ym0506 commented on issue #39765: URL: https://github.com/apache/shardingsphere/issues/39765#issuecomment-5632984016
Thanks for offering to investigate, @PrakarshSrivastav! I can help check a proposed fix against the reproductions. The best starting point for an upstream change is the [test-only current-master reproducer](https://github.com/ym0506/shardingsphere/commit/92571c88362bf71b2b3e49d1ec94834fd67854a7) linked in the issue. It uses a deterministic overlap rather than timing or sleeps. The [standalone 5.5.3 reproducer](https://gist.github.com/ym0506/451d135eac9f2b793e1753721c3f7701) is another way to inspect the failure without building the reactor; it is separate release-specific evidence. For a fix, the useful checks are: - Both overlapping invocations export distinct completed spans, with each invocation's own SQL attributes, status, and exception event. - The exception path still works when `afterMethod` runs in `finally` after `onThrowing`; it must not finish an outer nested invocation. - Nested calls and a failure during `beforeMethod` do not leave or consume another invocation's state. The issue describes an experimental invocation-keyed approach, but its ownership and cleanup trade-offs remain open, including completion hooks skipped by dynamic plugin disabling. Please don't treat that prototype as a required implementation or a merge-ready fix. A smaller design would be welcome; the maintainers can guide the preferred state owner. No RouteContract installation is needed for these advice-level regressions. If you share a draft, I can compare its behavior with the existing repro cases. Codex assisted with this investigation and reply. -- 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]
