kamalcph commented on PR #15634: URL: https://github.com/apache/kafka/pull/15634#issuecomment-2029474481
> just curious. Does it happens only if remote storage is enabled? According to the description: > > > The follower sends the first FETCH request to the leader, the leader checks whether the isFollowerInSync, then expands the ISR. Also, parks the request in DelayedFetchPurgatory. If the replica was elected as leader before the fetch-response gets processed, then the new-leader will have wrong high-watermark. > > It looks like the issue is existent even though we don't use remote storage. For normal topic, once the replica becomes leader. It is able to [resolve/convert](https://sourcegraph.com/github.com/apache/kafka@40e87ae35beb389d6419d32130174d7c68fa4d19/-/blob/core/src/main/scala/kafka/log/UnifiedLog.scala#L319) the highwatermark offset (log-start-offset) to metadata by reading the segment from disk and then it updates the high-watermark to either current-leader-log-end-offset (or) the lowest LEO of all the eligible-isr replicas. In case of remote topic, the replica will fail to [resolve](https://sourcegraph.com/github.com/apache/kafka@40e87ae35beb389d6419d32130174d7c68fa4d19/-/blob/core/src/main/scala/kafka/log/UnifiedLog.scala#L319) the highwatermark offset (log-start-offset) to metadata since the segment won't be in local-disk, and then fail continuously. -- 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