SEZ9 commented on issue #12959: URL: https://github.com/apache/seatunnel/issues/12959#issuecomment-6407967286
Thanks @hutiefang76 — this is exactly the evidence that was missing. Fork commit `91d1078a3790cf0bcdfbda5cd235c28ec8a95219`, the five latest-stop entry points in `MysqlCDCStopModeSpecificIT`, the helper where the 60-second final-value assertion runs before the `FINISHED` wait and job exit check, and `mysqlcdc_stop_mode_latest.conf` (`stop.mode=latest`, `snapshot.split.size=20`) are enough to confirm the source location. The MySQL JDK 17 job log from run 39772122616 (141 tests, 0 failures, 1 error — timing out expecting `latest-stop` while still reading `bulk`) matches the ordering hypothesis, since the assertion expires while the snapshot phase is still in progress. Agreed with your framing: this remains a test-ordering hypothesis and is not evidence of product CDC data loss in the latest-stop path. Remaining asks before a test-only PR can be accepted: 1. Reorder the helper so the bounded job is awaited to `FINISHED` first, then the final sink values are asserted; keep the existing time bounds rather than enlarging them. 2. Run all five latest-stop entry points (not just `testMysqlCdcInitialStartupWithLatestStop`) with the reordered helper and attach the resulting Surefire/Failsafe reports alongside the current ones from artifact 12467842508 for comparison. 3. Show that the reordered assertion still fails if rows written after the snapshot phase are missing from the sink, so the change does not mask a real post-snapshot loss — a deliberately broken run or an explicit negative check is fine. 4. Since the October 22 execution is the only observed failure so far, please note in the PR whether the reordered test passes repeatedly (a few consecutive runs) rather than once. Once those are in place, open the PR against `dev` and link it here; a test-only change in this scope is welcome. <!-- streview-comment:2050 --> -- 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]
