DanielLeens commented on PR #11557: URL: https://github.com/apache/seatunnel/pull/11557#issuecomment-6139152667
Review/CI reconciliation on current head `5dc655e6a91`: - Pushed [`5dc655e6a91`](https://github.com/apache/seatunnel/commit/5dc655e6a912c3a6968c97a19316739dc7437663), which fixes three managed-runtime lifecycle defects identified in review: keep draining coordinator control events in `PREPARE_CLOSE`, avoid claiming coordinator ownership on the restore-operation thread, and retain Kafka pending splits until their reader has completed the managed epoch handshake. It also adds a restore-thread regression test and removes stale allowlist wording from the managed-runtime E2E. - F1 executed evidence: on the previous head `b00c3e0cdad`, all four unit-test matrix jobs passed. In the [JDK 8 Ubuntu job](https://github.com/DanielLeens/seatunnel/actions/runs/37762848399/job/113270515885), `ManagedSourceRegisterOperationSerializationTest` executed 2 tests with 0 failures/errors/skips. This covers both the mismatched same-factory/different-class payload rejection and the valid round trip. - Filed the explicitly requested legacy follow-up as [#12747](https://github.com/apache/seatunnel/issues/12747), covering the unchecked `AssignSplitOperation#readInternal` path and the typed-override behavior in `TaskOperation#readInternal`. - The old docs thread is now resolved: `connector-allowlist` has been removed since `8dfbf773206`; lane selection is the default-off global flag plus connector-declared `ManagedSourceCapability`, and unsupported connectors fail closed. The benchmark request remains open. The existing `seatunnel-benchmarks` pipeline cannot measure this lane as-is: `BenchmarkSource` currently uses the legacy default capability, and every benchmark engine template omits `managed-source-runtime.enabled`, so reusing the current JMH case would silently benchmark the legacy path. A meaningful comparison needs a paired legacy/managed benchmark mode, explicit source capability opt-in, engine-template wiring, and an assertion that the managed lane was actually selected. I did not fold that broader performance fixture into this correctness repair. The new [Build run](https://github.com/DanielLeens/seatunnel/actions/runs/38352742729) is queued. No POM files were changed. -- 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]
