rhauch commented on pull request #10112: URL: https://github.com/apache/kafka/pull/10112#issuecomment-914547745
@C0urante, thanks for the feedback on my suggestion. I like your option 3, because it does allow the iteration to stop on each source partition as soon as it encounters the first unacknowledged record in each queue. I also think that the behavior with the suggested approach and your option 3 is still a lot better than the current situation. One question, though: you mention that it might be a problem if iterating over the submitted records takes longer than `offset.flush.timeout.ms`. But IIUC the `offset.flush.timeout.ms` would actually not be used anymore, as there actually are no timeouts as the offset commit thread doesn't block anymore. So, worst case, if task A has a ton of submitted records that have to be iterated over (e.g., fast producer and fast source task), it might slow the committing of offsets for other tasks. (Again, this is not any worse than the current behavior.) But your option 2 would help with this at the risk of using more threads, and so we may also want to consider this to help ensure that no slowly-proceeding producer of a task blocks other offset commits. Of course, another option might be to incur the iteration on the worker source task thread. That would essentially move the use of the queue(s) to the worker source task thread, tho we still need to get the offsets to the offset commit thread and so would likely have to keep the synchronization blocks around the offset writer snapshot. On one hand, that's putting more work onto the worker source task thread and making the offset thread super straightforward (snapshot and write); on the other it's putting the onus on the worker source task thread. Thoughts? -- 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: jira-unsubscr...@kafka.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
