dianfu commented on PR #29604: URL: https://github.com/apache/flink/pull/29604#issuecomment-6438873679
@bowenli86 You raised a good question. I’m inclined to close this PR based on the following reasons: - For whole-job cancellation and global failover, bundle closing is already asynchronous, and releasing the final shared lease tears down the Python worker. The benefit of this PR is therefore mainly avoiding unnecessary computation. During TM shutdown, SIGTERM triggers worker cleanup, while SIGKILL bypasses Java cleanup entirely. - The main regression risk without this PR is partial failover with a surviving shared-worker user. Skipping RemoteBundle.close() removes the cleanup opportunity for otherwise healthy bundles, potentially leaving registrations and pending output that can block surviving operators just as what you have pointed out. - Keeping the existing asynchronous cleanup seems safer until we can abort and clean up individual bundles without requiring computation to finish or terminating the shared worker. -- 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]
