Patch release 2.3.1 (#20021)

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: Etay Livne <53942171+EtayLivne@users.noreply.github.com>
Co-authored-by: Etay Livne <etay.livne@mobileye.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: thomas chaton <thomas@grid.ai>

.github/checkgroup.yml

@@ -242,9 +242,9 @@ subprojects:
       - "!*.md"
       - "!**/*.md"
     checks:
-      - "app-pytest (macOS-11, lightning, 3.8, latest)"
-      - "app-pytest (macOS-11, lightning, 3.8, oldest)"
-      - "app-pytest (macOS-11, app, 3.9, latest)"
+      - "app-pytest (macOS-12, lightning, 3.8, latest)"
+      - "app-pytest (macOS-12, lightning, 3.8, oldest)"
+      - "app-pytest (macOS-12, app, 3.9, latest)"
       - "app-pytest (macOS-12, app, 3.11, latest)"
       - "app-pytest (ubuntu-20.04, lightning, 3.8, latest)"
       - "app-pytest (ubuntu-20.04, lightning, 3.8, oldest)"
@@ -270,9 +270,9 @@ subprojects:
       - "!*.md"
       - "!**/*.md"
     checks:
-      - "app-examples (macOS-11, lightning, 3.9, latest)"
-      - "app-examples (macOS-11, lightning, 3.9, oldest)"
-      - "app-examples (macOS-11, app, 3.9, latest)"
+      - "app-examples (macOS-12, lightning, 3.9, latest)"
+      - "app-examples (macOS-12, lightning, 3.9, oldest)"
+      - "app-examples (macOS-12, app, 3.9, latest)"
       - "app-examples (ubuntu-20.04, lightning, 3.9, latest)"
       - "app-examples (ubuntu-20.04, lightning, 3.9, oldest)"
       - "app-examples (ubuntu-20.04, app, 3.9, latest)"

.github/workflows/ci-examples-app.yml

@@ -36,13 +36,13 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-20.04, macOS-11, windows-2022]
+        os: [ubuntu-20.04, macOS-12, windows-2022]
         pkg-name: ["lightning"]
         python-version: ["3.9"]
         requires: ["oldest", "latest"]
         include:
           # "app" installs the standalone package
-          - { os: "macOS-11", pkg-name: "app", python-version: "3.9", requires: "latest" }
+          - { os: "macOS-12", pkg-name: "app", python-version: "3.9", requires: "latest" }
           - { os: "ubuntu-20.04", pkg-name: "app", python-version: "3.9", requires: "latest" }
           - { os: "windows-2022", pkg-name: "app", python-version: "3.9", requires: "latest" }
     # Timeout: https://stackoverflow.com/a/59076067/4521646

.github/workflows/ci-tests-app.yml

@@ -38,7 +38,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: ["ubuntu-20.04", "macOS-11", "windows-2022"]
+        os: ["ubuntu-20.04", "macOS-12", "windows-2022"]
         pkg-name: ["lightning"]
         python-version: ["3.8"]
         requires: ["oldest", "latest"]
@@ -48,7 +48,7 @@ jobs:
           - { os: "ubuntu-22.04", pkg-name: "app", python-version: "3.11", requires: "latest" }
           - { os: "windows-2022", pkg-name: "app", python-version: "3.11", requires: "latest" }
           # "app" installs the standalone package
-          - { os: "macOS-11", pkg-name: "app", python-version: "3.9", requires: "latest" }
+          - { os: "macOS-12", pkg-name: "app", python-version: "3.9", requires: "latest" }
           - { os: "ubuntu-20.04", pkg-name: "app", python-version: "3.9", requires: "latest" }
           - { os: "windows-2022", pkg-name: "app", python-version: "3.8", requires: "latest" }
     # Timeout: https://stackoverflow.com/a/59076067/4521646

.github/workflows/docker-build.yml

@@ -83,7 +83,7 @@ jobs:
               gh_env.write("DOCKER_TAGS=" + ",".join(tags))
         shell: python
 
-      - uses: docker/build-push-action@v5
+      - uses: docker/build-push-action@v6
         with:
           build-args: |
             PYTHON_VERSION=${{ matrix.python_version }}
@@ -119,7 +119,7 @@ jobs:
         with:
           username: ${{ secrets.DOCKER_USERNAME }}
           password: ${{ secrets.DOCKER_PASSWORD }}
-      - uses: docker/build-push-action@v5
+      - uses: docker/build-push-action@v6
         with:
           build-args: |
             PYTHON_VERSION=${{ matrix.python_version }}
@@ -151,7 +151,7 @@ jobs:
       - name: Build Conda Docker
         # publish master/release
         continue-on-error: true
-        uses: docker/build-push-action@v5
+        uses: docker/build-push-action@v6
         with:
           file: dockers/nvidia/Dockerfile
           push: false

pyproject.toml

@@ -140,6 +140,7 @@ exclude = [
     "src/lightning/app/cli/component-template",
     "src/lightning/app/cli/pl-app-template",
     "src/lightning/app/cli/react-ui-template",
+    "src/lightning/app/launcher/utils.py",
 ]
 install_types = "True"
 non_interactive = "True"

requirements/app/app.txt

@@ -1,12 +1,12 @@
-lightning-cloud == 0.5.69  # Must be pinned to ensure compatibility
+lightning-cloud == 0.5.70  # Must be pinned to ensure compatibility
 packaging
 typing-extensions >=4.4.0, <4.10.0
 deepdiff >=5.7.0, <6.6.0
 fsspec[http] >=2022.5.0, <2023.11.0
 croniter >=1.3.0, <1.5.0  # strict; TODO: for now until we find something more robust.
 traitlets >=5.3.0, <5.12.0
 arrow >=1.2.0, <1.3.0
-lightning-utilities >=0.8.0, <0.12.0
+lightning-utilities >=0.10.0, <0.12.0
 beautifulsoup4 >=4.8.0, <4.13.0
 inquirer >=2.10.0, <3.2.0
 psutil <5.9.6
@@ -27,3 +27,5 @@ urllib3 <2.0.0
 uvicorn <0.24.0
 websocket-client <1.7.0
 websockets <11.1.0
+numpy >=1.17.2, <2.0
+msgpack

requirements/ci.txt

@@ -1,6 +1,7 @@
-setuptools
-wheel
+setuptools <70.1.1
+wheel <0.44.0
 awscli >=1.30.0, <1.31.0
 twine ==4.0.1
+importlib-metadata <8.0.0
 wget
-packaging
+packaging <24.2

requirements/fabric/base.txt

@@ -6,4 +6,4 @@ torch >=2.0.0, <2.4.0
 fsspec[http] >=2022.5.0, <2024.4.0
 packaging >=20.0, <=23.1
 typing-extensions >=4.4.0, <4.10.0
-lightning-utilities >=0.8.0, <0.12.0
+lightning-utilities >=0.10.0, <0.12.0

requirements/pytorch/base.txt

@@ -9,4 +9,4 @@ fsspec[http] >=2022.5.0, <2024.4.0
 torchmetrics >=0.7.0, <1.3.0  # needed for using fixed compare_version
 packaging >=20.0, <=23.1
 typing-extensions >=4.4.0, <4.10.0
-lightning-utilities >=0.8.0, <0.12.0
+lightning-utilities >=0.10.0, <0.12.0

requirements/pytorch/extra.txt

@@ -3,8 +3,8 @@
 
 # extended list of package dependencies to reach full functionality
 matplotlib>3.1, <3.9.0
-omegaconf >=2.0.5, <2.4.0
-hydra-core >=1.0.5, <1.4.0
+omegaconf >=2.2.3, <2.4.0
+hydra-core >=1.2.0, <1.4.0
 jsonargparse[signatures] >=4.27.7, <4.28.0
 rich >=12.3.0, <13.6.0
 tensorboardX >=2.2, <2.7.0  # min version is set by torch.onnx missing attribute

src/lightning/app/cli/lightning_cli.py

@@ -40,7 +40,19 @@
 from lightning.app.cli.lightning_cli_delete import delete
 from lightning.app.cli.lightning_cli_launch import launch
 from lightning.app.cli.lightning_cli_list import get_list
-from lightning.app.core.constants import ENABLE_APP_COMMENT_COMMAND_EXECUTION, get_lightning_cloud_url
+from lightning.app.core.constants import (
+    APP_SERVER_HOST,
+    APP_SERVER_PORT,
+    ENABLE_APP_COMMENT_COMMAND_EXECUTION,
+    get_lightning_cloud_url,
+)
+from lightning.app.launcher.launcher import (
+    run_lightning_flow,
+    run_lightning_work,
+    serve_frontend,
+    start_application_server,
+    start_flow_and_servers,
+)
 from lightning.app.runners.cloud import CloudRuntime
 from lightning.app.runners.runtime import dispatch
 from lightning.app.runners.runtime_type import RuntimeType
@@ -393,3 +405,99 @@ def _prepare_file(file: str) -> str:
         return file
 
     raise FileNotFoundError(f"The provided file {file} hasn't been found.")
+
+
+@run.command("server")
+@click.argument("file", type=click.Path(exists=True))
+@click.option("--queue-id", help="ID for identifying queue", default="", type=str)
+@click.option("--host", help="Application running host", default=APP_SERVER_HOST, type=str)
+@click.option("--port", help="Application running port", default=APP_SERVER_PORT, type=int)
+def run_server(file: str, queue_id: str, host: str, port: int) -> None:
+    """It takes the application file as input, build the application object and then use that to run the application
+    server.
+
+    This is used by the cloud runners to start the status server for the application
+
+    """
+    logger.debug(f"Run Server: {file} {queue_id} {host} {port}")
+    start_application_server(file, host, port, queue_id=queue_id)
+
+
+@run.command("flow")
+@click.argument("file", type=click.Path(exists=True))
+@click.option("--queue-id", help="ID for identifying queue", default="", type=str)
+@click.option("--base-url", help="Base url at which the app server is hosted", default="")
+def run_flow(file: str, queue_id: str, base_url: str) -> None:
+    """It takes the application file as input, build the application object, proxy all the work components and then run
+    the application flow defined in the root component.
+
+    It does exactly what a singleprocess dispatcher would do but with proxied work components.
+
+    """
+    logger.debug(f"Run Flow: {file} {queue_id} {base_url}")
+    run_lightning_flow(file, queue_id=queue_id, base_url=base_url)
+
+
+@run.command("work")
+@click.argument("file", type=click.Path(exists=True))
+@click.option("--work-name", type=str)
+@click.option("--queue-id", help="ID for identifying queue", default="", type=str)
+def run_work(file: str, work_name: str, queue_id: str) -> None:
+    """Unlike other entrypoints, this command will take the file path or module details for a work component and run
+    that by fetching the states from the queues."""
+    logger.debug(f"Run Work: {file} {work_name} {queue_id}")
+    run_lightning_work(
+        file=file,
+        work_name=work_name,
+        queue_id=queue_id,
+    )
+
+
+@run.command("frontend")
+@click.argument("file", type=click.Path(exists=True))
+@click.option("--flow-name")
+@click.option("--host")
+@click.option("--port", type=int)
+def run_frontend(file: str, flow_name: str, host: str, port: int) -> None:
+    """Serve the frontend specified by the given flow."""
+    logger.debug(f"Run Frontend: {file} {flow_name} {host}")
+    serve_frontend(file=file, flow_name=flow_name, host=host, port=port)
+
+
+@run.command("flow-and-servers")
+@click.argument("file", type=click.Path(exists=True))
+@click.option("--queue-id", help="ID for identifying queue", default="", type=str)
+@click.option("--base-url", help="Base url at which the app server is hosted", default="")
+@click.option("--host", help="Application running host", default=APP_SERVER_HOST, type=str)
+@click.option("--port", help="Application running port", default=APP_SERVER_PORT, type=int)
+@click.option(
+    "--flow-port",
+    help="Pair of flow name and frontend port",
+    type=(str, int),
+    multiple=True,
+)
+def run_flow_and_servers(
+    file: str,
+    base_url: str,
+    queue_id: str,
+    host: str,
+    port: int,
+    flow_port: Tuple[Tuple[str, int]],
+) -> None:
+    """It takes the application file as input, build the application object and then use that to run the application
+    flow defined in the root component, the application server and all the flow frontends.
+
+    This is used by the cloud runners to start the flow, the status server and all frontends for the application
+
+    """
+    logger.debug(f"Run Flow: {file} {queue_id} {base_url}")
+    logger.debug(f"Run Server: {file} {queue_id} {host} {port}.")
+    logger.debug(f"Run Frontend's: {flow_port}")
+    start_flow_and_servers(
+        entrypoint_file=file,
+        base_url=base_url,
+        queue_id=queue_id,
+        host=host,
+        port=port,
+        flow_names_and_ports=flow_port,
+    )

src/lightning/app/core/app.py

@@ -35,6 +35,7 @@
     FLOW_DURATION_SAMPLES,
     FLOW_DURATION_THRESHOLD,
     FRONTEND_DIR,
+    SHOULD_START_WORKS_WITH_FLOW,
     STATE_ACCUMULATE_WAIT,
 )
 from lightning.app.core.queues import BaseQueue
@@ -144,6 +145,7 @@ def __init__(
         self.threads: List[threading.Thread] = []
         self.exception = None
         self.collect_changes: bool = True
+        self._should_start_works_with_flow: bool = SHOULD_START_WORKS_WITH_FLOW
 
         self.status: Optional[AppStatus] = None
         # TODO: Enable ready locally for opening the UI.
@@ -733,6 +735,9 @@ def _send_flow_to_work_deltas(self, state: dict) -> None:
                 self.flow_to_work_delta_queues[w.name].put(deep_diff)
 
     def _start_with_flow_works(self) -> None:
+        if not self._should_start_works_with_flow:
+            return
+
         for w in self.works:
             if w._start_with_flow:
                 parallel = w.parallel

src/lightning/app/core/constants.py

@@ -12,9 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import json
 import os
 from pathlib import Path
-from typing import Optional
+from typing import Any, Optional
 
 import lightning_cloud.env
 
@@ -101,6 +102,37 @@ def get_lightning_cloud_url() -> str:
 
 BATCH_DELTA_COUNT = int(os.getenv("BATCH_DELTA_COUNT", "128"))
 CHECK_ERROR_QUEUE_INTERVAL = float(os.getenv("CHECK_ERROR_QUEUE_INTERVAL", "30"))
+SHOULD_START_WORKS_WITH_FLOW = bool(int(os.getenv("SHOULD_START_WORKS_WITH_FLOW", "1")))
+IS_RUNNING_IN_FLOW = os.getenv("LIGHTNING_CLOUD_WORK_NAME", None) is None
+
+
+class DistributedPluginChecker:
+    def __init__(self) -> None:
+        self.distributed_arguments = os.getenv("DISTRIBUTED_ARGUMENTS", None)
+        if self.distributed_arguments:
+            self.distributed_arguments = json.loads(self.distributed_arguments)
+
+        self.running_distributed_plugin = False
+
+        if self.distributed_arguments and os.getenv("LIGHTNING_CLOUD_WORK_NAME"):
+            self.running_distributed_plugin = True
+
+    def __bool__(self) -> bool:
+        return self.running_distributed_plugin
+
+    def should_create_work(self, work: Any) -> bool:
+        if not self.distributed_arguments:
+            return True
+
+        num_nodes = self.distributed_arguments.get("num_instances", 0)
+        node_rank = int(work.name.split(".")[-1])
+
+        # Only the start with flow works are skipped for performance purposes
+        return node_rank >= num_nodes
+
+
+# TODO (tchaton): Add LitData and JobPlugin optimizations
+PLUGIN_CHECKER = IS_DISTRIBUTED_PLUGIN = DistributedPluginChecker()
 
 
 def enable_multiple_works_in_default_container() -> bool:

src/lightning/app/core/flow.py

@@ -836,6 +836,11 @@ def load_state_dict(self, flow_state, children_states, strict) -> None:
             elif strict:
                 raise ValueError(f"The component {child_name} wasn't instantiated for the component {self.name}")
 
+    def stop_works(self, works: List[Any]) -> None:
+        if self._backend is None:
+            raise RuntimeError("Your flow should have a backend attached. Found None.")
+        self._backend.stop_works(works)
+
 
 class _RootFlow(LightningFlow):
     def __init__(self, work: LightningWork) -> None: