reduce TransitEvent struct size

CHANGELOG.md

@@ -84,6 +84,8 @@
   backend thread has increased throughput around 5%
 - Simplified the `TransitEventBuffer` in the backend worker thread, resulting in a minor throughput improvement of
   approximately 1%.
+- Optimised the size of the `TransitEvent` structure, reducing its memory footprint and improving backend thread
+  throughput by an additional 3%.
 - Added an optional fsync interval to control the minimum time between consecutive fsync calls, reducing disk wear from
   frequent fsync operations. This option is only applicable when fsync is
   enabled. ([#557](https://github.com/odygrd/quill/issues/557))

CMakeLists.txt

@@ -187,7 +187,6 @@ set(HEADER_FILES
         include/quill/core/DynamicFormatArgStore.h
         include/quill/core/Codec.h
         include/quill/core/Filesystem.h
-        include/quill/core/FormatBuffer.h
         include/quill/core/FrontendOptions.h
         include/quill/core/InlinedVector.h
         include/quill/core/LoggerBase.h

benchmarks/backend_throughput/quill_backend_throughput.cpp

@@ -14,16 +14,15 @@ static constexpr size_t total_iterations = 4'000'000;
 int main()
 {
   // main thread affinity
-  quill::detail::set_cpu_affinity(0);
+  quill::detail::set_cpu_affinity(1);
 
   quill::BackendOptions backend_options;
   backend_options.cpu_affinity = 5;
+  backend_options.sleep_duration = std::chrono::nanoseconds{0};
 
   // Start the logging backend thread and give it some tiem to init
   quill::Backend::start(backend_options);
 
-  std::this_thread::sleep_for(std::chrono::milliseconds{100});
-
   // Create a file sink to write to a file
   std::shared_ptr<quill::Sink> file_sink = quill::Frontend::create_or_get_sink<quill::FileSink>(
     "quill_backend_total_time.log",
@@ -41,7 +40,8 @@ int main()
       "%(time) [%(thread_id)] %(short_source_location) %(log_level) %(message)", "%H:%M:%S.%Qns",
       quill::Timezone::LocalTime, false});
 
-  quill::Frontend::preallocate();
+  LOG_INFO(logger, "preallocate");
+  logger->flush_log(0);
 
   // start counting the time until backend worker finishes
   auto const start_time = std::chrono::steady_clock::now();

benchmarks/hot_path_latency/quill_hot_path_rdtsc_clock.cpp

@@ -28,18 +28,13 @@ void quill_benchmark(std::vector<uint16_t> const& thread_count_array,
   /** - MAIN THREAD START - Logger setup if any **/
 
   /** - Setup Quill **/
-  // main thread affinity
-  quill::detail::set_cpu_affinity(0);
-
   quill::BackendOptions backend_options;
   backend_options.cpu_affinity = 5;
   backend_options.sleep_duration = std::chrono::nanoseconds{0};
 
   // Start the logging backend thread and give it some tiem to init
   quill::Backend::start(backend_options);
 
-  std::this_thread::sleep_for(std::chrono::milliseconds{100});
-
   // wait for the backend thread to start
   std::this_thread::sleep_for(std::chrono::seconds(1));
 
@@ -62,9 +57,11 @@ void quill_benchmark(std::vector<uint16_t> const& thread_count_array,
 
   /** LOGGING THREAD FUNCTIONS - on_start, on_exit, log_func must be implemented **/
   /** those run on a several thread(s). It can be one or multiple threads based on THREAD_LIST_COUNT config */
-  auto on_start = []() {
+  auto on_start = [logger]()
+  {
     // on thread start
-    Frontend::preallocate();
+    LOG_INFO(logger, "preallocate");
+    logger->flush_log(0);
   };
 
   auto on_exit = [logger]()

benchmarks/hot_path_latency/quill_hot_path_system_clock.cpp

@@ -28,9 +28,6 @@ void quill_benchmark(std::vector<uint16_t> const& thread_count_array,
   /** - MAIN THREAD START - Logger setup if any **/
 
   /** - Setup Quill **/
-  // main thread affinity
-  quill::detail::set_cpu_affinity(0);
-
   quill::BackendOptions backend_options;
   backend_options.cpu_affinity = 5;
   backend_options.sleep_duration = std::chrono::nanoseconds{0};
@@ -63,10 +60,11 @@ void quill_benchmark(std::vector<uint16_t> const& thread_count_array,
 
   /** LOGGING THREAD FUNCTIONS - on_start, on_exit, log_func must be implemented **/
   /** those run on a several thread(s). It can be one or multiple threads based on THREAD_LIST_COUNT config */
-  auto on_start = []()
+  auto on_start = [logger]()
   {
     // on thread start
-    Frontend::preallocate();
+    LOG_INFO(logger, "preallocate");
+    logger->flush_log(0);
   };
 
   auto on_exit = [logger]()

include/quill/backend/BackendOptions.h

@@ -55,7 +55,7 @@ struct BackendOptions
    * transit_events_hard_limit The backend will use a separate transit_event_buffer for each
    * frontend thread. The capacity must be a power of two.
    */
-  uint32_t transit_event_buffer_initial_capacity = 64;
+  uint32_t transit_event_buffer_initial_capacity = 128;
 
   /**
    * The backend gives priority to reading messages from the frontend queues of all