Luke Chen created KAFKA-19331:
---------------------------------
Summary: No error handling for leader unregistered in
applyLocalFollowersDelta
Key: KAFKA-19331
URL: https://issues.apache.org/jira/browse/KAFKA-19331
Project: Kafka
Issue Type: Bug
Reporter: Luke Chen
In ReplicaManager#applyLocalFollowersDelta, when we prepare for fetching from
the leader, we'll check if the leader node info is in metadata image. If
somehow it didn't include in the newImage, we'll log something like:
{code:java}
[2025-05-26 15:25:58,124] TRACE [Broker id=4] Unable to start fetching
quickstart-events-0 with topic ID Some(Dn9K0BB8QWuj4PqcJD0nrA) from leader
Some(2) because it is not alive. (state.change.logger)
[2025-05-26 15:25:58,124] INFO [Broker id=4] Started fetchers as part of
become-follower for 1 partitions (state.change.logger)
{code}
It's confusing to users to see it's unable to fetch, then start fetch. And in
the end, it's not actually fetching... We should handling the error well by
updating the `FailedPartition` and not updating other successful result status.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
