X-Git-Url: https://git.piment-noir.org/?a=blobdiff_plain;f=CHANGELOG.md;h=93289b8dce36a22b66dca782b2e391bac47df293;hb=de91b43a597f33ea95f6e3f24fe70b9b9827429b;hp=48733b72431fd1f858ebdcf0d9231a47f794bf79;hpb=741cdb341d12048ba0f1f55784254260ab49b4fa;p=poolifier.git

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 48733b72..93289b8d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,90 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.6.34] - 2023-08-24
+
+### Fixes
+
+- Avoid cascading tasks stealing under back pressure.
+
+### Changed
+
+- Add fastpath to queued tasks rescheduling.
+
+## [2.6.33] - 2023-08-24
+
+### Fixed
+
+- Fix queued tasks rescheduling.
+
+### Changed
+
+- Rename tasks queue options `queueMaxSize` to `size`.
+
+### Added
+
+- Task stealing scheduling algorithm if tasks queueing is enabled.
+
+## [2.6.32] - 2023-08-23
+
+### Fixed
+
+- Ensure no task can be executed when the pool is destroyed.
+
+### Added
+
+- Add `queueMaxSize` option to tasks queue options.
+- Add O(1) deque implementation implemented with doubly linked list and use it for tasks queueing.
+- Add tasks stealing algorithm when a worker node queue is back pressured if tasks queueing is enabled.
+
+## [2.6.31] - 2023-08-20
+
+### Fixed
+
+- Fix worker choice strategy retries mechanism in some edge cases.
+
+### Changed
+
+- Make orthogonal worker choice strategies tasks distribution and created dynamic worker usage.
+- Remove the experimental status of the `LEAST_ELU` worker choice strategy.
+
+## [2.6.30] - 2023-08-19
+
+### Fixed
+
+- Ensure pool event `backPressure` is emitted.
+- Ensure pool event `full` is emitted only once.
+- Ensure worker node cannot be instantiated without proper arguments.
+
+## [2.6.29] - 2023-08-18
+
+### Fixed
+
+- Fix race condition between ready and task functions worker message handling at startup.
+- Fix duplicate task function worker usage statistics computation per task function.
+- Update task function worker usage statistics if and only if there's at least two different task functions.
+- Fix race condition at task function worker usage executing task computation leading to negative value.
+
+### Added
+
+- Add back pressure detection on the worker node queue. Event `backPressure` is emitted when all worker node queues are full (worker node queue size >= poolMaxSize^2).
+- Use back pressure detection in worker choice strategies.
+- Add worker choice strategies retries mechanism if no worker is eligible.
+
+## [2.6.28] - 2023-08-16
+
+### Fixed
+
+- Ensure pool workers are properly initialized.
+
+### Added
+
+- HTTP server pool examples: express-cluster, express-hybrid.
+
+### Changed
+
+- Remove now useless branching in worker hot code path.
+
 ## [2.6.27] - 2023-08-15
 
 ### Fixed
@@ -72,7 +156,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 
 - Fix queued tasks redistribution on error task execution starvation.
-- Ensure task queueing per worker condition is untangled from the pool busyness semantic.
+- Ensure tasks queueing per worker condition is untangled from the pool busyness semantic.
 
 ### Changed
 
@@ -180,7 +264,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
-- Add safe helper `availableParallelism` to help sizing the pool.
+- Add safe helper `availableParallelism()` to help sizing the pool.
 
 ### Fixed
 
@@ -195,7 +279,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Fixed
 
 - Artificial version bump to 2.6.5 to workaround publication issue.
-- Ensure cluster pool destroy() gracefully shutdowns worker's server.
+- Ensure cluster pool `destroy()` gracefully shutdowns worker's server.
 - Ensure pool event is emitted before task error promise rejection.
 - Fix queued tasks count computation.
 
@@ -211,7 +295,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
-- Ensure cluster pool destroy() gracefully shutdowns worker's server.
+- Ensure cluster pool `destroy()` gracefully shutdowns worker's server.
 - Ensure pool event is emitted before task error promise rejection.
 - Fix queued tasks count computation.