SEZ9 commented on issue #12441: URL: https://github.com/apache/seatunnel/issues/12441#issuecomment-5825951039
Thanks @Vivek1106-04, recording this here works for me. Agreed: the regression is how this issue closes, and both parts must be shown red on the current blocking lock acquisition before the fix turns them green. On part 1 (same coordinator): this is the right shape. Hold `pendingCounter > 0`, start `startSavepoint()`, then call `tryTriggerPendingCheckpoint()` on the same coordinator. Red run: it blocks for the full drain on current `dev`. Green run: it returns within a bound far below the drain time, creates no pending checkpoint, and re-arms. That lines up with constraints 3 and 4 already listed above (no new non-savepoint pending checkpoint while the gate is set, re-arm rather than drop). Two asks so this stays deterministic rather than sleep-timed: - Gate the drain with a latch/future the test controls, so "blocks for the full drain" and "returns well under the drain time" are asserted against a test-owned signal, not wall-clock sleeps. - Cover the general, completed-point, and schema-change trigger paths in the same test, since all three must re-arm identically. On part 2 (cross-pipeline): agreed on the ordering dependency. It can only go red on a base that contains the shared dispatch pool, so either both parts land in this PR if the scheduler change merges first, or the fix lands with part 1 and part 2 follows immediately after. Thanks for offering to write part 2 yourself in that case; this issue stays open until part 2 has been shown red then green. For the PR description, please include for each part: the test name, the base revision it was run against, and the observed failure (blocking/starvation), followed by the passing run with the fix. Everything else above stands unchanged, including keeping the unexpected-trigger-failure policy out of this change. <!-- streview-comment:1303 --> -- 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]
