kerwin-zk commented on PR #13283: URL: https://github.com/apache/gluten/pull/13283#issuecomment-6183070298
The constructor lookup itself fixes the reported `NoSuchMethodException`, but the created `FailedShuffleCleaner` is not retained by `CelebornShuffleManager`. In `fetchCleanFailedShuffle`, assigning the new instance to the `failedShuffleCleaner` parameter only updates that local reference; Java passes object references by value, so `CelebornShuffleManager.failedShuffleCleaner` remains `null`. As a result, `CelebornShuffleManager.stop()` skips `stopFailedShuffleCleaner`, even though `FailedShuffleCleaner` starts a scheduled executor during construction. This leaves the cleaner thread and its `LifecycleManager` reference alive after the shuffle manager is stopped. Celeborn 0.6.3 avoids this by storing the cleaner in the manager field and stopping it during shutdown. Could we return the created cleaner through `fetchCleanFailedShuffle` / `stageRerun` (or otherwise assign it to the manager field), and add a regression test that enables `celeborn.client.spark.fetch.cleanFailedShuffle` and verifies both construction and shutdown? -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
