DanielLeens commented on issue #12413: URL: https://github.com/apache/seatunnel/issues/12413#issuecomment-5760330010
Classification: A / Zeta cancellation and final-metrics lifecycle. I re-checked current `dev`. When the last task completes, `TaskGroupExecutionTracker.taskDone()` calls `finishExecution(...)` and then calls `updateMetricsContextInImap()` synchronously before it completes the task-group future. That method creates a Hazelcast invocation and waits with `invoke.get()`. Its interruption handling records the event and restores the interrupt flag, but it does not move the invocation away from the interrupted `BlockingWorker`. That confirms the lifecycle risk described here. PR #12265 is about coordinator admission/lifecycle isolation and does not change this path; I found no open focused PR for it. A focused fix is welcome. It must keep the existing ordering that the terminal task-group result is published only after the final metrics attempt, but the attempt itself must run outside the cancelled worker. Please include the unit regression and the two-member case from this report: prove that a late interrupt cannot close the existing worker-master connection, that the terminal state still arrives within the asserted bound, and that a failed final metrics attempt remains best-effort. Please keep this separate from #12265 and link the PR here before starting so there is one implementation path. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
