This release focuses on distributed worker stability, shared-nothing execution isolation, and memory bounding across the event pipeline. Distributed workers now properly clean up workspace bundles and release coordinator resources on completion, alongside extended lease grace periods to prevent false expiration under transient coordinator load.
Fixed
- fix: preserve rescheduled file dependencies (#2618) @yohamta0
- fix: restore notification event delivery (#2619) @yohamta0
- fix: reduce event replay allocations (#2620) @yohamta0
- fix: reduce pending event allocations (#2621) @yohamta0
- perf: cache retry candidate files (#2622) @yohamta0
- fix: compact retained event IDs (#2624) @yohamta0
- fix: bound retry candidate cache (#2625) @yohamta0
- Fix router diagnostic value resolution (#2626) @yohamta0
- fix: bound event deduplication memory (#2627) @yohamta0
- fix: bound pending notification state (#2628) @yohamta0
- fix: reclaim workspace bundle storage (#2630) @yohamta0
- fix: extend distributed lease grace (#2631) @yohamta0
- fix: preserve monitor startup events (#2632) @yohamta0
- Fix remote context and worker deployment guidance (#2639) @yohamta0
- fix: release coordinator run resources (#2640) @yohamta0
- fix: make distributed workers shared-nothing (#2641) @yohamta0
Contributors
Thanks to our contributors for this release:
| Contribution | Contributor |
|---|---|
| bug: distributed run lease expired: worker dagu-worker-general-558449b8dd-b8qjh accepted the task claim ... failed to check coordinator health (#2629) | @pommetjehorlepiep (report) |
Full Changelog: v2.15.3...v2.15.4