SEZ9 commented on PR #11780: URL: https://github.com/apache/seatunnel/pull/11780#issuecomment-5366241662
Thanks for the thorough follow-up, @DanielLeens — both the formal review and the root-cause writeup are much appreciated, and I agree with your conclusion: this isn't flakiness. The `all-connectors-it-1` (`PostgresCDCIT#testAddFieldWithRestore`) and `all-connectors-it-6` (`OpengaussCDCIT`) failures are a deterministic regression from the unconditional restore hook. Your analysis matches what I see in the code: `IncrementalSourceReader` hands `checkpointTables` to the collector via `restoreSchema(...)` unconditionally, while `SeaTunnelRowDebeziumDeserializeSchema#restoreCheckpointProducedType(...)` returns early when `schemaChangeResolver == null`. That split-brain (deserializer on the new schema, collector pulled back to the old one) explains the NPE / row-arity failures on Postgres, OpenGauss, and MongoDB. The proposed fix on `DanielLeens:codex/fix-pr11780-resolver-gate-20260820` (commit `01fbe370236aadd358ec7ee99f5f9df610479e3d`) looks right to me — gating the collector-side restore on `debeziumDeserializationSchema.getSchemaChangeResolver() != null` keeps both sides behind the same condition, which is the minimal correct change here. Also confirmed on my side: the re-diff between `75ea8e5` and `cfe2ab9` showing `cfe2ab9` is a pure conflict-resolution merge is consistent with the intent — no logic changes were meant to land in that merge. Concrete remaining asks to move this forward: 1. Cherry-pick `01fbe370236aadd358ec7ee99f5f9df610479e3d` (or an equivalent change) onto this PR branch so the resolver gate lives in the PR itself. 2. Rebase/update the branch onto the latest `dev` and push to refresh the head, per the earlier request — the temporary follow-up PR has been closed, so all remaining work should continue here. 3. Once the new head is up, let's confirm `all-connectors-it-1` and `all-connectors-it-6` go green before we look at any remaining gates. Thanks again for keeping this moving — this is on the 3.0.0 path, so a quick turnaround on the rebase + cherry-pick would be great. <!-- streview-comment:418 --> -- 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]
