SEZ9 commented on PR #12578: URL: https://github.com/apache/seatunnel/pull/12578#issuecomment-6393754839
Thanks @hutiefang76 for posting the final status of run 39775915510 on ef53a70c6f72f5f6ae7c31fcc57b541eb3c849f3. Good to see the retried JDK 11 unit-test job (120703016460) pass. On the remaining engine-v2-it/JDK 11 failure (119919175848): as noted above, the checkpoint timeout in `BackpressureSlowSinkIT` is unrelated to the `SeaTunnelRow` byte-size cache this PR changes, so there is no need to investigate it here or add any source/test changes to this PR for it. Please just re-run that failed job once; if it times out the same way again, link the run and leave it there. The fork `workflow_dispatch` run does not attach as the PR-bound `Build` check, so a maintainer still needs to retry the failed `Notify test workflow` (38787676286 / 116603788523) to produce the required check. Separately from CI, none of the earlier review points are blockers, but it would help to know where each stands (or to see the follow-up pushed): 1. **PR12578-F1 (Performance, MEDIUM)** – the unconditional volatile write in `setField` adds a memory barrier to every field assignment on the row-construction hot path. Either make the invalidation conditional or explain why the cost is acceptable. 2. **PR12578-F2 (Docs, LOW)** – add Javadoc on `setField` describing the cache-invalidation contract and its limitation, and align the placement of the architecture-guide paragraph with its scope. 3. **PR12578-F3 / F5 (Test, LOW)** – add a case that mutates a nested row / via `getFields()` and asserts the cached size is *not* invalidated, pinning the documented limitation. 4. **PR12578-F4 / F5 (Test, LOW)** – add the cross-overload case on a single row (typed pre-warm followed by untyped reads, and vice versa) to show both overloads share the same cache and observe the invalidation. My source-level conclusion is unchanged: no source-side blocker. <!-- streview-comment:2029 --> -- 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]
