goutamadwant commented on PR #12313: URL: https://github.com/apache/seatunnel/pull/12313#issuecomment-5841704234
I ran `BackpressureSlowSinkIT` in a loop locally against `dev` (`deb16a3c3`), this PR, #12316, and both together. Setup: JDK 8, and JDK 11 with `-XX:ActiveProcessorCount=2`, 7-8 runs per variant per JDK, on a loaded 10-core macOS host. Nothing failed: base 0 / 16, this PR 0 / 16, #12316 0 / 16, both 0 / 15. I could not reproduce the CI starvation locally, so this only shows that the fixture change does not break the test. It does not show that it fixes the flake. On the approach: - The assertions are unchanged, and backpressure still holds with 500 splits of 4000 rows. The 1 s inter-split sleep drains about 500 of the 2048 queue slots. I agree with the point above that the release window recurs roughly every 9 s, not every second. - The trade-off is coverage. With this fixture the reader leaves the checkpoint lock once per split, so the test no longer exercises the case where barrier injection loses the lock to back-to-back polls. That case is the engine behaviour #12316 changes. If both PRs land, one variant of the test on the original `split.num = 1` workload would keep #12316 covered. -- 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]
