SEZ9 commented on issue #12382: URL: https://github.com/apache/seatunnel/issues/12382#issuecomment-5987157120
Thanks @xinnyuli, this is exactly the comparison that was asked for, and it settles the open question. In all three missing-row runs on 146a1b5c the stored commit-end LSN, the first LSN after restore, the LSN filtered as "already processed", and the start of the `id=15` WAL range are the same value (e.g. `0/2228138 -> 0/2228270` in n18), while the two runs on 3080371c that hit the same boundary resumed there, filtered nothing, and delivered the row. That confirms the ambiguous-boundary hypothesis for the reported symptom; the earlier evidence only made it plausible. I agree with your caveat on the rate: 3/30 vs 0/28 is not significant on its own, and we should not quote it as proof. The boundary rows are the evidence. Your note on "after" n10 is also useful: the filtered message at the stored LSN with nothing else on the stream until `0/2227898`, and `id=15` starting later and being delivered, is consistent with the fix only skipping what was genuinely processed. Current state: this is covered by the active #12454 and is not fixed in `dev` yet, so please do not open a second production-fix PR. Remaining asks: 1. Post this table (and the two run links, 37185551736 and 37187621867) as a comment on #12454 so the reviewers there see the real-run confirmation alongside the unit and IT coverage. 2. Keep the `pg_current_wal_insert_lsn()`/`pg_current_wal_lsn()` logging as test-only; if you think it would strengthen the deterministic boundary assertion, propose it on #12454 rather than as a separate change. 3. If you have the capacity, run the PR's own extended IT on 3080371c in the same stress-ng/no-agent setup and report whether the boundary is hit and `id=15` is delivered, since you noted that was not run here. I'll keep this issue open and linked to #12454 until that PR is merged. <!-- streview-comment:1523 --> -- 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]
