You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@kafka.apache.org by "David Arthur (Jira)" <ji...@apache.org> on 2023/11/08 17:20:00 UTC

[jira] [Created] (KAFKA-15799) ZK brokers incorrectly handle KRaft metadata snapshots

David Arthur created KAFKA-15799:
------------------------------------

             Summary: ZK brokers incorrectly handle KRaft metadata snapshots
                 Key: KAFKA-15799
                 URL: https://issues.apache.org/jira/browse/KAFKA-15799
             Project: Kafka
          Issue Type: Bug
            Reporter: David Arthur
            Assignee: David Arthur
             Fix For: 3.6.1


While working on the fix for KAFKA-15605, I noticed that ZK brokers are unconditionally merging data from UpdateMetadataRequest with their existing MetadataCache. This is not the correct behavior when handling a metadata snapshot from the KRaft controller. 

For example, if a topic was deleted in KRaft and not transmitted as part of a delta update (e.g., during a failover) then the ZK brokers will never remove the topic from their cache (until they restart and rebuild their cache).



--
This message was sent by Atlassian Jira
(v8.20.10#820010)