Repository navigation
v3.33.1
Postgres-backed deployments no longer silently stop scheduling work after a transient database blip, and Redis concurrency limits resolve to the right DAO on startup.
What's Changed
- fix(postgres-persistence): keep the unack schedule alive after a failed run (#1789) —
PostgresQueueDAOranprocessAllUnacksonscheduleAtFixedRatewith nothing catching inside the callback, so a single escaped exception (a dropped connection -- the retry template only retries deadlocks and serialization failures) cancelled every later run. From then on no popped message was ever un-popped, and workflows only the sweeper could advance stalled past their timeouts until a restart. The run is now wrapped so a failure is logged and the next tick picks the work back up once the database is reachable. Thanks @khuongntrd. (Fixes #326) - fix(postgres-persistence): reset popped flag when re-pushing queue messages (#1761) — re-pushing a queue message left its
poppedflag set, so the re-queued message could be skipped on the next poll. The flag is now reset on re-push. Thanks @noy-solvin. (Fixes #1667) - fix: mark RedisConcurrentExecutionLimitDAO as primary bean (#1765) — with more than one
ConcurrentExecutionLimitDAOon the classpath, Spring could not resolve which bean to inject for Redis concurrency limits. Marking the Redis implementation@Primarydisambiguates it. Thanks @utafrali. (Fixes #1753) - fix(publish): promote to Maven Central once per release, not once per module (#1826) — the Central promote step lived inside
subprojects {}and fired once per module, fanning a single release out into 51 deployments against the org's monthly quota. It is now one root-level task that runs after all module publishes. Thanks @nthmost-orkes.
Full Changelog: v3.33.0...v3.33.1