vbhanuchander-lang commented on PR #309: URL: https://github.com/apache/flink-connector-kafka/pull/309#issuecomment-6386200759
@MartijnVisser fair question, and sorry for the slow reply. Yes, I used an AI tool while writing this, and I understand the change: #300 added `readersWithNoMoreSplits` so each reader is told "no more splits" only once. But on a metadata change, `onHandleSubscribedStreamsFetch` closes and recreates every sub-enumerator, and the reader recreates its sub-readers too. So the set still holds entries from before the change, which suppresses the new round of signals, and the bounded job never finishes. Clearing it right after `closeAllEnumeratorsAndContexts()` resets it at the point where the old sub-enumerators go away. Thanks for reproducing it. You're right that it's narrower than I described: bounded sources force the discovery interval to -1, so it only bites when that's bypassed via `setProperty` after `build()`. You're also right about the retained cluster; I was wrong that only a switchover shows it, and with FLINK-31006 now merged that comment was out of date anyway. Pushed: rebased onto main, description and commit message narrowed to the builder case, the comment cut to one line, and the wrong claim dropped from the test comment. The test still fails on main without the clear and passes with it, and the dynamic source enumerator/reader suites are green. -- 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]
