junrao commented on pull request #8657: URL: https://github.com/apache/kafka/pull/8657#issuecomment-669330984
@chia7712 : The following is my thought after thinking about this a bit more. The changes that we made in DelayedJoin is complicated and it still doesn't completely solve the deadlock issue. Adding more complexity there to solve the issue is probably not ideal. As a compromise, I was thinking another approach. We could pass down a flag to ReplicaManager.appendRecords() to complete the delayed requests in a separate thread there. Only GroupCoordinator will set this flag. So, the background completeness check is limited to the offset_commit topic. Since this topic is low volume, a single thread is likely enough to handle the load. So, we don't have to make the number of thread configurable and don't have to worry about this thread being overwhelmed. The benefit of this approach is that the code is probably easier to understand since we could keep most existing logic in GroupCoordinator unchanged. What do you think? ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
