SEZ9 commented on PR #12440: URL: https://github.com/apache/seatunnel/pull/12440#issuecomment-6871239554
Thanks for pushing the restore-time reclaim in `3656e78b3a` — the direction is right, the three new tests in `HudiTwoPhaseCommitTest` are the ones I asked for, and the docs now describe the cleanup model honestly. CI is also green: Build 42974133073 and the Apache Build check pass on the current head `99f85eaa16276d7c3d431a7d131a94f166c900ec`. The one thing still blocking from my side is Issue 1 from the source review (5857369474): `HudiSinkAggregatedCommitter#reclaimOrphanedInstants` treats the Hudi table timeline as owned by a single committer. When several upstream tables in the multi-table sink map to the same Hudi base path, `restoreCommit` on one sub-committer can roll back a pending instant that another restored sub-committer still needs. That is a completed-checkpoint data-loss/recovery scenario, so a passing build does not clear it. What I need before approving (no new issues against this head): 1. Make the reclaim safe when multiple tables share one base path, while keeping the guarantee that a completed instant is never rolled back. 2. Add the shared-table restore regression test covering that scenario. 3. Update the EN/ZH docs to state which table-to-base-path mappings are supported under `EXACTLY_ONCE`. Issues 2 and 3 keep their earlier non-blocking disposition; feel free to fold them in or follow up separately. Once the fix and regression test are up I will do another full pass on the new head. <!-- streview-comment:2349 --> -- 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]
