SEZ9 commented on PR #11382: URL: https://github.com/apache/seatunnel/pull/11382#issuecomment-5594818673
@DanielLeens thanks for tracing the second shape rather than folding it into the F2 path — the distinction matters, and I agree with what you laid out: on the task-level restart via `SubPlan.prepareRestorePipeline()` (`SubPlan.java:498`) → `reportedPipelineRunning(pipelineId, false)` → `getCheckpointCoordinator(pipelineId)` (`CheckpointManager.java:212`), the coordinator is looked up from `coordinatorMap`, and `restoreCoordinator(false)` (`CheckpointCoordinator.java:706-708`) only does `isAllTaskReady.set(false)`, so `latestCompletedCheckpoint` stays the same in-memory object with no re-deserialization. One thing on my side: your comment is cut off at "nothing in `cleanPendingChe…", so I'm missing the conclusion of the branch. Could you post the rest? Specifically, for this coordinator-survives shape I want to be sure of the answer to the question that keeps F2/F4/F5 open or closed: - The tasks *are* recreated in this shape, so the `SchemaChangeDrainGuard` drain-ready state (process memory only, F4) is gone on the sink side. Since the coordinator is not reconstructed and `latestCompletedCheckpoint` is not re-processed, what code path re-sends the typed `CheckpointFinishedOperation` (or a typed abort, F5) to the new task instances? If the answer is "nothing does, but the guard doesn't need it because X", please spell out X — that's the part I can't confirm from the truncated text. - If there is no re-delivery in this branch, then the withdrawal of F2 only covers the coordinator-recreated shape, and I'd want to keep F2 open (narrowed to the task-level-restart case) rather than closed. Once I have the remaining half of your trace I'll update the F2/F4/F5 status accordingly. The other points (F1 race, F3/F8 wire-format safety, F6 single-slot tracking, F7 wiring/e2e coverage) stand as previously discussed and aren't affected by this branch. <!-- streview-comment:903 --> -- 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]
