Vivek1106-04 commented on PR #12623: URL: https://github.com/apache/seatunnel/pull/12623#issuecomment-6011357726
@SEZ9 Thanks for checking the follow-up. **F1 (merge order).** The description already makes the dependency explicit: "this PR should merge after #12511: on its own, a stop during the restore wait still lets the restore thread re-apply resources and briefly report the pipeline `RUNNING` before it ends; the #12511 re-check removes that round-trip." The "independent" wording is gone. I would rather not add the guard here: it would be the same post-sleep `isNeedRestore()` check that #12511 adds to `SubPlan`, and whichever PR lands second would have to remove the duplicate. **F2 (pipeline-level test).** I have updated the description to list all three new vertex tests and to link the pipeline-level assertions in #12511: `SavePointTest#testSavePointFailureDuringPipelineRestoreWaitEndsTheJob` and `CheckpointErrorRestoreEndTest#testCancelDuringPipelineRestoreWaitEndsTheJobCanceled`. I did not add a `SubPlan`-level test here on purpose. Once #12511 is in, a stop during the restore wait is ended by its re-check (`abandonRestore`), which does not wait on the task futures, so a `forceStopPipeline()`-during-restore-wait test would pass with or without this fix and could not catch a regression of it. The window only this PR closes is a stop that lands after that re-check and before the restarted tasks are deployed, which is a race that a pipeline-level test cannot hit deterministically. The vertex tests pin the contract that window relies on: `forceStop()` on a task whose state process is stopped completes its future with `CANCELED`, both for a fresh task and for one reset for restore. Both fail if the fix is reverted, regardless of `SubPlan`. The third test guards the fix itself: when the transition is skipped, the state process the fix started is stopped again. -- 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]
