jjj-n commented on PR #12265: URL: https://github.com/apache/seatunnel/pull/12265#issuecomment-5746840891
Thanks for the second review. I checked the cleanup branch against the merged dev parent and the remaining CI failures. The early-fire cleanup rescheduling at `processPendingJobCleanup` was introduced upstream by #11503, commit ec1b1b8b51a7aa39c208cf39473f8f44104b3939 (September 13). It is already present in this merge's dev parent c7304ace6e18d350314e92480df1fd3c0962f1f2 and is unchanged by this PR relative to that parent. It appeared between reviewed heads because 4a5fbc94 merged dev. I agree dedicated coverage would be useful, but it is not a cleanup behavior change introduced by the executor split. Could you reassess that blocking item against the base-to-head diff? The BackpressureSlowSinkIT change has a local differential result: the original 90-second loop observed only two additional completed checkpoints both with this PR and with only CoordinatorService replaced by the unmodified dev version. Observed checkpoint duration reached 34 seconds despite a 15-second trigger interval, and the old loop did not sample after its final sleep. The revised loop preserves at least 90 seconds of stress, at least three additional checkpoints, zero failed checkpoints and the backpressure metric assertions, with a 180-second upper bound. The final three embedded-engine integration tests passed together (154.707 seconds). This reproduces the completion-count failure; it does not claim to reproduce the earlier CI first-checkpoint timeout. The completed CI run on 4a5fbc94 is not fully green: https://github.com/jjj-n/seatunnel/actions/runs/35302309118 . All four Java 8/11 Windows/Ubuntu unit-test jobs and both Java 8/11 engine-v2-it jobs passed. Remaining failed jobs: - all-connectors-it-2 / Java 8: Maven Central timed out downloading hadoop-common:3.3.4 before connector-cdc-mongodb-e2e dependency resolution completed (job 105468814056). - all-connectors-it-1 / Java 8 and 11: NebulaGraphIT.startUp:109 fails at adminPool.init after the graph client cannot ping the server, before connector job execution (jobs 105468814101 and 105468814191). - rocketmq-connector-it / Java 11: 87 tests, 14 failures and 29 errors; logs show missing topic routes, including failures in producer data generation and across Flink/Spark/Zeta executions (job 105468814771). This points to shared broker/topic setup rather than establishing an executor regression, but I am not treating it as a confirmed harmless flake or claiming the full matrix passes. I am rerunning the failed/cancelled checks on the same head to distinguish transient failures from repeatable setup problems. The PR remains draft while those results are pending. -- 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]
