Luke Chen created KAFKA-19933:
---------------------------------
Summary: Make new voter caught up criteria strict
Key: KAFKA-19933
URL: https://issues.apache.org/jira/browse/KAFKA-19933
Project: Kafka
Issue Type: Improvement
Reporter: Luke Chen
In
[KIP-853|https://cwiki.apache.org/confluence/display/KAFKA/KIP-853%3A+KRaft+Controller+Membership+Changes#KIP853:KRaftControllerMembershipChanges-Handling.3],
when adding voter, we have to wait for the replica to catch up with the
leader. Currently, we consider a replica caught up for add voter, if they have
fetched within the last hour
([here|https://github.com/apache/kafka/blob/trunk/raft/src/main/java/org/apache/kafka/raft/internals/AddVoterHandler.java#L283]).
This is a loose check and in worst case, the new added voter might have 1 hour
of data lagged.
We can consider to compare the leader high watermark and the endOffset or
lastFetchLeaderLogEndOffset in the follower state, to make sure the follower's
log end offset has already exceeded the leader high watermark.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
