thereafter. +var ( + // StructuredSinks is the set of Structured sink instances to which logs + // should be written. + StructuredSinks []Structured + + // TextSinks is the set of Text sink instances to which logs should be + // written. + // + // These are registered separately from Structured sink implementations to + // avoid the need to repeat the work of formatting a message for each Text + // sink that writes it. The package-level Printf function writes to both sets + // independenty, so a given log destination should only register a Structured + // *or* a Text sink (not both). + TextSinks []Text +) + +type savedEntry struct { + meta *Meta + msg []byte +} + +// StructuredTextWrapper is a Structured sink which forwards logs to a set of Text sinks. +// +// The purpose of this sink is to allow applications to intercept logging calls before they are +// serialized and sent to Text sinks. For example, if one needs to redact PII from logging +// arguments before they reach STDERR, one solution would be to do the redacting in a Structured +// sink that forwards logs to a StructuredTextWrapper instance, and make STDERR a child of that +// StructuredTextWrapper instance. This is how one could set this up in their application: +// +// func init() { +// +// wrapper := logsink.StructuredTextWrapper{TextSinks: logsink.TextSinks} +// // sanitizersink will intercept logs and remove PII +// sanitizer := sanitizersink{Sink: &wrapper} +// logsink.StructuredSinks = append(logsink.StructuredSinks, &sanitizer) +// logsink.TextSinks = nil +// +// } +type StructuredTextWrapper struct { + // TextSinks is the set of Text sinks that should receive logs from this + // StructuredTextWrapper instance. + TextSinks []Text +} + +// Printf forwards logs to all Text sinks registered in the StructuredTextWrapper. +func (w *StructuredTextWrapper) Printf(meta *Meta, format string, args ...any) (n int, err error) { + return textPrintf(meta, w.TextSinks, format, args...) +} diff --git a/vendor/github.com/golang/glog/internal/logsink/logsink_fatal.go b/vendor/github.com/golang/glog/internal/logsink/logsink_fatal.go new file mode 100644 index 00000000000..3dc269abc21 --- /dev/null +++ b/vendor/github.com/golang/glog/internal/logsink/logsink_fatal.go @@ -0,0 +1,35 @@ +package logsink + +import ( + "sync/atomic" + "unsafe" +) + +func fatalMessageStore(e savedEntry) { + // Only put a new one in if we haven't assigned before. + atomic.CompareAndSwapPointer(&fatalMessage, nil, unsafe.Pointer(&e)) +} + +var fatalMessage unsafe.Pointer // savedEntry stored with CompareAndSwapPointer + +// FatalMessage returns the Meta and message contents of the first message +// logged with Fatal severity, or false if none has occurred. +func FatalMessage() (*Meta, []byte, bool) { + e := (*savedEntry)(atomic.LoadPointer(&fatalMessage)) + if e == nil { + return nil, nil, false + } + return e.meta, e.msg, true +} + +// DoNotUseRacyFatalMessage is FatalMessage, but worse. +// +//go:norace +//go:nosplit +func DoNotUseRacyFatalMessage() (*Meta, []byte, bool) { + e := (*savedEntry)(fatalMessage) + if e == nil { + return nil, nil, false + } + return e.meta, e.msg, true +} diff --git a/vendor/github.com/golang/glog/internal/stackdump/stackdump.go b/vendor/github.com/golang/glog/internal/stackdump/stackdump.go new file mode 100644 index 00000000000..3427c9d6bd0 --- /dev/null +++ b/vendor/github.com/golang/glog/internal/stackdump/stackdump.go @@ -0,0 +1,127 @@ +// Copyright 2023 Google Inc. All Rights Reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package stackdump provides wrappers for runtime.Stack and runtime.Callers +// with uniform support for skipping caller frames. +// +// ⚠ Unlike the functions in the runtime package, these may allocate a +// non-trivial quantity of memory: use them with care. ⚠ +package stackdump + +import ( + "bytes" + "runtime" +) + +// runtimeStackSelfFrames is 1 if runtime.Stack includes the call to +// runtime.Stack itself or 0 if it does not. +// +// As of 2016-04-27, the gccgo compiler includes runtime.Stack but the gc +// compiler does not. +var runtimeStackSelfFrames = func() int { + for n := 1 << 10; n < 1<<20; n *= 2 { + buf := make([]byte, n) + n := runtime.Stack(buf, false) + if bytes.Contains(buf[:n], []byte("runtime.Stack")) { + return 1 + } else if n < len(buf) || bytes.Count(buf, []byte("\n")) >= 3 { + return 0 + } + } + return 0 +}() + +// Stack is a stack dump for a single goroutine. +type Stack struct { + // Text is a representation of the stack dump in a human-readable format. + Text []byte + + // PC is a representation of the stack dump using raw program counter values. + PC []uintptr +} + +func (s Stack) String() string { return string(s.Text) } + +// Caller returns the Stack dump for the calling goroutine, starting skipDepth +// frames before the caller of Caller. (Caller(0) provides a dump starting at +// the caller of this function.) +func Caller(skipDepth int) Stack { + return Stack{ + Text: CallerText(skipDepth + 1), + PC: CallerPC(skipDepth + 1), + } +} + +// CallerText returns a textual dump of the stack starting skipDepth frames before +// the caller. (CallerText(0) provides a dump starting at the caller of this +// function.) +func CallerText(skipDepth int) []byte { + for n := 1 << 10; ; n *= 2 { + buf := make([]byte, n) + n := runtime.Stack(buf, false) + if n < len(buf) { + return pruneFrames(skipDepth+1+runtimeStackSelfFrames, buf[:n]) + } + } +} + +// CallerPC returns a dump of the program counters of the stack starting +// skipDepth frames before the caller. (CallerPC(0) provides a dump starting at +// the caller of this function.) +func CallerPC(skipDepth int) []uintptr { + for n := 1 << 8; ; n *= 2 { + buf := make([]uintptr, n) + n := runtime.Callers(skipDepth+2, buf) + if n < len(buf) { + return buf[:n] + } + } +} + +// pruneFrames removes the topmost skipDepth frames of the first goroutine in a +// textual stack dump. It overwrites the passed-in slice. +// +// If there are fewer than skipDepth frames in the first goroutine's stack, +// pruneFrames prunes it to an empty stack and leaves the remaining contents +// intact. +func pruneFrames(skipDepth int, stack []byte) []byte { + headerLen := 0 + for i, c := range stack { + if c == '\n' { + headerLen = i + 1 + break + } + } + if headerLen == 0 { + return stack // No header line - not a well-formed stack trace. + } + + skipLen := headerLen + skipNewlines := skipDepth * 2 + for ; skipLen < len(stack) && skipNewlines > 0; skipLen++ { + c := stack[skipLen] + if c != '\n' { + continue + } + skipNewlines-- + skipLen++ + if skipNewlines == 0 || skipLen == len(stack) || stack[skipLen] == '\n' { + break + } + } + + pruned := stack[skipLen-headerLen:] + copy(pruned, stack[:headerLen]) + return pruned +} diff --git a/vendor/github.com/golang/groupcache/LICENSE b/vendor/github.com/golang/groupcache/LICENSE new file mode 100644 index 00000000000..37ec93a14fd --- /dev/null +++ b/vendor/github.com/golang/groupcache/LICENSE @@ -0,0 +1,191 @@ +Apache License +Version 2.0, January 2004 +http://www.apache.org/licenses/ + +TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + +1. Definitions. + +"License" shall mean the terms and conditions for use, reproduction, and +distribution as defined by Sections 1 through 9 of this document. + +"Licensor" shall mean the copyright owner or entity authorized by the copyright +owner that is granting the License. + +"Legal Entity" shall mean the union of the acting entity and all other entities +that control, are controlled by, or are under common control with that entity. +For the purposes of this definition, "control" means (i) the power, direct or +indirect, to cause the direction or management of such entity, whether by +contract or otherwise, or (ii) ownership of fifty percent (50%) or more of the +outstanding shares, or (iii) beneficial ownership of such entity. + +"You" (or "Your") shall mean an individual or Legal Entity exercising +permissions granted by this License. + +"Source" form shall mean the preferred form for making modifications, including +but not limited to software source code, documentation source, and configuration +files. + +"Object" form shall mean any form resulting from mechanical transformation or +translation of a Source form, including but not limited to compiled object code, +generated documentation, and conversions to other media types. + +"Work" shall mean the work of authorship, whether in Source or Object form, made +available under the License, as indicated by a copyright notice that is included +in or attached to the work (an example is provided in the Appendix below). + +"Derivative Works" shall mean any work, whether in Source or Object form, that +is based on (or derived from) the Work and for which the editorial revisions, +annotations, elaborations, or other modifications represent, as a whole, an +original work of authorship. For the purposes of this License, Derivative Works +shall not include works that remain separable from, or merely link (or bind by +name) to the interfaces of, the Work and Derivative Works thereof. + +"Contribution" shall mean any work of authorship, including the original version +of the Work and any modifications or additions to that Work or Derivative Works +thereof, that is intentionally submitted to Licensor for inclusion in the Work +by the copyright owner or by an individual or Legal Entity authorized to submit +on behalf of the copyright owner. For the purposes of this definition, +"submitted" means any form of electronic, verbal, or written communication sent +to the Licensor or its representatives, including but not limited to +communication on electronic mailing lists, source code control systems, and +issue tracking systems that are managed by, or on behalf of, the Licensor for +the purpose of discussing and improving the Work, but excluding communication +that is conspicuously marked or otherwise designated in writing by the copyright +owner as "Not a Contribution." + +"Contributor" shall mean Licensor and any individual or Legal Entity on behalf +of whom a Contribution has been received by Licensor and subsequently +incorporated within the Work. + +2. Grant of Copyright License. + +Subject to the terms and conditions of this License, each Contributor hereby +grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, +irrevocable copyright license to reproduce, prepare Derivative Works of, +publicly display, publicly perform, sublicense, and distribute the Work and such +Derivative Works in Source or Object form. + +3. Grant of Patent License. + +Subject to the terms and conditions of this License, each Contributor hereby +grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, +irrevocable (except as stated in this section) patent license to make, have +made, use, offer to sell, sell, import, and otherwise transfer the Work, where +such license applies only to those patent claims licensable by such Contributor +that are necessarily infringed by their Contribution(s) alone or by combination +of their Contribution(s) with the Work to which such Contribution(s) was +submitted. If You institute patent litigation against any entity (including a +cross-claim or counterclaim in a lawsuit) alleging that the Work or a +Contribution incorporated within the Work constitutes direct or contributory +patent infringement, then any patent licenses granted to You under this License +for that Work shall terminate as of the date such litigation is filed. + +4. Redistribution. + +You may reproduce and distribute copies of the Work or Derivative Works thereof +in any medium, with or without modifications, and in Source or Object form, +provided that You meet the following conditions: + +You must give any other recipients of the Work or Derivative Works a copy of +this License; and +You must cause any modified files to carry prominent notices stating that You +changed the files; and +You must retain, in the Source form of any Derivative Works that You distribute, +all copyright, patent, trademark, and attribution notices from the Source form +of the Work, excluding those notices that do not pertain to any part of the +Derivative Works; and +If the Work includes a "NOTICE" text file as part of its distribution, then any +Derivative Works that You distribute must include a readable copy of the +attribution notices contained within such NOTICE file, excluding those notices +that do not pertain to any part of the Derivative Works, in at least one of the +following places: within a NOTICE text file distributed as part of the +Derivative Works; within the Source form or documentation, if provided along +with the Derivative Works; or, within a display generated by the Derivative +Works, if and wherever such third-party notices normally appear. The contents of +the NOTICE file are for informational purposes only and do not modify the +License. You may add Your own attribution notices within Derivative Works that +You distribute, alongside or as an addendum to the NOTICE text from the Work, +provided that such additional attribution notices cannot be construed as +modifying the License. +You may add Your own copyright statement to Your modifications and may provide +additional or different license terms and conditions for use, reproduction, or +distribution of Your modifications, or for any such Derivative Works as a whole, +provided Your use, reproduction, and distribution of the Work otherwise complies +with the conditions stated in this License. + +5. Submission of Contributions. + +Unless You explicitly state otherwise, any Contribution intentionally submitted +for inclusion in the Work by You to the Licensor shall be under the terms and +conditions of this License, without any additional terms or conditions. +Notwithstanding the above, nothing herein shall supersede or modify the terms of +any separate license agreement you may have executed with Licensor regarding +such Contributions. + +6. Trademarks. + +This License does not grant permission to use the trade names, trademarks, +service marks, or product names of the Licensor, except as required for +reasonable and customary use in describing the origin of the Work and +reproducing the content of the NOTICE file. + +7. Disclaimer of Warranty. + +Unless required by applicable law or agreed to in writing, Licensor provides the +Work (and each Contributor provides its Contributions) on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied, +including, without limitation, any warranties or conditions of TITLE, +NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE. You are +solely responsible for determining the appropriateness of using or +redistributing the Work and assume any risks associated with Your exercise of +permissions under this License. + +8. Limitation of Liability. + +In no event and under no legal theory, whether in tort (including negligence), +contract, or otherwise, unless required by applicable law (such as deliberate +and grossly negligent acts) or agreed to in writing, shall any Contributor be +liable to You for damages, including any direct, indirect, special, incidental, +or consequential damages of any character arising as a result of this License or +out of the use or inability to use the Work (including but not limited to +damages for loss of goodwill, work stoppage, computer failure or malfunction, or +any and all other commercial damages or losses), even if such Contributor has +been advised of the possibility of such damages. + +9. Accepting Warranty or Additional Liability. + +While redistributing the Work or Derivative Works thereof, You may choose to +offer, and charge a fee for, acceptance of support, warranty, indemnity, or +other liability obligations and/or rights consistent with this License. However, +in accepting such obligations, You may act only on Your own behalf and on Your +sole responsibility, not on behalf of any other Contributor, and only if You +agree to indemnify, defend, and hold each Contributor harmless for any liability +incurred by, or claims asserted against, such Contributor by reason of your +accepting any such warranty or additional liability. + +END OF TERMS AND CONDITIONS + +APPENDIX: How to apply the Apache License to your work + +To apply the Apache License to your work, attach the following boilerplate +notice, with the fields enclosed by brackets "[]" replaced with your own +identifying information. (Don't include the brackets!) The text should be +enclosed in the appropriate comment syntax for the file format. We also +recommend that a file or class name and description of purpose be included on +the same "printed page" as the copyright notice for easier identification within +third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/vendor/github.com/golang/groupcache/lru/lru.go b/vendor/github.com/golang/groupcache/lru/lru.go new file mode 100644 index 00000000000..eac1c7664f9 --- /dev/null +++ b/vendor/github.com/golang/groupcache/lru/lru.go @@ -0,0 +1,133 @@ +/* +Copyright 2013 Google Inc. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +*/ + +// Package lru implements an LRU cache. +package lru + +import "container/list" + +// Cache is an LRU cache. It is not safe for concurrent access. +type Cache struct { + // MaxEntries is the maximum number of cache entries before + // an item is evicted. Zero means no limit. + MaxEntries int + + // OnEvicted optionally specifies a callback function to be + // executed when an entry is purged from the cache. + OnEvicted func(key Key, value interface{}) + + ll *list.List + cache map[interface{}]*list.Element +} + +// A Key may be any value that is comparable. See http://golang.org/ref/spec#Comparison_operators +type Key interface{} + +type entry struct { + key Key + value interface{} +} + +// New creates a new Cache. +// If maxEntries is zero, the cache has no limit and it's assumed +// that eviction is done by the caller. +func New(maxEntries int) *Cache { + return &Cache{ + MaxEntries: maxEntries, + ll: list.New(), + cache: make(map[interface{}]*list.Element), + } +} + +// Add adds a value to the cache. +func (c *Cache) Add(key Key, value interface{}) { + if c.cache == nil { + c.cache = make(map[interface{}]*list.Element) + c.ll = list.New() + } + if ee, ok := c.cache[key]; ok { + c.ll.MoveToFront(ee) + ee.Value.(*entry).value = value + return + } + ele := c.ll.PushFront(&entry{key, value}) + c.cache[key] = ele + if c.MaxEntries != 0 && c.ll.Len() > c.MaxEntries { + c.RemoveOldest() + } +} + +// Get looks up a key's value from the cache. +func (c *Cache) Get(key Key) (value interface{}, ok bool) { + if c.cache == nil { + return + } + if ele, hit := c.cache[key]; hit { + c.ll.MoveToFront(ele) + return ele.Value.(*entry).value, true + } + return +} + +// Remove removes the provided key from the cache. +func (c *Cache) Remove(key Key) { + if c.cache == nil { + return + } + if ele, hit := c.cache[key]; hit { + c.removeElement(ele) + } +} + +// RemoveOldest removes the oldest item from the cache. +func (c *Cache) RemoveOldest() { + if c.cache == nil { + return + } + ele := c.ll.Back() + if ele != nil { + c.removeElement(ele) + } +} + +func (c *Cache) removeElement(e *list.Element) { + c.ll.Remove(e) + kv := e.Value.(*entry) + delete(c.cache, kv.key) + if c.OnEvicted != nil { + c.OnEvicted(kv.key, kv.value) + } +} + +// Len returns the number of items in the cache. +func (c *Cache) Len() int { + if c.cache == nil { + return 0 + } + return c.ll.Len() +} + +// Clear purges all stored items from the cache. +func (c *Cache) Clear() { + if c.OnEvicted != nil { + for _, e := range c.cache { + kv := e.Value.(*entry) + c.OnEvicted(kv.key, kv.value) + } + } + c.ll = nil + c.cache = nil +} diff --git a/vendor/github.com/google/flatbuffers/LICENSE.txt b/vendor/github.com/google/flatbuffers/LICENSE.txt new file mode 100644 index 00000000000..d6456956733 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/LICENSE.txt @@ -0,0 +1,202 @@ + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/vendor/github.com/google/flatbuffers/go/BUILD.bazel b/vendor/github.com/google/flatbuffers/go/BUILD.bazel new file mode 100644 index 00000000000..78bd8d81ada --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/BUILD.bazel @@ -0,0 +1,23 @@ +load("@io_bazel_rules_go//go:def.bzl", "go_library") + +alias( + name = "go_default_library", + actual = ":go", + visibility = ["//visibility:public"], +) + +go_library( + name = "go", + srcs = [ + "builder.go", + "doc.go", + "encode.go", + "grpc.go", + "lib.go", + "sizes.go", + "struct.go", + "table.go", + ], + importpath = "github.com/google/flatbuffers/go", + visibility = ["//visibility:public"], +) diff --git a/vendor/github.com/google/flatbuffers/go/builder.go b/vendor/github.com/google/flatbuffers/go/builder.go new file mode 100644 index 00000000000..0e763d7ac16 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/builder.go @@ -0,0 +1,786 @@ +package flatbuffers + +// Builder is a state machine for creating FlatBuffer objects. +// Use a Builder to construct object(s) starting from leaf nodes. +// +// A Builder constructs byte buffers in a last-first manner for simplicity and +// performance. +type Builder struct { + // `Bytes` gives raw access to the buffer. Most users will want to use + // FinishedBytes() instead. + Bytes []byte + + minalign int + vtable []UOffsetT + objectEnd UOffsetT + vtables []UOffsetT + head UOffsetT + nested bool + finished bool + + sharedStrings map[string]UOffsetT +} + +const fileIdentifierLength = 4 + +// NewBuilder initializes a Builder of size `initial_size`. +// The internal buffer is grown as needed. +func NewBuilder(initialSize int) *Builder { + if initialSize <= 0 { + initialSize = 0 + } + + b := &Builder{} + b.Bytes = make([]byte, initialSize) + b.head = UOffsetT(initialSize) + b.minalign = 1 + b.vtables = make([]UOffsetT, 0, 16) // sensible default capacity + return b +} + +// Reset truncates the underlying Builder buffer, facilitating alloc-free +// reuse of a Builder. It also resets bookkeeping data. +func (b *Builder) Reset() { + if b.Bytes != nil { + b.Bytes = b.Bytes[:cap(b.Bytes)] + } + + if b.vtables != nil { + b.vtables = b.vtables[:0] + } + + if b.vtable != nil { + b.vtable = b.vtable[:0] + } + + b.head = UOffsetT(len(b.Bytes)) + b.minalign = 1 + b.nested = false + b.finished = false +} + +// FinishedBytes returns a pointer to the written data in the byte buffer. +// Panics if the builder is not in a finished state (which is caused by calling +// `Finish()`). +func (b *Builder) FinishedBytes() []byte { + b.assertFinished() + return b.Bytes[b.Head():] +} + +// StartObject initializes bookkeeping for writing a new object. +func (b *Builder) StartObject(numfields int) { + b.assertNotNested() + b.nested = true + + // use 32-bit offsets so that arithmetic doesn't overflow. + if cap(b.vtable) < numfields || b.vtable == nil { + b.vtable = make([]UOffsetT, numfields) + } else { + b.vtable = b.vtable[:numfields] + for i := 0; i < len(b.vtable); i++ { + b.vtable[i] = 0 + } + } + + b.objectEnd = b.Offset() +} + +// WriteVtable serializes the vtable for the current object, if applicable. +// +// Before writing out the vtable, this checks pre-existing vtables for equality +// to this one. If an equal vtable is found, point the object to the existing +// vtable and return. +// +// Because vtable values are sensitive to alignment of object data, not all +// logically-equal vtables will be deduplicated. +// +// A vtable has the following format: +// +// +// * N, where N is the number of fields in +// the schema for this type. Includes deprecated fields. +// Thus, a vtable is made of 2 + N elements, each SizeVOffsetT bytes wide. +// +// An object has the following format: +// +// + +func (b *Builder) WriteVtable() (n UOffsetT) { + // Prepend a zero scalar to the object. Later in this function we'll + // write an offset here that points to the object's vtable: + b.PrependSOffsetT(0) + + objectOffset := b.Offset() + existingVtable := UOffsetT(0) + + // Trim vtable of trailing zeroes. + i := len(b.vtable) - 1 + for ; i >= 0 && b.vtable[i] == 0; i-- { + } + b.vtable = b.vtable[:i+1] + + // Search backwards through existing vtables, because similar vtables + // are likely to have been recently appended. See + // BenchmarkVtableDeduplication for a case in which this heuristic + // saves about 30% of the time used in writing objects with duplicate + // tables. + for i := len(b.vtables) - 1; i >= 0; i-- { + // Find the other vtable, which is associated with `i`: + vt2Offset := b.vtables[i] + vt2Start := len(b.Bytes) - int(vt2Offset) + vt2Len := GetVOffsetT(b.Bytes[vt2Start:]) + + metadata := VtableMetadataFields * SizeVOffsetT + vt2End := vt2Start + int(vt2Len) + vt2 := b.Bytes[vt2Start+metadata : vt2End] + + // Compare the other vtable to the one under consideration. + // If they are equal, store the offset and break: + if vtableEqual(b.vtable, objectOffset, vt2) { + existingVtable = vt2Offset + break + } + } + + if existingVtable == 0 { + // Did not find a vtable, so write this one to the buffer. + + // Write out the current vtable in reverse , because + // serialization occurs in last-first order: + for i := len(b.vtable) - 1; i >= 0; i-- { + var off UOffsetT + if b.vtable[i] != 0 { + // Forward reference to field; + // use 32bit number to assert no overflow: + off = objectOffset - b.vtable[i] + } + + b.PrependVOffsetT(VOffsetT(off)) + } + + // The two metadata fields are written last. + + // First, store the object bytesize: + objectSize := objectOffset - b.objectEnd + b.PrependVOffsetT(VOffsetT(objectSize)) + + // Second, store the vtable bytesize: + vBytes := (len(b.vtable) + VtableMetadataFields) * SizeVOffsetT + b.PrependVOffsetT(VOffsetT(vBytes)) + + // Next, write the offset to the new vtable in the + // already-allocated SOffsetT at the beginning of this object: + objectStart := SOffsetT(len(b.Bytes)) - SOffsetT(objectOffset) + WriteSOffsetT(b.Bytes[objectStart:], + SOffsetT(b.Offset())-SOffsetT(objectOffset)) + + // Finally, store this vtable in memory for future + // deduplication: + b.vtables = append(b.vtables, b.Offset()) + } else { + // Found a duplicate vtable. + + objectStart := SOffsetT(len(b.Bytes)) - SOffsetT(objectOffset) + b.head = UOffsetT(objectStart) + + // Write the offset to the found vtable in the + // already-allocated SOffsetT at the beginning of this object: + WriteSOffsetT(b.Bytes[b.head:], + SOffsetT(existingVtable)-SOffsetT(objectOffset)) + } + + b.vtable = b.vtable[:0] + return objectOffset +} + +// EndObject writes data necessary to finish object construction. +func (b *Builder) EndObject() UOffsetT { + b.assertNested() + n := b.WriteVtable() + b.nested = false + return n +} + +// Doubles the size of the byteslice, and copies the old data towards the +// end of the new byteslice (since we build the buffer backwards). +func (b *Builder) growByteBuffer() { + if (int64(len(b.Bytes)) & int64(0xC0000000)) != 0 { + panic("cannot grow buffer beyond 2 gigabytes") + } + newLen := len(b.Bytes) * 2 + if newLen == 0 { + newLen = 1 + } + + if cap(b.Bytes) >= newLen { + b.Bytes = b.Bytes[:newLen] + } else { + extension := make([]byte, newLen-len(b.Bytes)) + b.Bytes = append(b.Bytes, extension...) + } + + middle := newLen / 2 + copy(b.Bytes[middle:], b.Bytes[:middle]) +} + +// Head gives the start of useful data in the underlying byte buffer. +// Note: unlike other functions, this value is interpreted as from the left. +func (b *Builder) Head() UOffsetT { + return b.head +} + +// Offset relative to the end of the buffer. +func (b *Builder) Offset() UOffsetT { + return UOffsetT(len(b.Bytes)) - b.head +} + +// Pad places zeros at the current offset. +func (b *Builder) Pad(n int) { + for i := 0; i < n; i++ { + b.PlaceByte(0) + } +} + +// Prep prepares to write an element of `size` after `additional_bytes` +// have been written, e.g. if you write a string, you need to align such +// the int length field is aligned to SizeInt32, and the string data follows it +// directly. +// If all you need to do is align, `additionalBytes` will be 0. +func (b *Builder) Prep(size, additionalBytes int) { + // Track the biggest thing we've ever aligned to. + if size > b.minalign { + b.minalign = size + } + // Find the amount of alignment needed such that `size` is properly + // aligned after `additionalBytes`: + alignSize := (^(len(b.Bytes) - int(b.Head()) + additionalBytes)) + 1 + alignSize &= (size - 1) + + // Reallocate the buffer if needed: + for int(b.head) <= alignSize+size+additionalBytes { + oldBufSize := len(b.Bytes) + b.growByteBuffer() + b.head += UOffsetT(len(b.Bytes) - oldBufSize) + } + b.Pad(alignSize) +} + +// PrependSOffsetT prepends an SOffsetT, relative to where it will be written. +func (b *Builder) PrependSOffsetT(off SOffsetT) { + b.Prep(SizeSOffsetT, 0) // Ensure alignment is already done. + if !(UOffsetT(off) <= b.Offset()) { + panic("unreachable: off <= b.Offset()") + } + off2 := SOffsetT(b.Offset()) - off + SOffsetT(SizeSOffsetT) + b.PlaceSOffsetT(off2) +} + +// PrependUOffsetT prepends an UOffsetT, relative to where it will be written. +func (b *Builder) PrependUOffsetT(off UOffsetT) { + b.Prep(SizeUOffsetT, 0) // Ensure alignment is already done. + if !(off <= b.Offset()) { + panic("unreachable: off <= b.Offset()") + } + off2 := b.Offset() - off + UOffsetT(SizeUOffsetT) + b.PlaceUOffsetT(off2) +} + +// StartVector initializes bookkeeping for writing a new vector. +// +// A vector has the following format: +// +// +, where T is the type of elements of this vector. +func (b *Builder) StartVector(elemSize, numElems, alignment int) UOffsetT { + b.assertNotNested() + b.nested = true + b.Prep(SizeUint32, elemSize*numElems) + b.Prep(alignment, elemSize*numElems) // Just in case alignment > int. + return b.Offset() +} + +// EndVector writes data necessary to finish vector construction. +func (b *Builder) EndVector(vectorNumElems int) UOffsetT { + b.assertNested() + + // we already made space for this, so write without PrependUint32 + b.PlaceUOffsetT(UOffsetT(vectorNumElems)) + + b.nested = false + return b.Offset() +} + +// CreateSharedString Checks if the string is already written +// to the buffer before calling CreateString +func (b *Builder) CreateSharedString(s string) UOffsetT { + if b.sharedStrings == nil { + b.sharedStrings = make(map[string]UOffsetT) + } + if v, ok := b.sharedStrings[s]; ok { + return v + } + off := b.CreateString(s) + b.sharedStrings[s] = off + return off +} + +// CreateString writes a null-terminated string as a vector. +func (b *Builder) CreateString(s string) UOffsetT { + b.assertNotNested() + b.nested = true + + b.Prep(int(SizeUOffsetT), (len(s)+1)*SizeByte) + b.PlaceByte(0) + + l := UOffsetT(len(s)) + + b.head -= l + copy(b.Bytes[b.head:b.head+l], s) + + return b.EndVector(len(s)) +} + +// CreateByteString writes a byte slice as a string (null-terminated). +func (b *Builder) CreateByteString(s []byte) UOffsetT { + b.assertNotNested() + b.nested = true + + b.Prep(int(SizeUOffsetT), (len(s)+1)*SizeByte) + b.PlaceByte(0) + + l := UOffsetT(len(s)) + + b.head -= l + copy(b.Bytes[b.head:b.head+l], s) + + return b.EndVector(len(s)) +} + +// CreateByteVector writes a ubyte vector +func (b *Builder) CreateByteVector(v []byte) UOffsetT { + b.assertNotNested() + b.nested = true + + b.Prep(int(SizeUOffsetT), len(v)*SizeByte) + + l := UOffsetT(len(v)) + + b.head -= l + copy(b.Bytes[b.head:b.head+l], v) + + return b.EndVector(len(v)) +} + +func (b *Builder) assertNested() { + // If you get this assert, you're in an object while trying to write + // data that belongs outside of an object. + // To fix this, write non-inline data (like vectors) before creating + // objects. + if !b.nested { + panic("Incorrect creation order: must be inside object.") + } +} + +func (b *Builder) assertNotNested() { + // If you hit this, you're trying to construct a Table/Vector/String + // during the construction of its parent table (between the MyTableBuilder + // and builder.Finish()). + // Move the creation of these sub-objects to above the MyTableBuilder to + // not get this assert. + // Ignoring this assert may appear to work in simple cases, but the reason + // it is here is that storing objects in-line may cause vtable offsets + // to not fit anymore. It also leads to vtable duplication. + if b.nested { + panic("Incorrect creation order: object must not be nested.") + } +} + +func (b *Builder) assertFinished() { + // If you get this assert, you're attempting to get access a buffer + // which hasn't been finished yet. Be sure to call builder.Finish() + // with your root table. + // If you really need to access an unfinished buffer, use the Bytes + // buffer directly. + if !b.finished { + panic("Incorrect use of FinishedBytes(): must call 'Finish' first.") + } +} + +// PrependBoolSlot prepends a bool onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependBoolSlot(o int, x, d bool) { + val := byte(0) + if x { + val = 1 + } + def := byte(0) + if d { + def = 1 + } + b.PrependByteSlot(o, val, def) +} + +// PrependByteSlot prepends a byte onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependByteSlot(o int, x, d byte) { + if x != d { + b.PrependByte(x) + b.Slot(o) + } +} + +// PrependUint8Slot prepends a uint8 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependUint8Slot(o int, x, d uint8) { + if x != d { + b.PrependUint8(x) + b.Slot(o) + } +} + +// PrependUint16Slot prepends a uint16 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependUint16Slot(o int, x, d uint16) { + if x != d { + b.PrependUint16(x) + b.Slot(o) + } +} + +// PrependUint32Slot prepends a uint32 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependUint32Slot(o int, x, d uint32) { + if x != d { + b.PrependUint32(x) + b.Slot(o) + } +} + +// PrependUint64Slot prepends a uint64 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependUint64Slot(o int, x, d uint64) { + if x != d { + b.PrependUint64(x) + b.Slot(o) + } +} + +// PrependInt8Slot prepends a int8 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependInt8Slot(o int, x, d int8) { + if x != d { + b.PrependInt8(x) + b.Slot(o) + } +} + +// PrependInt16Slot prepends a int16 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependInt16Slot(o int, x, d int16) { + if x != d { + b.PrependInt16(x) + b.Slot(o) + } +} + +// PrependInt32Slot prepends a int32 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependInt32Slot(o int, x, d int32) { + if x != d { + b.PrependInt32(x) + b.Slot(o) + } +} + +// PrependInt64Slot prepends a int64 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependInt64Slot(o int, x, d int64) { + if x != d { + b.PrependInt64(x) + b.Slot(o) + } +} + +// PrependFloat32Slot prepends a float32 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependFloat32Slot(o int, x, d float32) { + if x != d { + b.PrependFloat32(x) + b.Slot(o) + } +} + +// PrependFloat64Slot prepends a float64 onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependFloat64Slot(o int, x, d float64) { + if x != d { + b.PrependFloat64(x) + b.Slot(o) + } +} + +// PrependUOffsetTSlot prepends an UOffsetT onto the object at vtable slot `o`. +// If value `x` equals default `d`, then the slot will be set to zero and no +// other data will be written. +func (b *Builder) PrependUOffsetTSlot(o int, x, d UOffsetT) { + if x != d { + b.PrependUOffsetT(x) + b.Slot(o) + } +} + +// PrependStructSlot prepends a struct onto the object at vtable slot `o`. +// Structs are stored inline, so nothing additional is being added. +// In generated code, `d` is always 0. +func (b *Builder) PrependStructSlot(voffset int, x, d UOffsetT) { + if x != d { + b.assertNested() + if x != b.Offset() { + panic("inline data write outside of object") + } + b.Slot(voffset) + } +} + +// Slot sets the vtable key `voffset` to the current location in the buffer. +func (b *Builder) Slot(slotnum int) { + b.vtable[slotnum] = UOffsetT(b.Offset()) +} + +// FinishWithFileIdentifier finalizes a buffer, pointing to the given `rootTable`. +// as well as applys a file identifier +func (b *Builder) FinishWithFileIdentifier(rootTable UOffsetT, fid []byte) { + if fid == nil || len(fid) != fileIdentifierLength { + panic("incorrect file identifier length") + } + // In order to add a file identifier to the flatbuffer message, we need + // to prepare an alignment and file identifier length + b.Prep(b.minalign, SizeInt32+fileIdentifierLength) + for i := fileIdentifierLength - 1; i >= 0; i-- { + // place the file identifier + b.PlaceByte(fid[i]) + } + // finish + b.Finish(rootTable) +} + +// Finish finalizes a buffer, pointing to the given `rootTable`. +func (b *Builder) Finish(rootTable UOffsetT) { + b.assertNotNested() + b.Prep(b.minalign, SizeUOffsetT) + b.PrependUOffsetT(rootTable) + b.finished = true +} + +// vtableEqual compares an unwritten vtable to a written vtable. +func vtableEqual(a []UOffsetT, objectStart UOffsetT, b []byte) bool { + if len(a)*SizeVOffsetT != len(b) { + return false + } + + for i := 0; i < len(a); i++ { + x := GetVOffsetT(b[i*SizeVOffsetT : (i+1)*SizeVOffsetT]) + + // Skip vtable entries that indicate a default value. + if x == 0 && a[i] == 0 { + continue + } + + y := SOffsetT(objectStart) - SOffsetT(a[i]) + if SOffsetT(x) != y { + return false + } + } + return true +} + +// PrependBool prepends a bool to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependBool(x bool) { + b.Prep(SizeBool, 0) + b.PlaceBool(x) +} + +// PrependUint8 prepends a uint8 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependUint8(x uint8) { + b.Prep(SizeUint8, 0) + b.PlaceUint8(x) +} + +// PrependUint16 prepends a uint16 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependUint16(x uint16) { + b.Prep(SizeUint16, 0) + b.PlaceUint16(x) +} + +// PrependUint32 prepends a uint32 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependUint32(x uint32) { + b.Prep(SizeUint32, 0) + b.PlaceUint32(x) +} + +// PrependUint64 prepends a uint64 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependUint64(x uint64) { + b.Prep(SizeUint64, 0) + b.PlaceUint64(x) +} + +// PrependInt8 prepends a int8 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependInt8(x int8) { + b.Prep(SizeInt8, 0) + b.PlaceInt8(x) +} + +// PrependInt16 prepends a int16 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependInt16(x int16) { + b.Prep(SizeInt16, 0) + b.PlaceInt16(x) +} + +// PrependInt32 prepends a int32 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependInt32(x int32) { + b.Prep(SizeInt32, 0) + b.PlaceInt32(x) +} + +// PrependInt64 prepends a int64 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependInt64(x int64) { + b.Prep(SizeInt64, 0) + b.PlaceInt64(x) +} + +// PrependFloat32 prepends a float32 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependFloat32(x float32) { + b.Prep(SizeFloat32, 0) + b.PlaceFloat32(x) +} + +// PrependFloat64 prepends a float64 to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependFloat64(x float64) { + b.Prep(SizeFloat64, 0) + b.PlaceFloat64(x) +} + +// PrependByte prepends a byte to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependByte(x byte) { + b.Prep(SizeByte, 0) + b.PlaceByte(x) +} + +// PrependVOffsetT prepends a VOffsetT to the Builder buffer. +// Aligns and checks for space. +func (b *Builder) PrependVOffsetT(x VOffsetT) { + b.Prep(SizeVOffsetT, 0) + b.PlaceVOffsetT(x) +} + +// PlaceBool prepends a bool to the Builder, without checking for space. +func (b *Builder) PlaceBool(x bool) { + b.head -= UOffsetT(SizeBool) + WriteBool(b.Bytes[b.head:], x) +} + +// PlaceUint8 prepends a uint8 to the Builder, without checking for space. +func (b *Builder) PlaceUint8(x uint8) { + b.head -= UOffsetT(SizeUint8) + WriteUint8(b.Bytes[b.head:], x) +} + +// PlaceUint16 prepends a uint16 to the Builder, without checking for space. +func (b *Builder) PlaceUint16(x uint16) { + b.head -= UOffsetT(SizeUint16) + WriteUint16(b.Bytes[b.head:], x) +} + +// PlaceUint32 prepends a uint32 to the Builder, without checking for space. +func (b *Builder) PlaceUint32(x uint32) { + b.head -= UOffsetT(SizeUint32) + WriteUint32(b.Bytes[b.head:], x) +} + +// PlaceUint64 prepends a uint64 to the Builder, without checking for space. +func (b *Builder) PlaceUint64(x uint64) { + b.head -= UOffsetT(SizeUint64) + WriteUint64(b.Bytes[b.head:], x) +} + +// PlaceInt8 prepends a int8 to the Builder, without checking for space. +func (b *Builder) PlaceInt8(x int8) { + b.head -= UOffsetT(SizeInt8) + WriteInt8(b.Bytes[b.head:], x) +} + +// PlaceInt16 prepends a int16 to the Builder, without checking for space. +func (b *Builder) PlaceInt16(x int16) { + b.head -= UOffsetT(SizeInt16) + WriteInt16(b.Bytes[b.head:], x) +} + +// PlaceInt32 prepends a int32 to the Builder, without checking for space. +func (b *Builder) PlaceInt32(x int32) { + b.head -= UOffsetT(SizeInt32) + WriteInt32(b.Bytes[b.head:], x) +} + +// PlaceInt64 prepends a int64 to the Builder, without checking for space. +func (b *Builder) PlaceInt64(x int64) { + b.head -= UOffsetT(SizeInt64) + WriteInt64(b.Bytes[b.head:], x) +} + +// PlaceFloat32 prepends a float32 to the Builder, without checking for space. +func (b *Builder) PlaceFloat32(x float32) { + b.head -= UOffsetT(SizeFloat32) + WriteFloat32(b.Bytes[b.head:], x) +} + +// PlaceFloat64 prepends a float64 to the Builder, without checking for space. +func (b *Builder) PlaceFloat64(x float64) { + b.head -= UOffsetT(SizeFloat64) + WriteFloat64(b.Bytes[b.head:], x) +} + +// PlaceByte prepends a byte to the Builder, without checking for space. +func (b *Builder) PlaceByte(x byte) { + b.head -= UOffsetT(SizeByte) + WriteByte(b.Bytes[b.head:], x) +} + +// PlaceVOffsetT prepends a VOffsetT to the Builder, without checking for space. +func (b *Builder) PlaceVOffsetT(x VOffsetT) { + b.head -= UOffsetT(SizeVOffsetT) + WriteVOffsetT(b.Bytes[b.head:], x) +} + +// PlaceSOffsetT prepends a SOffsetT to the Builder, without checking for space. +func (b *Builder) PlaceSOffsetT(x SOffsetT) { + b.head -= UOffsetT(SizeSOffsetT) + WriteSOffsetT(b.Bytes[b.head:], x) +} + +// PlaceUOffsetT prepends a UOffsetT to the Builder, without checking for space. +func (b *Builder) PlaceUOffsetT(x UOffsetT) { + b.head -= UOffsetT(SizeUOffsetT) + WriteUOffsetT(b.Bytes[b.head:], x) +} diff --git a/vendor/github.com/google/flatbuffers/go/doc.go b/vendor/github.com/google/flatbuffers/go/doc.go new file mode 100644 index 00000000000..694edc763d8 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/doc.go @@ -0,0 +1,3 @@ +// Package flatbuffers provides facilities to read and write flatbuffers +// objects. +package flatbuffers diff --git a/vendor/github.com/google/flatbuffers/go/encode.go b/vendor/github.com/google/flatbuffers/go/encode.go new file mode 100644 index 00000000000..72d4f3a1521 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/encode.go @@ -0,0 +1,238 @@ +package flatbuffers + +import ( + "math" +) + +type ( + // A SOffsetT stores a signed offset into arbitrary data. + SOffsetT int32 + // A UOffsetT stores an unsigned offset into vector data. + UOffsetT uint32 + // A VOffsetT stores an unsigned offset in a vtable. + VOffsetT uint16 +) + +const ( + // VtableMetadataFields is the count of metadata fields in each vtable. + VtableMetadataFields = 2 +) + +// GetByte decodes a little-endian byte from a byte slice. +func GetByte(buf []byte) byte { + return byte(GetUint8(buf)) +} + +// GetBool decodes a little-endian bool from a byte slice. +func GetBool(buf []byte) bool { + return buf[0] == 1 +} + +// GetUint8 decodes a little-endian uint8 from a byte slice. +func GetUint8(buf []byte) (n uint8) { + n = uint8(buf[0]) + return +} + +// GetUint16 decodes a little-endian uint16 from a byte slice. +func GetUint16(buf []byte) (n uint16) { + _ = buf[1] // Force one bounds check. See: golang.org/issue/14808 + n |= uint16(buf[0]) + n |= uint16(buf[1]) << 8 + return +} + +// GetUint32 decodes a little-endian uint32 from a byte slice. +func GetUint32(buf []byte) (n uint32) { + _ = buf[3] // Force one bounds check. See: golang.org/issue/14808 + n |= uint32(buf[0]) + n |= uint32(buf[1]) << 8 + n |= uint32(buf[2]) << 16 + n |= uint32(buf[3]) << 24 + return +} + +// GetUint64 decodes a little-endian uint64 from a byte slice. +func GetUint64(buf []byte) (n uint64) { + _ = buf[7] // Force one bounds check. See: golang.org/issue/14808 + n |= uint64(buf[0]) + n |= uint64(buf[1]) << 8 + n |= uint64(buf[2]) << 16 + n |= uint64(buf[3]) << 24 + n |= uint64(buf[4]) << 32 + n |= uint64(buf[5]) << 40 + n |= uint64(buf[6]) << 48 + n |= uint64(buf[7]) << 56 + return +} + +// GetInt8 decodes a little-endian int8 from a byte slice. +func GetInt8(buf []byte) (n int8) { + n = int8(buf[0]) + return +} + +// GetInt16 decodes a little-endian int16 from a byte slice. +func GetInt16(buf []byte) (n int16) { + _ = buf[1] // Force one bounds check. See: golang.org/issue/14808 + n |= int16(buf[0]) + n |= int16(buf[1]) << 8 + return +} + +// GetInt32 decodes a little-endian int32 from a byte slice. +func GetInt32(buf []byte) (n int32) { + _ = buf[3] // Force one bounds check. See: golang.org/issue/14808 + n |= int32(buf[0]) + n |= int32(buf[1]) << 8 + n |= int32(buf[2]) << 16 + n |= int32(buf[3]) << 24 + return +} + +// GetInt64 decodes a little-endian int64 from a byte slice. +func GetInt64(buf []byte) (n int64) { + _ = buf[7] // Force one bounds check. See: golang.org/issue/14808 + n |= int64(buf[0]) + n |= int64(buf[1]) << 8 + n |= int64(buf[2]) << 16 + n |= int64(buf[3]) << 24 + n |= int64(buf[4]) << 32 + n |= int64(buf[5]) << 40 + n |= int64(buf[6]) << 48 + n |= int64(buf[7]) << 56 + return +} + +// GetFloat32 decodes a little-endian float32 from a byte slice. +func GetFloat32(buf []byte) float32 { + x := GetUint32(buf) + return math.Float32frombits(x) +} + +// GetFloat64 decodes a little-endian float64 from a byte slice. +func GetFloat64(buf []byte) float64 { + x := GetUint64(buf) + return math.Float64frombits(x) +} + +// GetUOffsetT decodes a little-endian UOffsetT from a byte slice. +func GetUOffsetT(buf []byte) UOffsetT { + return UOffsetT(GetInt32(buf)) +} + +// GetSOffsetT decodes a little-endian SOffsetT from a byte slice. +func GetSOffsetT(buf []byte) SOffsetT { + return SOffsetT(GetInt32(buf)) +} + +// GetVOffsetT decodes a little-endian VOffsetT from a byte slice. +func GetVOffsetT(buf []byte) VOffsetT { + return VOffsetT(GetUint16(buf)) +} + +// WriteByte encodes a little-endian uint8 into a byte slice. +func WriteByte(buf []byte, n byte) { + WriteUint8(buf, uint8(n)) +} + +// WriteBool encodes a little-endian bool into a byte slice. +func WriteBool(buf []byte, b bool) { + buf[0] = 0 + if b { + buf[0] = 1 + } +} + +// WriteUint8 encodes a little-endian uint8 into a byte slice. +func WriteUint8(buf []byte, n uint8) { + buf[0] = byte(n) +} + +// WriteUint16 encodes a little-endian uint16 into a byte slice. +func WriteUint16(buf []byte, n uint16) { + _ = buf[1] // Force one bounds check. See: golang.org/issue/14808 + buf[0] = byte(n) + buf[1] = byte(n >> 8) +} + +// WriteUint32 encodes a little-endian uint32 into a byte slice. +func WriteUint32(buf []byte, n uint32) { + _ = buf[3] // Force one bounds check. See: golang.org/issue/14808 + buf[0] = byte(n) + buf[1] = byte(n >> 8) + buf[2] = byte(n >> 16) + buf[3] = byte(n >> 24) +} + +// WriteUint64 encodes a little-endian uint64 into a byte slice. +func WriteUint64(buf []byte, n uint64) { + _ = buf[7] // Force one bounds check. See: golang.org/issue/14808 + buf[0] = byte(n) + buf[1] = byte(n >> 8) + buf[2] = byte(n >> 16) + buf[3] = byte(n >> 24) + buf[4] = byte(n >> 32) + buf[5] = byte(n >> 40) + buf[6] = byte(n >> 48) + buf[7] = byte(n >> 56) +} + +// WriteInt8 encodes a little-endian int8 into a byte slice. +func WriteInt8(buf []byte, n int8) { + buf[0] = byte(n) +} + +// WriteInt16 encodes a little-endian int16 into a byte slice. +func WriteInt16(buf []byte, n int16) { + _ = buf[1] // Force one bounds check. See: golang.org/issue/14808 + buf[0] = byte(n) + buf[1] = byte(n >> 8) +} + +// WriteInt32 encodes a little-endian int32 into a byte slice. +func WriteInt32(buf []byte, n int32) { + _ = buf[3] // Force one bounds check. See: golang.org/issue/14808 + buf[0] = byte(n) + buf[1] = byte(n >> 8) + buf[2] = byte(n >> 16) + buf[3] = byte(n >> 24) +} + +// WriteInt64 encodes a little-endian int64 into a byte slice. +func WriteInt64(buf []byte, n int64) { + _ = buf[7] // Force one bounds check. See: golang.org/issue/14808 + buf[0] = byte(n) + buf[1] = byte(n >> 8) + buf[2] = byte(n >> 16) + buf[3] = byte(n >> 24) + buf[4] = byte(n >> 32) + buf[5] = byte(n >> 40) + buf[6] = byte(n >> 48) + buf[7] = byte(n >> 56) +} + +// WriteFloat32 encodes a little-endian float32 into a byte slice. +func WriteFloat32(buf []byte, n float32) { + WriteUint32(buf, math.Float32bits(n)) +} + +// WriteFloat64 encodes a little-endian float64 into a byte slice. +func WriteFloat64(buf []byte, n float64) { + WriteUint64(buf, math.Float64bits(n)) +} + +// WriteVOffsetT encodes a little-endian VOffsetT into a byte slice. +func WriteVOffsetT(buf []byte, n VOffsetT) { + WriteUint16(buf, uint16(n)) +} + +// WriteSOffsetT encodes a little-endian SOffsetT into a byte slice. +func WriteSOffsetT(buf []byte, n SOffsetT) { + WriteInt32(buf, int32(n)) +} + +// WriteUOffsetT encodes a little-endian UOffsetT into a byte slice. +func WriteUOffsetT(buf []byte, n UOffsetT) { + WriteUint32(buf, uint32(n)) +} diff --git a/vendor/github.com/google/flatbuffers/go/grpc.go b/vendor/github.com/google/flatbuffers/go/grpc.go new file mode 100644 index 00000000000..15f1a510d3b --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/grpc.go @@ -0,0 +1,38 @@ +package flatbuffers + +// Codec implements gRPC-go Codec which is used to encode and decode messages. +var Codec = "flatbuffers" + +// FlatbuffersCodec defines the interface gRPC uses to encode and decode messages. Note +// that implementations of this interface must be thread safe; a Codec's +// methods can be called from concurrent goroutines. +type FlatbuffersCodec struct{} + +// Marshal returns the wire format of v. +func (FlatbuffersCodec) Marshal(v interface{}) ([]byte, error) { + return v.(*Builder).FinishedBytes(), nil +} + +// Unmarshal parses the wire format into v. +func (FlatbuffersCodec) Unmarshal(data []byte, v interface{}) error { + v.(flatbuffersInit).Init(data, GetUOffsetT(data)) + return nil +} + +// String old gRPC Codec interface func +func (FlatbuffersCodec) String() string { + return Codec +} + +// Name returns the name of the Codec implementation. The returned string +// will be used as part of content type in transmission. The result must be +// static; the result cannot change between calls. +// +// add Name() for ForceCodec interface +func (FlatbuffersCodec) Name() string { + return Codec +} + +type flatbuffersInit interface { + Init(data []byte, i UOffsetT) +} diff --git a/vendor/github.com/google/flatbuffers/go/lib.go b/vendor/github.com/google/flatbuffers/go/lib.go new file mode 100644 index 00000000000..adfce52efe5 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/lib.go @@ -0,0 +1,13 @@ +package flatbuffers + +// FlatBuffer is the interface that represents a flatbuffer. +type FlatBuffer interface { + Table() Table + Init(buf []byte, i UOffsetT) +} + +// GetRootAs is a generic helper to initialize a FlatBuffer with the provided buffer bytes and its data offset. +func GetRootAs(buf []byte, offset UOffsetT, fb FlatBuffer) { + n := GetUOffsetT(buf[offset:]) + fb.Init(buf, n+offset) +} diff --git a/vendor/github.com/google/flatbuffers/go/sizes.go b/vendor/github.com/google/flatbuffers/go/sizes.go new file mode 100644 index 00000000000..ba221698455 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/sizes.go @@ -0,0 +1,55 @@ +package flatbuffers + +import ( + "unsafe" +) + +const ( + // See http://golang.org/ref/spec#Numeric_types + + // SizeUint8 is the byte size of a uint8. + SizeUint8 = 1 + // SizeUint16 is the byte size of a uint16. + SizeUint16 = 2 + // SizeUint32 is the byte size of a uint32. + SizeUint32 = 4 + // SizeUint64 is the byte size of a uint64. + SizeUint64 = 8 + + // SizeInt8 is the byte size of a int8. + SizeInt8 = 1 + // SizeInt16 is the byte size of a int16. + SizeInt16 = 2 + // SizeInt32 is the byte size of a int32. + SizeInt32 = 4 + // SizeInt64 is the byte size of a int64. + SizeInt64 = 8 + + // SizeFloat32 is the byte size of a float32. + SizeFloat32 = 4 + // SizeFloat64 is the byte size of a float64. + SizeFloat64 = 8 + + // SizeByte is the byte size of a byte. + // The `byte` type is aliased (by Go definition) to uint8. + SizeByte = 1 + + // SizeBool is the byte size of a bool. + // The `bool` type is aliased (by flatbuffers convention) to uint8. + SizeBool = 1 + + // SizeSOffsetT is the byte size of an SOffsetT. + // The `SOffsetT` type is aliased (by flatbuffers convention) to int32. + SizeSOffsetT = 4 + // SizeUOffsetT is the byte size of an UOffsetT. + // The `UOffsetT` type is aliased (by flatbuffers convention) to uint32. + SizeUOffsetT = 4 + // SizeVOffsetT is the byte size of an VOffsetT. + // The `VOffsetT` type is aliased (by flatbuffers convention) to uint16. + SizeVOffsetT = 2 +) + +// byteSliceToString converts a []byte to string without a heap allocation. +func byteSliceToString(b []byte) string { + return *(*string)(unsafe.Pointer(&b)) +} diff --git a/vendor/github.com/google/flatbuffers/go/struct.go b/vendor/github.com/google/flatbuffers/go/struct.go new file mode 100644 index 00000000000..11258f715d4 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/struct.go @@ -0,0 +1,8 @@ +package flatbuffers + +// Struct wraps a byte slice and provides read access to its data. +// +// Structs do not have a vtable. +type Struct struct { + Table +} diff --git a/vendor/github.com/google/flatbuffers/go/table.go b/vendor/github.com/google/flatbuffers/go/table.go new file mode 100644 index 00000000000..b273146fad4 --- /dev/null +++ b/vendor/github.com/google/flatbuffers/go/table.go @@ -0,0 +1,505 @@ +package flatbuffers + +// Table wraps a byte slice and provides read access to its data. +// +// The variable `Pos` indicates the root of the FlatBuffers object therein. +type Table struct { + Bytes []byte + Pos UOffsetT // Always < 1<<31. +} + +// Offset provides access into the Table's vtable. +// +// Fields which are deprecated are ignored by checking against the vtable's length. +func (t *Table) Offset(vtableOffset VOffsetT) VOffsetT { + vtable := UOffsetT(SOffsetT(t.Pos) - t.GetSOffsetT(t.Pos)) + if vtableOffset < t.GetVOffsetT(vtable) { + return t.GetVOffsetT(vtable + UOffsetT(vtableOffset)) + } + return 0 +} + +// Indirect retrieves the relative offset stored at `offset`. +func (t *Table) Indirect(off UOffsetT) UOffsetT { + return off + GetUOffsetT(t.Bytes[off:]) +} + +// String gets a string from data stored inside the flatbuffer. +func (t *Table) String(off UOffsetT) string { + b := t.ByteVector(off) + return byteSliceToString(b) +} + +// ByteVector gets a byte slice from data stored inside the flatbuffer. +func (t *Table) ByteVector(off UOffsetT) []byte { + off += GetUOffsetT(t.Bytes[off:]) + start := off + UOffsetT(SizeUOffsetT) + length := GetUOffsetT(t.Bytes[off:]) + return t.Bytes[start : start+length] +} + +// VectorLen retrieves the length of the vector whose offset is stored at +// "off" in this object. +func (t *Table) VectorLen(off UOffsetT) int { + off += t.Pos + off += GetUOffsetT(t.Bytes[off:]) + return int(GetUOffsetT(t.Bytes[off:])) +} + +// Vector retrieves the start of data of the vector whose offset is stored +// at "off" in this object. +func (t *Table) Vector(off UOffsetT) UOffsetT { + off += t.Pos + x := off + GetUOffsetT(t.Bytes[off:]) + // data starts after metadata containing the vector length + x += UOffsetT(SizeUOffsetT) + return x +} + +// Union initializes any Table-derived type to point to the union at the given +// offset. +func (t *Table) Union(t2 *Table, off UOffsetT) { + off += t.Pos + t2.Pos = off + t.GetUOffsetT(off) + t2.Bytes = t.Bytes +} + +// GetBool retrieves a bool at the given offset. +func (t *Table) GetBool(off UOffsetT) bool { + return GetBool(t.Bytes[off:]) +} + +// GetByte retrieves a byte at the given offset. +func (t *Table) GetByte(off UOffsetT) byte { + return GetByte(t.Bytes[off:]) +} + +// GetUint8 retrieves a uint8 at the given offset. +func (t *Table) GetUint8(off UOffsetT) uint8 { + return GetUint8(t.Bytes[off:]) +} + +// GetUint16 retrieves a uint16 at the given offset. +func (t *Table) GetUint16(off UOffsetT) uint16 { + return GetUint16(t.Bytes[off:]) +} + +// GetUint32 retrieves a uint32 at the given offset. +func (t *Table) GetUint32(off UOffsetT) uint32 { + return GetUint32(t.Bytes[off:]) +} + +// GetUint64 retrieves a uint64 at the given offset. +func (t *Table) GetUint64(off UOffsetT) uint64 { + return GetUint64(t.Bytes[off:]) +} + +// GetInt8 retrieves a int8 at the given offset. +func (t *Table) GetInt8(off UOffsetT) int8 { + return GetInt8(t.Bytes[off:]) +} + +// GetInt16 retrieves a int16 at the given offset. +func (t *Table) GetInt16(off UOffsetT) int16 { + return GetInt16(t.Bytes[off:]) +} + +// GetInt32 retrieves a int32 at the given offset. +func (t *Table) GetInt32(off UOffsetT) int32 { + return GetInt32(t.Bytes[off:]) +} + +// GetInt64 retrieves a int64 at the given offset. +func (t *Table) GetInt64(off UOffsetT) int64 { + return GetInt64(t.Bytes[off:]) +} + +// GetFloat32 retrieves a float32 at the given offset. +func (t *Table) GetFloat32(off UOffsetT) float32 { + return GetFloat32(t.Bytes[off:]) +} + +// GetFloat64 retrieves a float64 at the given offset. +func (t *Table) GetFloat64(off UOffsetT) float64 { + return GetFloat64(t.Bytes[off:]) +} + +// GetUOffsetT retrieves a UOffsetT at the given offset. +func (t *Table) GetUOffsetT(off UOffsetT) UOffsetT { + return GetUOffsetT(t.Bytes[off:]) +} + +// GetVOffsetT retrieves a VOffsetT at the given offset. +func (t *Table) GetVOffsetT(off UOffsetT) VOffsetT { + return GetVOffsetT(t.Bytes[off:]) +} + +// GetSOffsetT retrieves a SOffsetT at the given offset. +func (t *Table) GetSOffsetT(off UOffsetT) SOffsetT { + return GetSOffsetT(t.Bytes[off:]) +} + +// GetBoolSlot retrieves the bool that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetBoolSlot(slot VOffsetT, d bool) bool { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetBool(t.Pos + UOffsetT(off)) +} + +// GetByteSlot retrieves the byte that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetByteSlot(slot VOffsetT, d byte) byte { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetByte(t.Pos + UOffsetT(off)) +} + +// GetInt8Slot retrieves the int8 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetInt8Slot(slot VOffsetT, d int8) int8 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetInt8(t.Pos + UOffsetT(off)) +} + +// GetUint8Slot retrieves the uint8 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetUint8Slot(slot VOffsetT, d uint8) uint8 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetUint8(t.Pos + UOffsetT(off)) +} + +// GetInt16Slot retrieves the int16 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetInt16Slot(slot VOffsetT, d int16) int16 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetInt16(t.Pos + UOffsetT(off)) +} + +// GetUint16Slot retrieves the uint16 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetUint16Slot(slot VOffsetT, d uint16) uint16 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetUint16(t.Pos + UOffsetT(off)) +} + +// GetInt32Slot retrieves the int32 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetInt32Slot(slot VOffsetT, d int32) int32 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetInt32(t.Pos + UOffsetT(off)) +} + +// GetUint32Slot retrieves the uint32 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetUint32Slot(slot VOffsetT, d uint32) uint32 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetUint32(t.Pos + UOffsetT(off)) +} + +// GetInt64Slot retrieves the int64 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetInt64Slot(slot VOffsetT, d int64) int64 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetInt64(t.Pos + UOffsetT(off)) +} + +// GetUint64Slot retrieves the uint64 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetUint64Slot(slot VOffsetT, d uint64) uint64 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetUint64(t.Pos + UOffsetT(off)) +} + +// GetFloat32Slot retrieves the float32 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetFloat32Slot(slot VOffsetT, d float32) float32 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetFloat32(t.Pos + UOffsetT(off)) +} + +// GetFloat64Slot retrieves the float64 that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetFloat64Slot(slot VOffsetT, d float64) float64 { + off := t.Offset(slot) + if off == 0 { + return d + } + + return t.GetFloat64(t.Pos + UOffsetT(off)) +} + +// GetVOffsetTSlot retrieves the VOffsetT that the given vtable location +// points to. If the vtable value is zero, the default value `d` +// will be returned. +func (t *Table) GetVOffsetTSlot(slot VOffsetT, d VOffsetT) VOffsetT { + off := t.Offset(slot) + if off == 0 { + return d + } + return VOffsetT(off) +} + +// MutateBool updates a bool at the given offset. +func (t *Table) MutateBool(off UOffsetT, n bool) bool { + WriteBool(t.Bytes[off:], n) + return true +} + +// MutateByte updates a Byte at the given offset. +func (t *Table) MutateByte(off UOffsetT, n byte) bool { + WriteByte(t.Bytes[off:], n) + return true +} + +// MutateUint8 updates a Uint8 at the given offset. +func (t *Table) MutateUint8(off UOffsetT, n uint8) bool { + WriteUint8(t.Bytes[off:], n) + return true +} + +// MutateUint16 updates a Uint16 at the given offset. +func (t *Table) MutateUint16(off UOffsetT, n uint16) bool { + WriteUint16(t.Bytes[off:], n) + return true +} + +// MutateUint32 updates a Uint32 at the given offset. +func (t *Table) MutateUint32(off UOffsetT, n uint32) bool { + WriteUint32(t.Bytes[off:], n) + return true +} + +// MutateUint64 updates a Uint64 at the given offset. +func (t *Table) MutateUint64(off UOffsetT, n uint64) bool { + WriteUint64(t.Bytes[off:], n) + return true +} + +// MutateInt8 updates a Int8 at the given offset. +func (t *Table) MutateInt8(off UOffsetT, n int8) bool { + WriteInt8(t.Bytes[off:], n) + return true +} + +// MutateInt16 updates a Int16 at the given offset. +func (t *Table) MutateInt16(off UOffsetT, n int16) bool { + WriteInt16(t.Bytes[off:], n) + return true +} + +// MutateInt32 updates a Int32 at the given offset. +func (t *Table) MutateInt32(off UOffsetT, n int32) bool { + WriteInt32(t.Bytes[off:], n) + return true +} + +// MutateInt64 updates a Int64 at the given offset. +func (t *Table) MutateInt64(off UOffsetT, n int64) bool { + WriteInt64(t.Bytes[off:], n) + return true +} + +// MutateFloat32 updates a Float32 at the given offset. +func (t *Table) MutateFloat32(off UOffsetT, n float32) bool { + WriteFloat32(t.Bytes[off:], n) + return true +} + +// MutateFloat64 updates a Float64 at the given offset. +func (t *Table) MutateFloat64(off UOffsetT, n float64) bool { + WriteFloat64(t.Bytes[off:], n) + return true +} + +// MutateUOffsetT updates a UOffsetT at the given offset. +func (t *Table) MutateUOffsetT(off UOffsetT, n UOffsetT) bool { + WriteUOffsetT(t.Bytes[off:], n) + return true +} + +// MutateVOffsetT updates a VOffsetT at the given offset. +func (t *Table) MutateVOffsetT(off UOffsetT, n VOffsetT) bool { + WriteVOffsetT(t.Bytes[off:], n) + return true +} + +// MutateSOffsetT updates a SOffsetT at the given offset. +func (t *Table) MutateSOffsetT(off UOffsetT, n SOffsetT) bool { + WriteSOffsetT(t.Bytes[off:], n) + return true +} + +// MutateBoolSlot updates the bool at given vtable location +func (t *Table) MutateBoolSlot(slot VOffsetT, n bool) bool { + if off := t.Offset(slot); off != 0 { + t.MutateBool(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateByteSlot updates the byte at given vtable location +func (t *Table) MutateByteSlot(slot VOffsetT, n byte) bool { + if off := t.Offset(slot); off != 0 { + t.MutateByte(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateInt8Slot updates the int8 at given vtable location +func (t *Table) MutateInt8Slot(slot VOffsetT, n int8) bool { + if off := t.Offset(slot); off != 0 { + t.MutateInt8(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateUint8Slot updates the uint8 at given vtable location +func (t *Table) MutateUint8Slot(slot VOffsetT, n uint8) bool { + if off := t.Offset(slot); off != 0 { + t.MutateUint8(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateInt16Slot updates the int16 at given vtable location +func (t *Table) MutateInt16Slot(slot VOffsetT, n int16) bool { + if off := t.Offset(slot); off != 0 { + t.MutateInt16(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateUint16Slot updates the uint16 at given vtable location +func (t *Table) MutateUint16Slot(slot VOffsetT, n uint16) bool { + if off := t.Offset(slot); off != 0 { + t.MutateUint16(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateInt32Slot updates the int32 at given vtable location +func (t *Table) MutateInt32Slot(slot VOffsetT, n int32) bool { + if off := t.Offset(slot); off != 0 { + t.MutateInt32(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateUint32Slot updates the uint32 at given vtable location +func (t *Table) MutateUint32Slot(slot VOffsetT, n uint32) bool { + if off := t.Offset(slot); off != 0 { + t.MutateUint32(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateInt64Slot updates the int64 at given vtable location +func (t *Table) MutateInt64Slot(slot VOffsetT, n int64) bool { + if off := t.Offset(slot); off != 0 { + t.MutateInt64(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateUint64Slot updates the uint64 at given vtable location +func (t *Table) MutateUint64Slot(slot VOffsetT, n uint64) bool { + if off := t.Offset(slot); off != 0 { + t.MutateUint64(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateFloat32Slot updates the float32 at given vtable location +func (t *Table) MutateFloat32Slot(slot VOffsetT, n float32) bool { + if off := t.Offset(slot); off != 0 { + t.MutateFloat32(t.Pos+UOffsetT(off), n) + return true + } + + return false +} + +// MutateFloat64Slot updates the float64 at given vtable location +func (t *Table) MutateFloat64Slot(slot VOffsetT, n float64) bool { + if off := t.Offset(slot); off != 0 { + t.MutateFloat64(t.Pos+UOffsetT(off), n) + return true + } + + return false +} diff --git a/vendor/go.opencensus.io/.gitignore b/vendor/go.opencensus.io/.gitignore new file mode 100644 index 00000000000..74a6db472e5 --- /dev/null +++ b/vendor/go.opencensus.io/.gitignore @@ -0,0 +1,9 @@ +/.idea/ + +# go.opencensus.io/exporter/aws +/exporter/aws/ + +# Exclude vendor, use dep ensure after checkout: +/vendor/github.com/ +/vendor/golang.org/ +/vendor/google.golang.org/ diff --git a/vendor/go.opencensus.io/AUTHORS b/vendor/go.opencensus.io/AUTHORS new file mode 100644 index 00000000000..e491a9e7f78 --- /dev/null +++ b/vendor/go.opencensus.io/AUTHORS @@ -0,0 +1 @@ +Google Inc. diff --git a/vendor/go.opencensus.io/CONTRIBUTING.md b/vendor/go.opencensus.io/CONTRIBUTING.md new file mode 100644 index 00000000000..1ba3962c8bf --- /dev/null +++ b/vendor/go.opencensus.io/CONTRIBUTING.md @@ -0,0 +1,63 @@ +# How to contribute + +We'd love to accept your patches and contributions to this project. There are +just a few small guidelines you need to follow. + +## Contributor License Agreement + +Contributions to this project must be accompanied by a Contributor License +Agreement. You (or your employer) retain the copyright to your contribution, +this simply gives us permission to use and redistribute your contributions as +part of the project. Head over to to see +your current agreements on file or to sign a new one. + +You generally only need to submit a CLA once, so if you've already submitted one +(even if it was for a different project), you probably don't need to do it +again. + +## Code reviews + +All submissions, including submissions by project members, require review. We +use GitHub pull requests for this purpose. Consult [GitHub Help] for more +information on using pull requests. + +[GitHub Help]: https://help.github.com/articles/about-pull-requests/ + +## Instructions + +Fork the repo, checkout the upstream repo to your GOPATH by: + +``` +$ go get -d go.opencensus.io +``` + +Add your fork as an origin: + +``` +cd $(go env GOPATH)/src/go.opencensus.io +git remote add fork git@github.com:YOUR_GITHUB_USERNAME/opencensus-go.git +``` + +Run tests: + +``` +$ make install-tools # Only first time. +$ make +``` + +Checkout a new branch, make modifications and push the branch to your fork: + +``` +$ git checkout -b feature +# edit files +$ git commit +$ git push fork feature +``` + +Open a pull request against the main opencensus-go repo. + +## General Notes +This project uses Appveyor and Travis for CI. + +The dependencies are managed with `go mod` if you work with the sources under your +`$GOPATH` you need to set the environment variable `GO111MODULE=on`. \ No newline at end of file diff --git a/vendor/go.opencensus.io/LICENSE b/vendor/go.opencensus.io/LICENSE new file mode 100644 index 00000000000..7a4a3ea2424 --- /dev/null +++ b/vendor/go.opencensus.io/LICENSE @@ -0,0 +1,202 @@ + + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. \ No newline at end of file diff --git a/vendor/go.opencensus.io/Makefile b/vendor/go.opencensus.io/Makefile new file mode 100644 index 00000000000..d896edc9968 --- /dev/null +++ b/vendor/go.opencensus.io/Makefile @@ -0,0 +1,97 @@ +# TODO: Fix this on windows. +ALL_SRC := $(shell find . -name '*.go' \ + -not -path './vendor/*' \ + -not -path '*/gen-go/*' \ + -type f | sort) +ALL_PKGS := $(shell go list $(sort $(dir $(ALL_SRC)))) + +GOTEST_OPT?=-v -race -timeout 30s +GOTEST_OPT_WITH_COVERAGE = $(GOTEST_OPT) -coverprofile=coverage.txt -covermode=atomic +GOTEST=go test +GOIMPORTS=goimports +GOLINT=golint +GOVET=go vet +EMBEDMD=embedmd +# TODO decide if we need to change these names. +TRACE_ID_LINT_EXCEPTION="type name will be used as trace.TraceID by other packages" +TRACE_OPTION_LINT_EXCEPTION="type name will be used as trace.TraceOptions by other packages" +README_FILES := $(shell find . -name '*README.md' | sort | tr '\n' ' ') + +.DEFAULT_GOAL := imports-lint-vet-embedmd-test + +.PHONY: imports-lint-vet-embedmd-test +imports-lint-vet-embedmd-test: imports lint vet embedmd test + +# TODO enable test-with-coverage in tavis +.PHONY: travis-ci +travis-ci: imports lint vet embedmd test test-386 + +all-pkgs: + @echo $(ALL_PKGS) | tr ' ' '\n' | sort + +all-srcs: + @echo $(ALL_SRC) | tr ' ' '\n' | sort + +.PHONY: test +test: + $(GOTEST) $(GOTEST_OPT) $(ALL_PKGS) + +.PHONY: test-386 +test-386: + GOARCH=386 $(GOTEST) -v -timeout 30s $(ALL_PKGS) + +.PHONY: test-with-coverage +test-with-coverage: + $(GOTEST) $(GOTEST_OPT_WITH_COVERAGE) $(ALL_PKGS) + +.PHONY: imports +imports: + @IMPORTSOUT=`$(GOIMPORTS) -l $(ALL_SRC) 2>&1`; \ + if [ "$$IMPORTSOUT" ]; then \ + echo "$(GOIMPORTS) FAILED => goimports the following files:\n"; \ + echo "$$IMPORTSOUT\n"; \ + exit 1; \ + else \ + echo "Imports finished successfully"; \ + fi + +.PHONY: lint +lint: + @LINTOUT=`$(GOLINT) $(ALL_PKGS) | grep -v $(TRACE_ID_LINT_EXCEPTION) | grep -v $(TRACE_OPTION_LINT_EXCEPTION) 2>&1`; \ + if [ "$$LINTOUT" ]; then \ + echo "$(GOLINT) FAILED => clean the following lint errors:\n"; \ + echo "$$LINTOUT\n"; \ + exit 1; \ + else \ + echo "Lint finished successfully"; \ + fi + +.PHONY: vet +vet: + # TODO: Understand why go vet downloads "github.com/google/go-cmp v0.2.0" + @VETOUT=`$(GOVET) ./... | grep -v "go: downloading" 2>&1`; \ + if [ "$$VETOUT" ]; then \ + echo "$(GOVET) FAILED => go vet the following files:\n"; \ + echo "$$VETOUT\n"; \ + exit 1; \ + else \ + echo "Vet finished successfully"; \ + fi + +.PHONY: embedmd +embedmd: + @EMBEDMDOUT=`$(EMBEDMD) -d $(README_FILES) 2>&1`; \ + if [ "$$EMBEDMDOUT" ]; then \ + echo "$(EMBEDMD) FAILED => embedmd the following files:\n"; \ + echo "$$EMBEDMDOUT\n"; \ + exit 1; \ + else \ + echo "Embedmd finished successfully"; \ + fi + +.PHONY: install-tools +install-tools: + go install golang.org/x/lint/golint@latest + go install golang.org/x/tools/cmd/cover@latest + go install golang.org/x/tools/cmd/goimports@latest + go install github.com/rakyll/embedmd@latest diff --git a/vendor/go.opencensus.io/README.md b/vendor/go.opencensus.io/README.md new file mode 100644 index 00000000000..1d7e837116f --- /dev/null +++ b/vendor/go.opencensus.io/README.md @@ -0,0 +1,267 @@ +# OpenCensus Libraries for Go + +[![Build Status][travis-image]][travis-url] +[![Windows Build Status][appveyor-image]][appveyor-url] +[![GoDoc][godoc-image]][godoc-url] +[![Gitter chat][gitter-image]][gitter-url] + +OpenCensus Go is a Go implementation of OpenCensus, a toolkit for +collecting application performance and behavior monitoring data. +Currently it consists of three major components: tags, stats and tracing. + +#### OpenCensus and OpenTracing have merged to form OpenTelemetry, which serves as the next major version of OpenCensus and OpenTracing. OpenTelemetry will offer backwards compatibility with existing OpenCensus integrations, and we will continue to make security patches to existing OpenCensus libraries for two years. Read more about the merger [here](https://medium.com/opentracing/a-roadmap-to-convergence-b074e5815289). + +## Installation + +``` +$ go get -u go.opencensus.io +``` + +The API of this project is still evolving, see: [Deprecation Policy](#deprecation-policy). +The use of vendoring or a dependency management tool is recommended. + +## Prerequisites + +OpenCensus Go libraries require Go 1.8 or later. + +## Getting Started + +The easiest way to get started using OpenCensus in your application is to use an existing +integration with your RPC framework: + +* [net/http](https://godoc.org/go.opencensus.io/plugin/ochttp) +* [gRPC](https://godoc.org/go.opencensus.io/plugin/ocgrpc) +* [database/sql](https://godoc.org/github.com/opencensus-integrations/ocsql) +* [Go kit](https://godoc.org/github.com/go-kit/kit/tracing/opencensus) +* [Groupcache](https://godoc.org/github.com/orijtech/groupcache) +* [Caddy webserver](https://godoc.org/github.com/orijtech/caddy) +* [MongoDB](https://godoc.org/github.com/orijtech/mongo-go-driver) +* [Redis gomodule/redigo](https://godoc.org/github.com/orijtech/redigo) +* [Redis goredis/redis](https://godoc.org/github.com/orijtech/redis) +* [Memcache](https://godoc.org/github.com/orijtech/gomemcache) + +If you're using a framework not listed here, you could either implement your own middleware for your +framework or use [custom stats](#stats) and [spans](#spans) directly in your application. + +## Exporters + +OpenCensus can export instrumentation data to various backends. +OpenCensus has exporter implementations for the following, users +can implement their own exporters by implementing the exporter interfaces +([stats](https://godoc.org/go.opencensus.io/stats/view#Exporter), +[trace](https://godoc.org/go.opencensus.io/trace#Exporter)): + +* [Prometheus][exporter-prom] for stats +* [OpenZipkin][exporter-zipkin] for traces +* [Stackdriver][exporter-stackdriver] Monitoring for stats and Trace for traces +* [Jaeger][exporter-jaeger] for traces +* [AWS X-Ray][exporter-xray] for traces +* [Datadog][exporter-datadog] for stats and traces +* [Graphite][exporter-graphite] for stats +* [Honeycomb][exporter-honeycomb] for traces +* [New Relic][exporter-newrelic] for stats and traces + +## Overview + +![OpenCensus Overview](https://i.imgur.com/cf4ElHE.jpg) + +In a microservices environment, a user request may go through +multiple services until there is a response. OpenCensus allows +you to instrument your services and collect diagnostics data all +through your services end-to-end. + +## Tags + +Tags represent propagated key-value pairs. They are propagated using `context.Context` +in the same process or can be encoded to be transmitted on the wire. Usually, this will +be handled by an integration plugin, e.g. `ocgrpc.ServerHandler` and `ocgrpc.ClientHandler` +for gRPC. + +Package `tag` allows adding or modifying tags in the current context. + +[embedmd]:# (internal/readme/tags.go new) +```go +ctx, err := tag.New(ctx, + tag.Insert(osKey, "macOS-10.12.5"), + tag.Upsert(userIDKey, "cde36753ed"), +) +if err != nil { + log.Fatal(err) +} +``` + +## Stats + +OpenCensus is a low-overhead framework even if instrumentation is always enabled. +In order to be so, it is optimized to make recording of data points fast +and separate from the data aggregation. + +OpenCensus stats collection happens in two stages: + +* Definition of measures and recording of data points +* Definition of views and aggregation of the recorded data + +### Recording + +Measurements are data points associated with a measure. +Recording implicitly tags the set of Measurements with the tags from the +provided context: + +[embedmd]:# (internal/readme/stats.go record) +```go +stats.Record(ctx, videoSize.M(102478)) +``` + +### Views + +Views are how Measures are aggregated. You can think of them as queries over the +set of recorded data points (measurements). + +Views have two parts: the tags to group by and the aggregation type used. + +Currently three types of aggregations are supported: +* CountAggregation is used to count the number of times a sample was recorded. +* DistributionAggregation is used to provide a histogram of the values of the samples. +* SumAggregation is used to sum up all sample values. + +[embedmd]:# (internal/readme/stats.go aggs) +```go +distAgg := view.Distribution(1<<32, 2<<32, 3<<32) +countAgg := view.Count() +sumAgg := view.Sum() +``` + +Here we create a view with the DistributionAggregation over our measure. + +[embedmd]:# (internal/readme/stats.go view) +```go +if err := view.Register(&view.View{ + Name: "example.com/video_size_distribution", + Description: "distribution of processed video size over time", + Measure: videoSize, + Aggregation: view.Distribution(1<<32, 2<<32, 3<<32), +}); err != nil { + log.Fatalf("Failed to register view: %v", err) +} +``` + +Register begins collecting data for the view. Registered views' data will be +exported via the registered exporters. + +## Traces + +A distributed trace tracks the progression of a single user request as +it is handled by the services and processes that make up an application. +Each step is called a span in the trace. Spans include metadata about the step, +including especially the time spent in the step, called the span’s latency. + +Below you see a trace and several spans underneath it. + +![Traces and spans](https://i.imgur.com/7hZwRVj.png) + +### Spans + +Span is the unit step in a trace. Each span has a name, latency, status and +additional metadata. + +Below we are starting a span for a cache read and ending it +when we are done: + +[embedmd]:# (internal/readme/trace.go startend) +```go +ctx, span := trace.StartSpan(ctx, "cache.Get") +defer span.End() + +// Do work to get from cache. +``` + +### Propagation + +Spans can have parents or can be root spans if they don't have any parents. +The current span is propagated in-process and across the network to allow associating +new child spans with the parent. + +In the same process, `context.Context` is used to propagate spans. +`trace.StartSpan` creates a new span as a root if the current context +doesn't contain a span. Or, it creates a child of the span that is +already in current context. The returned context can be used to keep +propagating the newly created span in the current context. + +[embedmd]:# (internal/readme/trace.go startend) +```go +ctx, span := trace.StartSpan(ctx, "cache.Get") +defer span.End() + +// Do work to get from cache. +``` + +Across the network, OpenCensus provides different propagation +methods for different protocols. + +* gRPC integrations use the OpenCensus' [binary propagation format](https://godoc.org/go.opencensus.io/trace/propagation). +* HTTP integrations use Zipkin's [B3](https://github.com/openzipkin/b3-propagation) + by default but can be configured to use a custom propagation method by setting another + [propagation.HTTPFormat](https://godoc.org/go.opencensus.io/trace/propagation#HTTPFormat). + +## Execution Tracer + +With Go 1.11, OpenCensus Go will support integration with the Go execution tracer. +See [Debugging Latency in Go](https://medium.com/observability/debugging-latency-in-go-1-11-9f97a7910d68) +for an example of their mutual use. + +## Profiles + +OpenCensus tags can be applied as profiler labels +for users who are on Go 1.9 and above. + +[embedmd]:# (internal/readme/tags.go profiler) +```go +ctx, err = tag.New(ctx, + tag.Insert(osKey, "macOS-10.12.5"), + tag.Insert(userIDKey, "fff0989878"), +) +if err != nil { + log.Fatal(err) +} +tag.Do(ctx, func(ctx context.Context) { + // Do work. + // When profiling is on, samples will be + // recorded with the key/values from the tag map. +}) +``` + +A screenshot of the CPU profile from the program above: + +![CPU profile](https://i.imgur.com/jBKjlkw.png) + +## Deprecation Policy + +Before version 1.0.0, the following deprecation policy will be observed: + +No backwards-incompatible changes will be made except for the removal of symbols that have +been marked as *Deprecated* for at least one minor release (e.g. 0.9.0 to 0.10.0). A release +removing the *Deprecated* functionality will be made no sooner than 28 days after the first +release in which the functionality was marked *Deprecated*. + +[travis-image]: https://travis-ci.org/census-instrumentation/opencensus-go.svg?branch=master +[travis-url]: https://travis-ci.org/census-instrumentation/opencensus-go +[appveyor-image]: https://ci.appveyor.com/api/projects/status/vgtt29ps1783ig38?svg=true +[appveyor-url]: https://ci.appveyor.com/project/opencensusgoteam/opencensus-go/branch/master +[godoc-image]: https://godoc.org/go.opencensus.io?status.svg +[godoc-url]: https://godoc.org/go.opencensus.io +[gitter-image]: https://badges.gitter.im/census-instrumentation/lobby.svg +[gitter-url]: https://gitter.im/census-instrumentation/lobby?utm_source=badge&utm_medium=badge&utm_campaign=pr-badge&utm_content=badge + + +[new-ex]: https://godoc.org/go.opencensus.io/tag#example-NewMap +[new-replace-ex]: https://godoc.org/go.opencensus.io/tag#example-NewMap--Replace + +[exporter-prom]: https://godoc.org/contrib.go.opencensus.io/exporter/prometheus +[exporter-stackdriver]: https://godoc.org/contrib.go.opencensus.io/exporter/stackdriver +[exporter-zipkin]: https://godoc.org/contrib.go.opencensus.io/exporter/zipkin +[exporter-jaeger]: https://godoc.org/contrib.go.opencensus.io/exporter/jaeger +[exporter-xray]: https://github.com/census-ecosystem/opencensus-go-exporter-aws +[exporter-datadog]: https://github.com/DataDog/opencensus-go-exporter-datadog +[exporter-graphite]: https://github.com/census-ecosystem/opencensus-go-exporter-graphite +[exporter-honeycomb]: https://github.com/honeycombio/opencensus-exporter +[exporter-newrelic]: https://github.com/newrelic/newrelic-opencensus-exporter-go diff --git a/vendor/go.opencensus.io/appveyor.yml b/vendor/go.opencensus.io/appveyor.yml new file mode 100644 index 00000000000..d08f0edaff9 --- /dev/null +++ b/vendor/go.opencensus.io/appveyor.yml @@ -0,0 +1,24 @@ +version: "{build}" + +platform: x64 + +clone_folder: c:\gopath\src\go.opencensus.io + +environment: + GOPATH: 'c:\gopath' + GO111MODULE: 'on' + CGO_ENABLED: '0' # See: https://github.com/appveyor/ci/issues/2613 + +stack: go 1.11 + +before_test: + - go version + - go env + +build: false +deploy: false + +test_script: + - cd %APPVEYOR_BUILD_FOLDER% + - go build -v .\... + - go test -v .\... # No -race because cgo is disabled diff --git a/vendor/go.opencensus.io/internal/internal.go b/vendor/go.opencensus.io/internal/internal.go new file mode 100644 index 00000000000..81dc7183ec3 --- /dev/null +++ b/vendor/go.opencensus.io/internal/internal.go @@ -0,0 +1,37 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package internal // import "go.opencensus.io/internal" + +import ( + "fmt" + "time" + + opencensus "go.opencensus.io" +) + +// UserAgent is the user agent to be added to the outgoing +// requests from the exporters. +var UserAgent = fmt.Sprintf("opencensus-go/%s", opencensus.Version()) + +// MonotonicEndTime returns the end time at present +// but offset from start, monotonically. +// +// The monotonic clock is used in subtractions hence +// the duration since start added back to start gives +// end as a monotonic time. +// See https://golang.org/pkg/time/#hdr-Monotonic_Clocks +func MonotonicEndTime(start time.Time) time.Time { + return start.Add(time.Since(start)) +} diff --git a/vendor/go.opencensus.io/internal/sanitize.go b/vendor/go.opencensus.io/internal/sanitize.go new file mode 100644 index 00000000000..de8ccf236c4 --- /dev/null +++ b/vendor/go.opencensus.io/internal/sanitize.go @@ -0,0 +1,50 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package internal + +import ( + "strings" + "unicode" +) + +const labelKeySizeLimit = 100 + +// Sanitize returns a string that is trunacated to 100 characters if it's too +// long, and replaces non-alphanumeric characters to underscores. +func Sanitize(s string) string { + if len(s) == 0 { + return s + } + if len(s) > labelKeySizeLimit { + s = s[:labelKeySizeLimit] + } + s = strings.Map(sanitizeRune, s) + if unicode.IsDigit(rune(s[0])) { + s = "key_" + s + } + if s[0] == '_' { + s = "key" + s + } + return s +} + +// converts anything that is not a letter or digit to an underscore +func sanitizeRune(r rune) rune { + if unicode.IsLetter(r) || unicode.IsDigit(r) { + return r + } + // Everything else turns into an underscore + return '_' +} diff --git a/vendor/go.opencensus.io/internal/traceinternals.go b/vendor/go.opencensus.io/internal/traceinternals.go new file mode 100644 index 00000000000..073af7b473a --- /dev/null +++ b/vendor/go.opencensus.io/internal/traceinternals.go @@ -0,0 +1,53 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package internal + +import ( + "time" +) + +// Trace allows internal access to some trace functionality. +// TODO(#412): remove this +var Trace interface{} + +// LocalSpanStoreEnabled true if the local span store is enabled. +var LocalSpanStoreEnabled bool + +// BucketConfiguration stores the number of samples to store for span buckets +// for successful and failed spans for a particular span name. +type BucketConfiguration struct { + Name string + MaxRequestsSucceeded int + MaxRequestsErrors int +} + +// PerMethodSummary is a summary of the spans stored for a single span name. +type PerMethodSummary struct { + Active int + LatencyBuckets []LatencyBucketSummary + ErrorBuckets []ErrorBucketSummary +} + +// LatencyBucketSummary is a summary of a latency bucket. +type LatencyBucketSummary struct { + MinLatency, MaxLatency time.Duration + Size int +} + +// ErrorBucketSummary is a summary of an error bucket. +type ErrorBucketSummary struct { + ErrorCode int32 + Size int +} diff --git a/vendor/go.opencensus.io/opencensus.go b/vendor/go.opencensus.io/opencensus.go new file mode 100644 index 00000000000..11e31f421c5 --- /dev/null +++ b/vendor/go.opencensus.io/opencensus.go @@ -0,0 +1,21 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package opencensus contains Go support for OpenCensus. +package opencensus // import "go.opencensus.io" + +// Version is the current release version of OpenCensus in use. +func Version() string { + return "0.24.0" +} diff --git a/vendor/go.opencensus.io/trace/basetypes.go b/vendor/go.opencensus.io/trace/basetypes.go new file mode 100644 index 00000000000..c8e26ed6355 --- /dev/null +++ b/vendor/go.opencensus.io/trace/basetypes.go @@ -0,0 +1,129 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "fmt" + "time" +) + +type ( + // TraceID is a 16-byte identifier for a set of spans. + TraceID [16]byte + + // SpanID is an 8-byte identifier for a single span. + SpanID [8]byte +) + +func (t TraceID) String() string { + return fmt.Sprintf("%02x", t[:]) +} + +func (s SpanID) String() string { + return fmt.Sprintf("%02x", s[:]) +} + +// Annotation represents a text annotation with a set of attributes and a timestamp. +type Annotation struct { + Time time.Time + Message string + Attributes map[string]interface{} +} + +// Attribute represents a key-value pair on a span, link or annotation. +// Construct with one of: BoolAttribute, Int64Attribute, or StringAttribute. +type Attribute struct { + key string + value interface{} +} + +// Key returns the attribute's key +func (a *Attribute) Key() string { + return a.key +} + +// Value returns the attribute's value +func (a *Attribute) Value() interface{} { + return a.value +} + +// BoolAttribute returns a bool-valued attribute. +func BoolAttribute(key string, value bool) Attribute { + return Attribute{key: key, value: value} +} + +// Int64Attribute returns an int64-valued attribute. +func Int64Attribute(key string, value int64) Attribute { + return Attribute{key: key, value: value} +} + +// Float64Attribute returns a float64-valued attribute. +func Float64Attribute(key string, value float64) Attribute { + return Attribute{key: key, value: value} +} + +// StringAttribute returns a string-valued attribute. +func StringAttribute(key string, value string) Attribute { + return Attribute{key: key, value: value} +} + +// LinkType specifies the relationship between the span that had the link +// added, and the linked span. +type LinkType int32 + +// LinkType values. +const ( + LinkTypeUnspecified LinkType = iota // The relationship of the two spans is unknown. + LinkTypeChild // The linked span is a child of the current span. + LinkTypeParent // The linked span is the parent of the current span. +) + +// Link represents a reference from one span to another span. +type Link struct { + TraceID TraceID + SpanID SpanID + Type LinkType + // Attributes is a set of attributes on the link. + Attributes map[string]interface{} +} + +// MessageEventType specifies the type of message event. +type MessageEventType int32 + +// MessageEventType values. +const ( + MessageEventTypeUnspecified MessageEventType = iota // Unknown event type. + MessageEventTypeSent // Indicates a sent RPC message. + MessageEventTypeRecv // Indicates a received RPC message. +) + +// MessageEvent represents an event describing a message sent or received on the network. +type MessageEvent struct { + Time time.Time + EventType MessageEventType + MessageID int64 + UncompressedByteSize int64 + CompressedByteSize int64 +} + +// Status is the status of a Span. +type Status struct { + // Code is a status code. Zero indicates success. + // + // If Code will be propagated to Google APIs, it ideally should be a value from + // https://github.com/googleapis/googleapis/blob/master/google/rpc/code.proto . + Code int32 + Message string +} diff --git a/vendor/go.opencensus.io/trace/config.go b/vendor/go.opencensus.io/trace/config.go new file mode 100644 index 00000000000..775f8274faa --- /dev/null +++ b/vendor/go.opencensus.io/trace/config.go @@ -0,0 +1,86 @@ +// Copyright 2018, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "sync" + + "go.opencensus.io/trace/internal" +) + +// Config represents the global tracing configuration. +type Config struct { + // DefaultSampler is the default sampler used when creating new spans. + DefaultSampler Sampler + + // IDGenerator is for internal use only. + IDGenerator internal.IDGenerator + + // MaxAnnotationEventsPerSpan is max number of annotation events per span + MaxAnnotationEventsPerSpan int + + // MaxMessageEventsPerSpan is max number of message events per span + MaxMessageEventsPerSpan int + + // MaxAnnotationEventsPerSpan is max number of attributes per span + MaxAttributesPerSpan int + + // MaxLinksPerSpan is max number of links per span + MaxLinksPerSpan int +} + +var configWriteMu sync.Mutex + +const ( + // DefaultMaxAnnotationEventsPerSpan is default max number of annotation events per span + DefaultMaxAnnotationEventsPerSpan = 32 + + // DefaultMaxMessageEventsPerSpan is default max number of message events per span + DefaultMaxMessageEventsPerSpan = 128 + + // DefaultMaxAttributesPerSpan is default max number of attributes per span + DefaultMaxAttributesPerSpan = 32 + + // DefaultMaxLinksPerSpan is default max number of links per span + DefaultMaxLinksPerSpan = 32 +) + +// ApplyConfig applies changes to the global tracing configuration. +// +// Fields not provided in the given config are going to be preserved. +func ApplyConfig(cfg Config) { + configWriteMu.Lock() + defer configWriteMu.Unlock() + c := *config.Load().(*Config) + if cfg.DefaultSampler != nil { + c.DefaultSampler = cfg.DefaultSampler + } + if cfg.IDGenerator != nil { + c.IDGenerator = cfg.IDGenerator + } + if cfg.MaxAnnotationEventsPerSpan > 0 { + c.MaxAnnotationEventsPerSpan = cfg.MaxAnnotationEventsPerSpan + } + if cfg.MaxMessageEventsPerSpan > 0 { + c.MaxMessageEventsPerSpan = cfg.MaxMessageEventsPerSpan + } + if cfg.MaxAttributesPerSpan > 0 { + c.MaxAttributesPerSpan = cfg.MaxAttributesPerSpan + } + if cfg.MaxLinksPerSpan > 0 { + c.MaxLinksPerSpan = cfg.MaxLinksPerSpan + } + config.Store(&c) +} diff --git a/vendor/go.opencensus.io/trace/doc.go b/vendor/go.opencensus.io/trace/doc.go new file mode 100644 index 00000000000..7a1616a55c5 --- /dev/null +++ b/vendor/go.opencensus.io/trace/doc.go @@ -0,0 +1,52 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +/* +Package trace contains support for OpenCensus distributed tracing. + +The following assumes a basic familiarity with OpenCensus concepts. +See http://opencensus.io + +# Exporting Traces + +To export collected tracing data, register at least one exporter. You can use +one of the provided exporters or write your own. + + trace.RegisterExporter(exporter) + +By default, traces will be sampled relatively rarely. To change the sampling +frequency for your entire program, call ApplyConfig. Use a ProbabilitySampler +to sample a subset of traces, or use AlwaysSample to collect a trace on every run: + + trace.ApplyConfig(trace.Config{DefaultSampler: trace.AlwaysSample()}) + +Be careful about using trace.AlwaysSample in a production application with +significant traffic: a new trace will be started and exported for every request. + +# Adding Spans to a Trace + +A trace consists of a tree of spans. In Go, the current span is carried in a +context.Context. + +It is common to want to capture all the activity of a function call in a span. For +this to work, the function must take a context.Context as a parameter. Add these two +lines to the top of the function: + + ctx, span := trace.StartSpan(ctx, "example.com/Run") + defer span.End() + +StartSpan will create a new top-level span if the context +doesn't contain another span, otherwise it will create a child span. +*/ +package trace // import "go.opencensus.io/trace" diff --git a/vendor/go.opencensus.io/trace/evictedqueue.go b/vendor/go.opencensus.io/trace/evictedqueue.go new file mode 100644 index 00000000000..ffc264f23d2 --- /dev/null +++ b/vendor/go.opencensus.io/trace/evictedqueue.go @@ -0,0 +1,38 @@ +// Copyright 2019, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +type evictedQueue struct { + queue []interface{} + capacity int + droppedCount int +} + +func newEvictedQueue(capacity int) *evictedQueue { + eq := &evictedQueue{ + capacity: capacity, + queue: make([]interface{}, 0), + } + + return eq +} + +func (eq *evictedQueue) add(value interface{}) { + if len(eq.queue) == eq.capacity { + eq.queue = eq.queue[1:] + eq.droppedCount++ + } + eq.queue = append(eq.queue, value) +} diff --git a/vendor/go.opencensus.io/trace/export.go b/vendor/go.opencensus.io/trace/export.go new file mode 100644 index 00000000000..e0d9a4b99e9 --- /dev/null +++ b/vendor/go.opencensus.io/trace/export.go @@ -0,0 +1,97 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "sync" + "sync/atomic" + "time" +) + +// Exporter is a type for functions that receive sampled trace spans. +// +// The ExportSpan method should be safe for concurrent use and should return +// quickly; if an Exporter takes a significant amount of time to process a +// SpanData, that work should be done on another goroutine. +// +// The SpanData should not be modified, but a pointer to it can be kept. +type Exporter interface { + ExportSpan(s *SpanData) +} + +type exportersMap map[Exporter]struct{} + +var ( + exporterMu sync.Mutex + exporters atomic.Value +) + +// RegisterExporter adds to the list of Exporters that will receive sampled +// trace spans. +// +// Binaries can register exporters, libraries shouldn't register exporters. +func RegisterExporter(e Exporter) { + exporterMu.Lock() + new := make(exportersMap) + if old, ok := exporters.Load().(exportersMap); ok { + for k, v := range old { + new[k] = v + } + } + new[e] = struct{}{} + exporters.Store(new) + exporterMu.Unlock() +} + +// UnregisterExporter removes from the list of Exporters the Exporter that was +// registered with the given name. +func UnregisterExporter(e Exporter) { + exporterMu.Lock() + new := make(exportersMap) + if old, ok := exporters.Load().(exportersMap); ok { + for k, v := range old { + new[k] = v + } + } + delete(new, e) + exporters.Store(new) + exporterMu.Unlock() +} + +// SpanData contains all the information collected by a Span. +type SpanData struct { + SpanContext + ParentSpanID SpanID + SpanKind int + Name string + StartTime time.Time + // The wall clock time of EndTime will be adjusted to always be offset + // from StartTime by the duration of the span. + EndTime time.Time + // The values of Attributes each have type string, bool, or int64. + Attributes map[string]interface{} + Annotations []Annotation + MessageEvents []MessageEvent + Status + Links []Link + HasRemoteParent bool + DroppedAttributeCount int + DroppedAnnotationCount int + DroppedMessageEventCount int + DroppedLinkCount int + + // ChildSpanCount holds the number of child span created for this span. + ChildSpanCount int +} diff --git a/vendor/go.opencensus.io/trace/internal/internal.go b/vendor/go.opencensus.io/trace/internal/internal.go new file mode 100644 index 00000000000..7e808d8f30e --- /dev/null +++ b/vendor/go.opencensus.io/trace/internal/internal.go @@ -0,0 +1,22 @@ +// Copyright 2018, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package internal provides trace internals. +package internal + +// IDGenerator allows custom generators for TraceId and SpanId. +type IDGenerator interface { + NewTraceID() [16]byte + NewSpanID() [8]byte +} diff --git a/vendor/go.opencensus.io/trace/lrumap.go b/vendor/go.opencensus.io/trace/lrumap.go new file mode 100644 index 00000000000..80095a5f6c0 --- /dev/null +++ b/vendor/go.opencensus.io/trace/lrumap.go @@ -0,0 +1,61 @@ +// Copyright 2019, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "github.com/golang/groupcache/lru" +) + +// A simple lru.Cache wrapper that tracks the keys of the current contents and +// the cumulative number of evicted items. +type lruMap struct { + cacheKeys map[lru.Key]bool + cache *lru.Cache + droppedCount int +} + +func newLruMap(size int) *lruMap { + lm := &lruMap{ + cacheKeys: make(map[lru.Key]bool), + cache: lru.New(size), + droppedCount: 0, + } + lm.cache.OnEvicted = func(key lru.Key, value interface{}) { + delete(lm.cacheKeys, key) + lm.droppedCount++ + } + return lm +} + +func (lm lruMap) len() int { + return lm.cache.Len() +} + +func (lm lruMap) keys() []interface{} { + keys := make([]interface{}, 0, len(lm.cacheKeys)) + for k := range lm.cacheKeys { + keys = append(keys, k) + } + return keys +} + +func (lm *lruMap) add(key, value interface{}) { + lm.cacheKeys[lru.Key(key)] = true + lm.cache.Add(lru.Key(key), value) +} + +func (lm *lruMap) get(key interface{}) (interface{}, bool) { + return lm.cache.Get(key) +} diff --git a/vendor/go.opencensus.io/trace/sampling.go b/vendor/go.opencensus.io/trace/sampling.go new file mode 100644 index 00000000000..71c10f9e3b4 --- /dev/null +++ b/vendor/go.opencensus.io/trace/sampling.go @@ -0,0 +1,75 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "encoding/binary" +) + +const defaultSamplingProbability = 1e-4 + +// Sampler decides whether a trace should be sampled and exported. +type Sampler func(SamplingParameters) SamplingDecision + +// SamplingParameters contains the values passed to a Sampler. +type SamplingParameters struct { + ParentContext SpanContext + TraceID TraceID + SpanID SpanID + Name string + HasRemoteParent bool +} + +// SamplingDecision is the value returned by a Sampler. +type SamplingDecision struct { + Sample bool +} + +// ProbabilitySampler returns a Sampler that samples a given fraction of traces. +// +// It also samples spans whose parents are sampled. +func ProbabilitySampler(fraction float64) Sampler { + if !(fraction >= 0) { + fraction = 0 + } else if fraction >= 1 { + return AlwaysSample() + } + + traceIDUpperBound := uint64(fraction * (1 << 63)) + return Sampler(func(p SamplingParameters) SamplingDecision { + if p.ParentContext.IsSampled() { + return SamplingDecision{Sample: true} + } + x := binary.BigEndian.Uint64(p.TraceID[0:8]) >> 1 + return SamplingDecision{Sample: x < traceIDUpperBound} + }) +} + +// AlwaysSample returns a Sampler that samples every trace. +// Be careful about using this sampler in a production application with +// significant traffic: a new trace will be started and exported for every +// request. +func AlwaysSample() Sampler { + return func(p SamplingParameters) SamplingDecision { + return SamplingDecision{Sample: true} + } +} + +// NeverSample returns a Sampler that samples no traces. +func NeverSample() Sampler { + return func(p SamplingParameters) SamplingDecision { + return SamplingDecision{Sample: false} + } +} diff --git a/vendor/go.opencensus.io/trace/spanbucket.go b/vendor/go.opencensus.io/trace/spanbucket.go new file mode 100644 index 00000000000..fbabad34c00 --- /dev/null +++ b/vendor/go.opencensus.io/trace/spanbucket.go @@ -0,0 +1,130 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "time" +) + +// samplePeriod is the minimum time between accepting spans in a single bucket. +const samplePeriod = time.Second + +// defaultLatencies contains the default latency bucket bounds. +// TODO: consider defaults, make configurable +var defaultLatencies = [...]time.Duration{ + 10 * time.Microsecond, + 100 * time.Microsecond, + time.Millisecond, + 10 * time.Millisecond, + 100 * time.Millisecond, + time.Second, + 10 * time.Second, + time.Minute, +} + +// bucket is a container for a set of spans for a particular error code or latency range. +type bucket struct { + nextTime time.Time // next time we can accept a span + buffer []*SpanData // circular buffer of spans + nextIndex int // location next SpanData should be placed in buffer + overflow bool // whether the circular buffer has wrapped around +} + +func makeBucket(bufferSize int) bucket { + return bucket{ + buffer: make([]*SpanData, bufferSize), + } +} + +// add adds a span to the bucket, if nextTime has been reached. +func (b *bucket) add(s *SpanData) { + if s.EndTime.Before(b.nextTime) { + return + } + if len(b.buffer) == 0 { + return + } + b.nextTime = s.EndTime.Add(samplePeriod) + b.buffer[b.nextIndex] = s + b.nextIndex++ + if b.nextIndex == len(b.buffer) { + b.nextIndex = 0 + b.overflow = true + } +} + +// size returns the number of spans in the bucket. +func (b *bucket) size() int { + if b.overflow { + return len(b.buffer) + } + return b.nextIndex +} + +// span returns the ith span in the bucket. +func (b *bucket) span(i int) *SpanData { + if !b.overflow { + return b.buffer[i] + } + if i < len(b.buffer)-b.nextIndex { + return b.buffer[b.nextIndex+i] + } + return b.buffer[b.nextIndex+i-len(b.buffer)] +} + +// resize changes the size of the bucket to n, keeping up to n existing spans. +func (b *bucket) resize(n int) { + cur := b.size() + newBuffer := make([]*SpanData, n) + if cur < n { + for i := 0; i < cur; i++ { + newBuffer[i] = b.span(i) + } + b.buffer = newBuffer + b.nextIndex = cur + b.overflow = false + return + } + for i := 0; i < n; i++ { + newBuffer[i] = b.span(i + cur - n) + } + b.buffer = newBuffer + b.nextIndex = 0 + b.overflow = true +} + +// latencyBucket returns the appropriate bucket number for a given latency. +func latencyBucket(latency time.Duration) int { + i := 0 + for i < len(defaultLatencies) && latency >= defaultLatencies[i] { + i++ + } + return i +} + +// latencyBucketBounds returns the lower and upper bounds for a latency bucket +// number. +// +// The lower bound is inclusive, the upper bound is exclusive (except for the +// last bucket.) +func latencyBucketBounds(index int) (lower time.Duration, upper time.Duration) { + if index == 0 { + return 0, defaultLatencies[index] + } + if index == len(defaultLatencies) { + return defaultLatencies[index-1], 1<<63 - 1 + } + return defaultLatencies[index-1], defaultLatencies[index] +} diff --git a/vendor/go.opencensus.io/trace/spanstore.go b/vendor/go.opencensus.io/trace/spanstore.go new file mode 100644 index 00000000000..e601f76f2c8 --- /dev/null +++ b/vendor/go.opencensus.io/trace/spanstore.go @@ -0,0 +1,308 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "sync" + "time" + + "go.opencensus.io/internal" +) + +const ( + maxBucketSize = 100000 + defaultBucketSize = 10 +) + +var ( + ssmu sync.RWMutex // protects spanStores + spanStores = make(map[string]*spanStore) +) + +// This exists purely to avoid exposing internal methods used by z-Pages externally. +type internalOnly struct{} + +func init() { + //TODO(#412): remove + internal.Trace = &internalOnly{} +} + +// ReportActiveSpans returns the active spans for the given name. +func (i internalOnly) ReportActiveSpans(name string) []*SpanData { + s := spanStoreForName(name) + if s == nil { + return nil + } + var out []*SpanData + s.mu.Lock() + defer s.mu.Unlock() + for activeSpan := range s.active { + if s, ok := activeSpan.(*span); ok { + out = append(out, s.makeSpanData()) + } + } + return out +} + +// ReportSpansByError returns a sample of error spans. +// +// If code is nonzero, only spans with that status code are returned. +func (i internalOnly) ReportSpansByError(name string, code int32) []*SpanData { + s := spanStoreForName(name) + if s == nil { + return nil + } + var out []*SpanData + s.mu.Lock() + defer s.mu.Unlock() + if code != 0 { + if b, ok := s.errors[code]; ok { + for _, sd := range b.buffer { + if sd == nil { + break + } + out = append(out, sd) + } + } + } else { + for _, b := range s.errors { + for _, sd := range b.buffer { + if sd == nil { + break + } + out = append(out, sd) + } + } + } + return out +} + +// ConfigureBucketSizes sets the number of spans to keep per latency and error +// bucket for different span names. +func (i internalOnly) ConfigureBucketSizes(bcs []internal.BucketConfiguration) { + for _, bc := range bcs { + latencyBucketSize := bc.MaxRequestsSucceeded + if latencyBucketSize < 0 { + latencyBucketSize = 0 + } + if latencyBucketSize > maxBucketSize { + latencyBucketSize = maxBucketSize + } + errorBucketSize := bc.MaxRequestsErrors + if errorBucketSize < 0 { + errorBucketSize = 0 + } + if errorBucketSize > maxBucketSize { + errorBucketSize = maxBucketSize + } + spanStoreSetSize(bc.Name, latencyBucketSize, errorBucketSize) + } +} + +// ReportSpansPerMethod returns a summary of what spans are being stored for each span name. +func (i internalOnly) ReportSpansPerMethod() map[string]internal.PerMethodSummary { + out := make(map[string]internal.PerMethodSummary) + ssmu.RLock() + defer ssmu.RUnlock() + for name, s := range spanStores { + s.mu.Lock() + p := internal.PerMethodSummary{ + Active: len(s.active), + } + for code, b := range s.errors { + p.ErrorBuckets = append(p.ErrorBuckets, internal.ErrorBucketSummary{ + ErrorCode: code, + Size: b.size(), + }) + } + for i, b := range s.latency { + min, max := latencyBucketBounds(i) + p.LatencyBuckets = append(p.LatencyBuckets, internal.LatencyBucketSummary{ + MinLatency: min, + MaxLatency: max, + Size: b.size(), + }) + } + s.mu.Unlock() + out[name] = p + } + return out +} + +// ReportSpansByLatency returns a sample of successful spans. +// +// minLatency is the minimum latency of spans to be returned. +// maxLatency, if nonzero, is the maximum latency of spans to be returned. +func (i internalOnly) ReportSpansByLatency(name string, minLatency, maxLatency time.Duration) []*SpanData { + s := spanStoreForName(name) + if s == nil { + return nil + } + var out []*SpanData + s.mu.Lock() + defer s.mu.Unlock() + for i, b := range s.latency { + min, max := latencyBucketBounds(i) + if i+1 != len(s.latency) && max <= minLatency { + continue + } + if maxLatency != 0 && maxLatency < min { + continue + } + for _, sd := range b.buffer { + if sd == nil { + break + } + if minLatency != 0 || maxLatency != 0 { + d := sd.EndTime.Sub(sd.StartTime) + if d < minLatency { + continue + } + if maxLatency != 0 && d > maxLatency { + continue + } + } + out = append(out, sd) + } + } + return out +} + +// spanStore keeps track of spans stored for a particular span name. +// +// It contains all active spans; a sample of spans for failed requests, +// categorized by error code; and a sample of spans for successful requests, +// bucketed by latency. +type spanStore struct { + mu sync.Mutex // protects everything below. + active map[SpanInterface]struct{} + errors map[int32]*bucket + latency []bucket + maxSpansPerErrorBucket int +} + +// newSpanStore creates a span store. +func newSpanStore(name string, latencyBucketSize int, errorBucketSize int) *spanStore { + s := &spanStore{ + active: make(map[SpanInterface]struct{}), + latency: make([]bucket, len(defaultLatencies)+1), + maxSpansPerErrorBucket: errorBucketSize, + } + for i := range s.latency { + s.latency[i] = makeBucket(latencyBucketSize) + } + return s +} + +// spanStoreForName returns the spanStore for the given name. +// +// It returns nil if it doesn't exist. +func spanStoreForName(name string) *spanStore { + var s *spanStore + ssmu.RLock() + s, _ = spanStores[name] + ssmu.RUnlock() + return s +} + +// spanStoreForNameCreateIfNew returns the spanStore for the given name. +// +// It creates it if it didn't exist. +func spanStoreForNameCreateIfNew(name string) *spanStore { + ssmu.RLock() + s, ok := spanStores[name] + ssmu.RUnlock() + if ok { + return s + } + ssmu.Lock() + defer ssmu.Unlock() + s, ok = spanStores[name] + if ok { + return s + } + s = newSpanStore(name, defaultBucketSize, defaultBucketSize) + spanStores[name] = s + return s +} + +// spanStoreSetSize resizes the spanStore for the given name. +// +// It creates it if it didn't exist. +func spanStoreSetSize(name string, latencyBucketSize int, errorBucketSize int) { + ssmu.RLock() + s, ok := spanStores[name] + ssmu.RUnlock() + if ok { + s.resize(latencyBucketSize, errorBucketSize) + return + } + ssmu.Lock() + defer ssmu.Unlock() + s, ok = spanStores[name] + if ok { + s.resize(latencyBucketSize, errorBucketSize) + return + } + s = newSpanStore(name, latencyBucketSize, errorBucketSize) + spanStores[name] = s +} + +func (s *spanStore) resize(latencyBucketSize int, errorBucketSize int) { + s.mu.Lock() + for i := range s.latency { + s.latency[i].resize(latencyBucketSize) + } + for _, b := range s.errors { + b.resize(errorBucketSize) + } + s.maxSpansPerErrorBucket = errorBucketSize + s.mu.Unlock() +} + +// add adds a span to the active bucket of the spanStore. +func (s *spanStore) add(span SpanInterface) { + s.mu.Lock() + s.active[span] = struct{}{} + s.mu.Unlock() +} + +// finished removes a span from the active set, and adds a corresponding +// SpanData to a latency or error bucket. +func (s *spanStore) finished(span SpanInterface, sd *SpanData) { + latency := sd.EndTime.Sub(sd.StartTime) + if latency < 0 { + latency = 0 + } + code := sd.Status.Code + + s.mu.Lock() + delete(s.active, span) + if code == 0 { + s.latency[latencyBucket(latency)].add(sd) + } else { + if s.errors == nil { + s.errors = make(map[int32]*bucket) + } + if b := s.errors[code]; b != nil { + b.add(sd) + } else { + b := makeBucket(s.maxSpansPerErrorBucket) + s.errors[code] = &b + b.add(sd) + } + } + s.mu.Unlock() +} diff --git a/vendor/go.opencensus.io/trace/status_codes.go b/vendor/go.opencensus.io/trace/status_codes.go new file mode 100644 index 00000000000..ec60effd108 --- /dev/null +++ b/vendor/go.opencensus.io/trace/status_codes.go @@ -0,0 +1,37 @@ +// Copyright 2018, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +// Status codes for use with Span.SetStatus. These correspond to the status +// codes used by gRPC defined here: https://github.com/googleapis/googleapis/blob/master/google/rpc/code.proto +const ( + StatusCodeOK = 0 + StatusCodeCancelled = 1 + StatusCodeUnknown = 2 + StatusCodeInvalidArgument = 3 + StatusCodeDeadlineExceeded = 4 + StatusCodeNotFound = 5 + StatusCodeAlreadyExists = 6 + StatusCodePermissionDenied = 7 + StatusCodeResourceExhausted = 8 + StatusCodeFailedPrecondition = 9 + StatusCodeAborted = 10 + StatusCodeOutOfRange = 11 + StatusCodeUnimplemented = 12 + StatusCodeInternal = 13 + StatusCodeUnavailable = 14 + StatusCodeDataLoss = 15 + StatusCodeUnauthenticated = 16 +) diff --git a/vendor/go.opencensus.io/trace/trace.go b/vendor/go.opencensus.io/trace/trace.go new file mode 100644 index 00000000000..861df9d3913 --- /dev/null +++ b/vendor/go.opencensus.io/trace/trace.go @@ -0,0 +1,595 @@ +// Copyright 2017, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "context" + crand "crypto/rand" + "encoding/binary" + "fmt" + "math/rand" + "sync" + "sync/atomic" + "time" + + "go.opencensus.io/internal" + "go.opencensus.io/trace/tracestate" +) + +type tracer struct{} + +var _ Tracer = &tracer{} + +// Span represents a span of a trace. It has an associated SpanContext, and +// stores data accumulated while the span is active. +// +// Ideally users should interact with Spans by calling the functions in this +// package that take a Context parameter. +type span struct { + // data contains information recorded about the span. + // + // It will be non-nil if we are exporting the span or recording events for it. + // Otherwise, data is nil, and the Span is simply a carrier for the + // SpanContext, so that the trace ID is propagated. + data *SpanData + mu sync.Mutex // protects the contents of *data (but not the pointer value.) + spanContext SpanContext + + // lruAttributes are capped at configured limit. When the capacity is reached an oldest entry + // is removed to create room for a new entry. + lruAttributes *lruMap + + // annotations are stored in FIFO queue capped by configured limit. + annotations *evictedQueue + + // messageEvents are stored in FIFO queue capped by configured limit. + messageEvents *evictedQueue + + // links are stored in FIFO queue capped by configured limit. + links *evictedQueue + + // spanStore is the spanStore this span belongs to, if any, otherwise it is nil. + *spanStore + endOnce sync.Once + + executionTracerTaskEnd func() // ends the execution tracer span +} + +// IsRecordingEvents returns true if events are being recorded for this span. +// Use this check to avoid computing expensive annotations when they will never +// be used. +func (s *span) IsRecordingEvents() bool { + if s == nil { + return false + } + return s.data != nil +} + +// TraceOptions contains options associated with a trace span. +type TraceOptions uint32 + +// IsSampled returns true if the span will be exported. +func (sc SpanContext) IsSampled() bool { + return sc.TraceOptions.IsSampled() +} + +// setIsSampled sets the TraceOptions bit that determines whether the span will be exported. +func (sc *SpanContext) setIsSampled(sampled bool) { + if sampled { + sc.TraceOptions |= 1 + } else { + sc.TraceOptions &= ^TraceOptions(1) + } +} + +// IsSampled returns true if the span will be exported. +func (t TraceOptions) IsSampled() bool { + return t&1 == 1 +} + +// SpanContext contains the state that must propagate across process boundaries. +// +// SpanContext is not an implementation of context.Context. +// TODO: add reference to external Census docs for SpanContext. +type SpanContext struct { + TraceID TraceID + SpanID SpanID + TraceOptions TraceOptions + Tracestate *tracestate.Tracestate +} + +type contextKey struct{} + +// FromContext returns the Span stored in a context, or nil if there isn't one. +func (t *tracer) FromContext(ctx context.Context) *Span { + s, _ := ctx.Value(contextKey{}).(*Span) + return s +} + +// NewContext returns a new context with the given Span attached. +func (t *tracer) NewContext(parent context.Context, s *Span) context.Context { + return context.WithValue(parent, contextKey{}, s) +} + +// All available span kinds. Span kind must be either one of these values. +const ( + SpanKindUnspecified = iota + SpanKindServer + SpanKindClient +) + +// StartOptions contains options concerning how a span is started. +type StartOptions struct { + // Sampler to consult for this Span. If provided, it is always consulted. + // + // If not provided, then the behavior differs based on whether + // the parent of this Span is remote, local, or there is no parent. + // In the case of a remote parent or no parent, the + // default sampler (see Config) will be consulted. Otherwise, + // when there is a non-remote parent, no new sampling decision will be made: + // we will preserve the sampling of the parent. + Sampler Sampler + + // SpanKind represents the kind of a span. If none is set, + // SpanKindUnspecified is used. + SpanKind int +} + +// StartOption apply changes to StartOptions. +type StartOption func(*StartOptions) + +// WithSpanKind makes new spans to be created with the given kind. +func WithSpanKind(spanKind int) StartOption { + return func(o *StartOptions) { + o.SpanKind = spanKind + } +} + +// WithSampler makes new spans to be be created with a custom sampler. +// Otherwise, the global sampler is used. +func WithSampler(sampler Sampler) StartOption { + return func(o *StartOptions) { + o.Sampler = sampler + } +} + +// StartSpan starts a new child span of the current span in the context. If +// there is no span in the context, creates a new trace and span. +// +// Returned context contains the newly created span. You can use it to +// propagate the returned span in process. +func (t *tracer) StartSpan(ctx context.Context, name string, o ...StartOption) (context.Context, *Span) { + var opts StartOptions + var parent SpanContext + if p := t.FromContext(ctx); p != nil { + if ps, ok := p.internal.(*span); ok { + ps.addChild() + } + parent = p.SpanContext() + } + for _, op := range o { + op(&opts) + } + span := startSpanInternal(name, parent != SpanContext{}, parent, false, opts) + + ctx, end := startExecutionTracerTask(ctx, name) + span.executionTracerTaskEnd = end + extSpan := NewSpan(span) + return t.NewContext(ctx, extSpan), extSpan +} + +// StartSpanWithRemoteParent starts a new child span of the span from the given parent. +// +// If the incoming context contains a parent, it ignores. StartSpanWithRemoteParent is +// preferred for cases where the parent is propagated via an incoming request. +// +// Returned context contains the newly created span. You can use it to +// propagate the returned span in process. +func (t *tracer) StartSpanWithRemoteParent(ctx context.Context, name string, parent SpanContext, o ...StartOption) (context.Context, *Span) { + var opts StartOptions + for _, op := range o { + op(&opts) + } + span := startSpanInternal(name, parent != SpanContext{}, parent, true, opts) + ctx, end := startExecutionTracerTask(ctx, name) + span.executionTracerTaskEnd = end + extSpan := NewSpan(span) + return t.NewContext(ctx, extSpan), extSpan +} + +func startSpanInternal(name string, hasParent bool, parent SpanContext, remoteParent bool, o StartOptions) *span { + s := &span{} + s.spanContext = parent + + cfg := config.Load().(*Config) + if gen, ok := cfg.IDGenerator.(*defaultIDGenerator); ok { + // lazy initialization + gen.init() + } + + if !hasParent { + s.spanContext.TraceID = cfg.IDGenerator.NewTraceID() + } + s.spanContext.SpanID = cfg.IDGenerator.NewSpanID() + sampler := cfg.DefaultSampler + + if !hasParent || remoteParent || o.Sampler != nil { + // If this span is the child of a local span and no Sampler is set in the + // options, keep the parent's TraceOptions. + // + // Otherwise, consult the Sampler in the options if it is non-nil, otherwise + // the default sampler. + if o.Sampler != nil { + sampler = o.Sampler + } + s.spanContext.setIsSampled(sampler(SamplingParameters{ + ParentContext: parent, + TraceID: s.spanContext.TraceID, + SpanID: s.spanContext.SpanID, + Name: name, + HasRemoteParent: remoteParent}).Sample) + } + + if !internal.LocalSpanStoreEnabled && !s.spanContext.IsSampled() { + return s + } + + s.data = &SpanData{ + SpanContext: s.spanContext, + StartTime: time.Now(), + SpanKind: o.SpanKind, + Name: name, + HasRemoteParent: remoteParent, + } + s.lruAttributes = newLruMap(cfg.MaxAttributesPerSpan) + s.annotations = newEvictedQueue(cfg.MaxAnnotationEventsPerSpan) + s.messageEvents = newEvictedQueue(cfg.MaxMessageEventsPerSpan) + s.links = newEvictedQueue(cfg.MaxLinksPerSpan) + + if hasParent { + s.data.ParentSpanID = parent.SpanID + } + if internal.LocalSpanStoreEnabled { + var ss *spanStore + ss = spanStoreForNameCreateIfNew(name) + if ss != nil { + s.spanStore = ss + ss.add(s) + } + } + + return s +} + +// End ends the span. +func (s *span) End() { + if s == nil { + return + } + if s.executionTracerTaskEnd != nil { + s.executionTracerTaskEnd() + } + if !s.IsRecordingEvents() { + return + } + s.endOnce.Do(func() { + exp, _ := exporters.Load().(exportersMap) + mustExport := s.spanContext.IsSampled() && len(exp) > 0 + if s.spanStore != nil || mustExport { + sd := s.makeSpanData() + sd.EndTime = internal.MonotonicEndTime(sd.StartTime) + if s.spanStore != nil { + s.spanStore.finished(s, sd) + } + if mustExport { + for e := range exp { + e.ExportSpan(sd) + } + } + } + }) +} + +// makeSpanData produces a SpanData representing the current state of the Span. +// It requires that s.data is non-nil. +func (s *span) makeSpanData() *SpanData { + var sd SpanData + s.mu.Lock() + sd = *s.data + if s.lruAttributes.len() > 0 { + sd.Attributes = s.lruAttributesToAttributeMap() + sd.DroppedAttributeCount = s.lruAttributes.droppedCount + } + if len(s.annotations.queue) > 0 { + sd.Annotations = s.interfaceArrayToAnnotationArray() + sd.DroppedAnnotationCount = s.annotations.droppedCount + } + if len(s.messageEvents.queue) > 0 { + sd.MessageEvents = s.interfaceArrayToMessageEventArray() + sd.DroppedMessageEventCount = s.messageEvents.droppedCount + } + if len(s.links.queue) > 0 { + sd.Links = s.interfaceArrayToLinksArray() + sd.DroppedLinkCount = s.links.droppedCount + } + s.mu.Unlock() + return &sd +} + +// SpanContext returns the SpanContext of the span. +func (s *span) SpanContext() SpanContext { + if s == nil { + return SpanContext{} + } + return s.spanContext +} + +// SetName sets the name of the span, if it is recording events. +func (s *span) SetName(name string) { + if !s.IsRecordingEvents() { + return + } + s.mu.Lock() + s.data.Name = name + s.mu.Unlock() +} + +// SetStatus sets the status of the span, if it is recording events. +func (s *span) SetStatus(status Status) { + if !s.IsRecordingEvents() { + return + } + s.mu.Lock() + s.data.Status = status + s.mu.Unlock() +} + +func (s *span) interfaceArrayToLinksArray() []Link { + linksArr := make([]Link, 0, len(s.links.queue)) + for _, value := range s.links.queue { + linksArr = append(linksArr, value.(Link)) + } + return linksArr +} + +func (s *span) interfaceArrayToMessageEventArray() []MessageEvent { + messageEventArr := make([]MessageEvent, 0, len(s.messageEvents.queue)) + for _, value := range s.messageEvents.queue { + messageEventArr = append(messageEventArr, value.(MessageEvent)) + } + return messageEventArr +} + +func (s *span) interfaceArrayToAnnotationArray() []Annotation { + annotationArr := make([]Annotation, 0, len(s.annotations.queue)) + for _, value := range s.annotations.queue { + annotationArr = append(annotationArr, value.(Annotation)) + } + return annotationArr +} + +func (s *span) lruAttributesToAttributeMap() map[string]interface{} { + attributes := make(map[string]interface{}, s.lruAttributes.len()) + for _, key := range s.lruAttributes.keys() { + value, ok := s.lruAttributes.get(key) + if ok { + keyStr := key.(string) + attributes[keyStr] = value + } + } + return attributes +} + +func (s *span) copyToCappedAttributes(attributes []Attribute) { + for _, a := range attributes { + s.lruAttributes.add(a.key, a.value) + } +} + +func (s *span) addChild() { + if !s.IsRecordingEvents() { + return + } + s.mu.Lock() + s.data.ChildSpanCount++ + s.mu.Unlock() +} + +// AddAttributes sets attributes in the span. +// +// Existing attributes whose keys appear in the attributes parameter are overwritten. +func (s *span) AddAttributes(attributes ...Attribute) { + if !s.IsRecordingEvents() { + return + } + s.mu.Lock() + s.copyToCappedAttributes(attributes) + s.mu.Unlock() +} + +func (s *span) printStringInternal(attributes []Attribute, str string) { + now := time.Now() + var am map[string]interface{} + if len(attributes) != 0 { + am = make(map[string]interface{}, len(attributes)) + for _, attr := range attributes { + am[attr.key] = attr.value + } + } + s.mu.Lock() + s.annotations.add(Annotation{ + Time: now, + Message: str, + Attributes: am, + }) + s.mu.Unlock() +} + +// Annotate adds an annotation with attributes. +// Attributes can be nil. +func (s *span) Annotate(attributes []Attribute, str string) { + if !s.IsRecordingEvents() { + return + } + s.printStringInternal(attributes, str) +} + +// Annotatef adds an annotation with attributes. +func (s *span) Annotatef(attributes []Attribute, format string, a ...interface{}) { + if !s.IsRecordingEvents() { + return + } + s.printStringInternal(attributes, fmt.Sprintf(format, a...)) +} + +// AddMessageSendEvent adds a message send event to the span. +// +// messageID is an identifier for the message, which is recommended to be +// unique in this span and the same between the send event and the receive +// event (this allows to identify a message between the sender and receiver). +// For example, this could be a sequence id. +func (s *span) AddMessageSendEvent(messageID, uncompressedByteSize, compressedByteSize int64) { + if !s.IsRecordingEvents() { + return + } + now := time.Now() + s.mu.Lock() + s.messageEvents.add(MessageEvent{ + Time: now, + EventType: MessageEventTypeSent, + MessageID: messageID, + UncompressedByteSize: uncompressedByteSize, + CompressedByteSize: compressedByteSize, + }) + s.mu.Unlock() +} + +// AddMessageReceiveEvent adds a message receive event to the span. +// +// messageID is an identifier for the message, which is recommended to be +// unique in this span and the same between the send event and the receive +// event (this allows to identify a message between the sender and receiver). +// For example, this could be a sequence id. +func (s *span) AddMessageReceiveEvent(messageID, uncompressedByteSize, compressedByteSize int64) { + if !s.IsRecordingEvents() { + return + } + now := time.Now() + s.mu.Lock() + s.messageEvents.add(MessageEvent{ + Time: now, + EventType: MessageEventTypeRecv, + MessageID: messageID, + UncompressedByteSize: uncompressedByteSize, + CompressedByteSize: compressedByteSize, + }) + s.mu.Unlock() +} + +// AddLink adds a link to the span. +func (s *span) AddLink(l Link) { + if !s.IsRecordingEvents() { + return + } + s.mu.Lock() + s.links.add(l) + s.mu.Unlock() +} + +func (s *span) String() string { + if s == nil { + return "" + } + if s.data == nil { + return fmt.Sprintf("span %s", s.spanContext.SpanID) + } + s.mu.Lock() + str := fmt.Sprintf("span %s %q", s.spanContext.SpanID, s.data.Name) + s.mu.Unlock() + return str +} + +var config atomic.Value // access atomically + +func init() { + config.Store(&Config{ + DefaultSampler: ProbabilitySampler(defaultSamplingProbability), + IDGenerator: &defaultIDGenerator{}, + MaxAttributesPerSpan: DefaultMaxAttributesPerSpan, + MaxAnnotationEventsPerSpan: DefaultMaxAnnotationEventsPerSpan, + MaxMessageEventsPerSpan: DefaultMaxMessageEventsPerSpan, + MaxLinksPerSpan: DefaultMaxLinksPerSpan, + }) +} + +type defaultIDGenerator struct { + sync.Mutex + + // Please keep these as the first fields + // so that these 8 byte fields will be aligned on addresses + // divisible by 8, on both 32-bit and 64-bit machines when + // performing atomic increments and accesses. + // See: + // * https://github.com/census-instrumentation/opencensus-go/issues/587 + // * https://github.com/census-instrumentation/opencensus-go/issues/865 + // * https://golang.org/pkg/sync/atomic/#pkg-note-BUG + nextSpanID uint64 + spanIDInc uint64 + + traceIDAdd [2]uint64 + traceIDRand *rand.Rand + + initOnce sync.Once +} + +// init initializes the generator on the first call to avoid consuming entropy +// unnecessarily. +func (gen *defaultIDGenerator) init() { + gen.initOnce.Do(func() { + // initialize traceID and spanID generators. + var rngSeed int64 + for _, p := range []interface{}{ + &rngSeed, &gen.traceIDAdd, &gen.nextSpanID, &gen.spanIDInc, + } { + binary.Read(crand.Reader, binary.LittleEndian, p) + } + gen.traceIDRand = rand.New(rand.NewSource(rngSeed)) + gen.spanIDInc |= 1 + }) +} + +// NewSpanID returns a non-zero span ID from a randomly-chosen sequence. +func (gen *defaultIDGenerator) NewSpanID() [8]byte { + var id uint64 + for id == 0 { + id = atomic.AddUint64(&gen.nextSpanID, gen.spanIDInc) + } + var sid [8]byte + binary.LittleEndian.PutUint64(sid[:], id) + return sid +} + +// NewTraceID returns a non-zero trace ID from a randomly-chosen sequence. +// mu should be held while this function is called. +func (gen *defaultIDGenerator) NewTraceID() [16]byte { + var tid [16]byte + // Construct the trace ID from two outputs of traceIDRand, with a constant + // added to each half for additional entropy. + gen.Lock() + binary.LittleEndian.PutUint64(tid[0:8], gen.traceIDRand.Uint64()+gen.traceIDAdd[0]) + binary.LittleEndian.PutUint64(tid[8:16], gen.traceIDRand.Uint64()+gen.traceIDAdd[1]) + gen.Unlock() + return tid +} diff --git a/vendor/go.opencensus.io/trace/trace_api.go b/vendor/go.opencensus.io/trace/trace_api.go new file mode 100644 index 00000000000..9e2c3a99926 --- /dev/null +++ b/vendor/go.opencensus.io/trace/trace_api.go @@ -0,0 +1,265 @@ +// Copyright 2020, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +package trace + +import ( + "context" +) + +// DefaultTracer is the tracer used when package-level exported functions are invoked. +var DefaultTracer Tracer = &tracer{} + +// Tracer can start spans and access context functions. +type Tracer interface { + + // StartSpan starts a new child span of the current span in the context. If + // there is no span in the context, creates a new trace and span. + // + // Returned context contains the newly created span. You can use it to + // propagate the returned span in process. + StartSpan(ctx context.Context, name string, o ...StartOption) (context.Context, *Span) + + // StartSpanWithRemoteParent starts a new child span of the span from the given parent. + // + // If the incoming context contains a parent, it ignores. StartSpanWithRemoteParent is + // preferred for cases where the parent is propagated via an incoming request. + // + // Returned context contains the newly created span. You can use it to + // propagate the returned span in process. + StartSpanWithRemoteParent(ctx context.Context, name string, parent SpanContext, o ...StartOption) (context.Context, *Span) + + // FromContext returns the Span stored in a context, or nil if there isn't one. + FromContext(ctx context.Context) *Span + + // NewContext returns a new context with the given Span attached. + NewContext(parent context.Context, s *Span) context.Context +} + +// StartSpan starts a new child span of the current span in the context. If +// there is no span in the context, creates a new trace and span. +// +// Returned context contains the newly created span. You can use it to +// propagate the returned span in process. +func StartSpan(ctx context.Context, name string, o ...StartOption) (context.Context, *Span) { + return DefaultTracer.StartSpan(ctx, name, o...) +} + +// StartSpanWithRemoteParent starts a new child span of the span from the given parent. +// +// If the incoming context contains a parent, it ignores. StartSpanWithRemoteParent is +// preferred for cases where the parent is propagated via an incoming request. +// +// Returned context contains the newly created span. You can use it to +// propagate the returned span in process. +func StartSpanWithRemoteParent(ctx context.Context, name string, parent SpanContext, o ...StartOption) (context.Context, *Span) { + return DefaultTracer.StartSpanWithRemoteParent(ctx, name, parent, o...) +} + +// FromContext returns the Span stored in a context, or a Span that is not +// recording events if there isn't one. +func FromContext(ctx context.Context) *Span { + return DefaultTracer.FromContext(ctx) +} + +// NewContext returns a new context with the given Span attached. +func NewContext(parent context.Context, s *Span) context.Context { + return DefaultTracer.NewContext(parent, s) +} + +// SpanInterface represents a span of a trace. It has an associated SpanContext, and +// stores data accumulated while the span is active. +// +// Ideally users should interact with Spans by calling the functions in this +// package that take a Context parameter. +type SpanInterface interface { + + // IsRecordingEvents returns true if events are being recorded for this span. + // Use this check to avoid computing expensive annotations when they will never + // be used. + IsRecordingEvents() bool + + // End ends the span. + End() + + // SpanContext returns the SpanContext of the span. + SpanContext() SpanContext + + // SetName sets the name of the span, if it is recording events. + SetName(name string) + + // SetStatus sets the status of the span, if it is recording events. + SetStatus(status Status) + + // AddAttributes sets attributes in the span. + // + // Existing attributes whose keys appear in the attributes parameter are overwritten. + AddAttributes(attributes ...Attribute) + + // Annotate adds an annotation with attributes. + // Attributes can be nil. + Annotate(attributes []Attribute, str string) + + // Annotatef adds an annotation with attributes. + Annotatef(attributes []Attribute, format string, a ...interface{}) + + // AddMessageSendEvent adds a message send event to the span. + // + // messageID is an identifier for the message, which is recommended to be + // unique in this span and the same between the send event and the receive + // event (this allows to identify a message between the sender and receiver). + // For example, this could be a sequence id. + AddMessageSendEvent(messageID, uncompressedByteSize, compressedByteSize int64) + + // AddMessageReceiveEvent adds a message receive event to the span. + // + // messageID is an identifier for the message, which is recommended to be + // unique in this span and the same between the send event and the receive + // event (this allows to identify a message between the sender and receiver). + // For example, this could be a sequence id. + AddMessageReceiveEvent(messageID, uncompressedByteSize, compressedByteSize int64) + + // AddLink adds a link to the span. + AddLink(l Link) + + // String prints a string representation of a span. + String() string +} + +// NewSpan is a convenience function for creating a *Span out of a *span +func NewSpan(s SpanInterface) *Span { + return &Span{internal: s} +} + +// Span is a struct wrapper around the SpanInt interface, which allows correctly handling +// nil spans, while also allowing the SpanInterface implementation to be swapped out. +type Span struct { + internal SpanInterface +} + +// Internal returns the underlying implementation of the Span +func (s *Span) Internal() SpanInterface { + return s.internal +} + +// IsRecordingEvents returns true if events are being recorded for this span. +// Use this check to avoid computing expensive annotations when they will never +// be used. +func (s *Span) IsRecordingEvents() bool { + if s == nil { + return false + } + return s.internal.IsRecordingEvents() +} + +// End ends the span. +func (s *Span) End() { + if s == nil { + return + } + s.internal.End() +} + +// SpanContext returns the SpanContext of the span. +func (s *Span) SpanContext() SpanContext { + if s == nil { + return SpanContext{} + } + return s.internal.SpanContext() +} + +// SetName sets the name of the span, if it is recording events. +func (s *Span) SetName(name string) { + if !s.IsRecordingEvents() { + return + } + s.internal.SetName(name) +} + +// SetStatus sets the status of the span, if it is recording events. +func (s *Span) SetStatus(status Status) { + if !s.IsRecordingEvents() { + return + } + s.internal.SetStatus(status) +} + +// AddAttributes sets attributes in the span. +// +// Existing attributes whose keys appear in the attributes parameter are overwritten. +func (s *Span) AddAttributes(attributes ...Attribute) { + if !s.IsRecordingEvents() { + return + } + s.internal.AddAttributes(attributes...) +} + +// Annotate adds an annotation with attributes. +// Attributes can be nil. +func (s *Span) Annotate(attributes []Attribute, str string) { + if !s.IsRecordingEvents() { + return + } + s.internal.Annotate(attributes, str) +} + +// Annotatef adds an annotation with attributes. +func (s *Span) Annotatef(attributes []Attribute, format string, a ...interface{}) { + if !s.IsRecordingEvents() { + return + } + s.internal.Annotatef(attributes, format, a...) +} + +// AddMessageSendEvent adds a message send event to the span. +// +// messageID is an identifier for the message, which is recommended to be +// unique in this span and the same between the send event and the receive +// event (this allows to identify a message between the sender and receiver). +// For example, this could be a sequence id. +func (s *Span) AddMessageSendEvent(messageID, uncompressedByteSize, compressedByteSize int64) { + if !s.IsRecordingEvents() { + return + } + s.internal.AddMessageSendEvent(messageID, uncompressedByteSize, compressedByteSize) +} + +// AddMessageReceiveEvent adds a message receive event to the span. +// +// messageID is an identifier for the message, which is recommended to be +// unique in this span and the same between the send event and the receive +// event (this allows to identify a message between the sender and receiver). +// For example, this could be a sequence id. +func (s *Span) AddMessageReceiveEvent(messageID, uncompressedByteSize, compressedByteSize int64) { + if !s.IsRecordingEvents() { + return + } + s.internal.AddMessageReceiveEvent(messageID, uncompressedByteSize, compressedByteSize) +} + +// AddLink adds a link to the span. +func (s *Span) AddLink(l Link) { + if !s.IsRecordingEvents() { + return + } + s.internal.AddLink(l) +} + +// String prints a string representation of a span. +func (s *Span) String() string { + if s == nil { + return "" + } + return s.internal.String() +} diff --git a/vendor/go.opencensus.io/trace/trace_go11.go b/vendor/go.opencensus.io/trace/trace_go11.go new file mode 100644 index 00000000000..b8fc1e495a9 --- /dev/null +++ b/vendor/go.opencensus.io/trace/trace_go11.go @@ -0,0 +1,33 @@ +// Copyright 2018, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//go:build go1.11 +// +build go1.11 + +package trace + +import ( + "context" + t "runtime/trace" +) + +func startExecutionTracerTask(ctx context.Context, name string) (context.Context, func()) { + if !t.IsEnabled() { + // Avoid additional overhead if + // runtime/trace is not enabled. + return ctx, func() {} + } + nctx, task := t.NewTask(ctx, name) + return nctx, task.End +} diff --git a/vendor/go.opencensus.io/trace/trace_nongo11.go b/vendor/go.opencensus.io/trace/trace_nongo11.go new file mode 100644 index 00000000000..da488fc8740 --- /dev/null +++ b/vendor/go.opencensus.io/trace/trace_nongo11.go @@ -0,0 +1,26 @@ +// Copyright 2018, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +//go:build !go1.11 +// +build !go1.11 + +package trace + +import ( + "context" +) + +func startExecutionTracerTask(ctx context.Context, name string) (context.Context, func()) { + return ctx, func() {} +} diff --git a/vendor/go.opencensus.io/trace/tracestate/tracestate.go b/vendor/go.opencensus.io/trace/tracestate/tracestate.go new file mode 100644 index 00000000000..2d6c713eb3a --- /dev/null +++ b/vendor/go.opencensus.io/trace/tracestate/tracestate.go @@ -0,0 +1,147 @@ +// Copyright 2018, OpenCensus Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Package tracestate implements support for the Tracestate header of the +// W3C TraceContext propagation format. +package tracestate + +import ( + "fmt" + "regexp" +) + +const ( + keyMaxSize = 256 + valueMaxSize = 256 + maxKeyValuePairs = 32 +) + +const ( + keyWithoutVendorFormat = `[a-z][_0-9a-z\-\*\/]{0,255}` + keyWithVendorFormat = `[a-z][_0-9a-z\-\*\/]{0,240}@[a-z][_0-9a-z\-\*\/]{0,13}` + keyFormat = `(` + keyWithoutVendorFormat + `)|(` + keyWithVendorFormat + `)` + valueFormat = `[\x20-\x2b\x2d-\x3c\x3e-\x7e]{0,255}[\x21-\x2b\x2d-\x3c\x3e-\x7e]` +) + +var keyValidationRegExp = regexp.MustCompile(`^(` + keyFormat + `)$`) +var valueValidationRegExp = regexp.MustCompile(`^(` + valueFormat + `)$`) + +// Tracestate represents tracing-system specific context in a list of key-value pairs. Tracestate allows different +// vendors propagate additional information and inter-operate with their legacy Id formats. +type Tracestate struct { + entries []Entry +} + +// Entry represents one key-value pair in a list of key-value pair of Tracestate. +type Entry struct { + // Key is an opaque string up to 256 characters printable. It MUST begin with a lowercase letter, + // and can only contain lowercase letters a-z, digits 0-9, underscores _, dashes -, asterisks *, and + // forward slashes /. + Key string + + // Value is an opaque string up to 256 characters printable ASCII RFC0020 characters (i.e., the + // range 0x20 to 0x7E) except comma , and =. + Value string +} + +// Entries returns a slice of Entry. +func (ts *Tracestate) Entries() []Entry { + if ts == nil { + return nil + } + return ts.entries +} + +func (ts *Tracestate) remove(key string) *Entry { + for index, entry := range ts.entries { + if entry.Key == key { + ts.entries = append(ts.entries[:index], ts.entries[index+1:]...) + return &entry + } + } + return nil +} + +func (ts *Tracestate) add(entries []Entry) error { + for _, entry := range entries { + ts.remove(entry.Key) + } + if len(ts.entries)+len(entries) > maxKeyValuePairs { + return fmt.Errorf("adding %d key-value pairs to current %d pairs exceeds the limit of %d", + len(entries), len(ts.entries), maxKeyValuePairs) + } + ts.entries = append(entries, ts.entries...) + return nil +} + +func isValid(entry Entry) bool { + return keyValidationRegExp.MatchString(entry.Key) && + valueValidationRegExp.MatchString(entry.Value) +} + +func containsDuplicateKey(entries ...Entry) (string, bool) { + keyMap := make(map[string]int) + for _, entry := range entries { + if _, ok := keyMap[entry.Key]; ok { + return entry.Key, true + } + keyMap[entry.Key] = 1 + } + return "", false +} + +func areEntriesValid(entries ...Entry) (*Entry, bool) { + for _, entry := range entries { + if !isValid(entry) { + return &entry, false + } + } + return nil, true +} + +// New creates a Tracestate object from a parent and/or entries (key-value pair). +// Entries from the parent are copied if present. The entries passed to this function +// are inserted in front of those copied from the parent. If an entry copied from the +// parent contains the same key as one of the entry in entries then the entry copied +// from the parent is removed. See add func. +// +// An error is returned with nil Tracestate if +// 1. one or more entry in entries is invalid. +// 2. two or more entries in the input entries have the same key. +// 3. the number of entries combined from the parent and the input entries exceeds maxKeyValuePairs. +// (duplicate entry is counted only once). +func New(parent *Tracestate, entries ...Entry) (*Tracestate, error) { + if parent == nil && len(entries) == 0 { + return nil, nil + } + if entry, ok := areEntriesValid(entries...); !ok { + return nil, fmt.Errorf("key-value pair {%s, %s} is invalid", entry.Key, entry.Value) + } + + if key, duplicate := containsDuplicateKey(entries...); duplicate { + return nil, fmt.Errorf("contains duplicate keys (%s)", key) + } + + tracestate := Tracestate{} + + if parent != nil && len(parent.entries) > 0 { + tracestate.entries = append([]Entry{}, parent.entries...) + } + + err := tracestate.add(entries) + if err != nil { + return nil, err + } + return &tracestate, nil +} diff --git a/vendor/modules.txt b/vendor/modules.txt index a756dc6cf81..c3874e4ae12 100644 --- a/vendor/modules.txt +++ b/vendor/modules.txt @@ -104,6 +104,21 @@ github.com/containerd/containerd/pkg/userns # github.com/davecgh/go-spew v1.1.1 ## explicit github.com/davecgh/go-spew/spew +# github.com/dgraph-io/badger/v4 v4.2.0 +## explicit; go 1.19 +github.com/dgraph-io/badger/v4 +github.com/dgraph-io/badger/v4/fb +github.com/dgraph-io/badger/v4/options +github.com/dgraph-io/badger/v4/pb +github.com/dgraph-io/badger/v4/skl +github.com/dgraph-io/badger/v4/table +github.com/dgraph-io/badger/v4/trie +github.com/dgraph-io/badger/v4/y +# github.com/dgraph-io/ristretto v0.1.1 +## explicit; go 1.12 +github.com/dgraph-io/ristretto +github.com/dgraph-io/ristretto/z +github.com/dgraph-io/ristretto/z/simd # github.com/docker/docker v24.0.9+incompatible ## explicit github.com/docker/docker/api/types/blkiodev @@ -169,6 +184,14 @@ github.com/go-task/slim-sprig github.com/gogo/protobuf/gogoproto github.com/gogo/protobuf/proto github.com/gogo/protobuf/protoc-gen-gogo/descriptor +# github.com/golang/glog v1.1.2 +## explicit; go 1.19 +github.com/golang/glog +github.com/golang/glog/internal/logsink +github.com/golang/glog/internal/stackdump +# github.com/golang/groupcache v0.0.0-20210331224755-41bb18bfe9da +## explicit +github.com/golang/groupcache/lru # github.com/golang/protobuf v1.5.3 ## explicit; go 1.9 github.com/golang/protobuf/jsonpb @@ -181,6 +204,9 @@ github.com/golang/protobuf/ptypes/timestamp # github.com/golang/snappy v0.0.4 ## explicit github.com/golang/snappy +# github.com/google/flatbuffers v1.12.1 +## explicit +github.com/google/flatbuffers/go # github.com/google/go-cmp v0.6.0 ## explicit; go 1.13 github.com/google/go-cmp/cmp @@ -513,6 +539,13 @@ go.etcd.io/etcd/server/v3/etcdserver/api/snap go.etcd.io/etcd/server/v3/etcdserver/api/snap/snappb go.etcd.io/etcd/server/v3/wal go.etcd.io/etcd/server/v3/wal/walpb +# go.opencensus.io v0.24.0 +## explicit; go 1.13 +go.opencensus.io +go.opencensus.io/internal +go.opencensus.io/trace +go.opencensus.io/trace/internal +go.opencensus.io/trace/tracestate # go.uber.org/multierr v1.11.0 ## explicit; go 1.19 go.uber.org/multierr