viktorsomogyi commented on PR #13975: URL: https://github.com/apache/kafka/pull/13975#issuecomment-1641908924
@mumrah correct, the problem happens because the broker doesn't have a populated metadata cache and it tries to auto-create a topic during a topic metadata request (and it might be that this topic otherwise already exists). The PR is trying to fix this problem too. I thought I'd put the solution to ZkAdminManager so that it's not too different from the KRaft solution as there the broker would forward a topic creation request to the controller and return `UNKNOWN_TOPIC_OR_PARTITION` to the client. It's important to note that this isn't the first metadata request that the consumer in the herder sends out but a subsequent one, so it had some connection to a valid broker but lost it and reconnected to another one which was just starting up and didn't have metadata. In the Jira description I used a single broker because it was easier to reproduce it that way but it can be reproduced in a multi-broker environment too of course. I didn't really consider how would the controller affect things here because as you say they start accepting requests before having metadata. -- 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
