CryoThrust commented on issue #12202: URL: https://github.com/apache/seatunnel/issues/12202#issuecomment-5594827382
I can help narrow this down, but I agree that adding a timeout around `join()` before identifying the blocking operation could mask the underlying failure. A useful first diagnostic slice may be a test-only or opt-in trace around the REJECT completion path, recording: - the job id and the exact phase entered/exited in `completeFailJob()` and `JobMaster.cleanJob()`; - elapsed time for `getJobDAGInfo()`, checkpoint cleanup, history writes, and pending-job cleanup; - whether the `JobMaster` monitor is held by `cancelJob()` or `stopJob()` when the cleanup starts. For a production-safe containment contract, I would separate two concerns: (1) make cleanup observability explicit so a thread dump can distinguish monitor contention from a Hazelcast operation stall, and (2) decide whether a bounded cleanup budget should leave the job in a documented terminal-but-partially-cleaned state, with retry/reconciliation on the coordinator, rather than simply converting a timeout into FAILED. Would maintainers prefer a minimal reproduction/instrumentation PR first, or should the investigation use an existing fault-tolerance test with an opt-in diagnostic flag? I can prepare the former without changing default scheduling behavior. -- 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]
