Luke Chen created KAFKA-20746:
---------------------------------
Summary: NPE in MetadataCache#toCluster
Key: KAFKA-20746
URL: https://issues.apache.org/jira/browse/KAFKA-20746
Project: Kafka
Issue Type: Bug
Affects Versions: 4.3.1, 4.2.0
Reporter: Luke Chen
In KAFKA-18225, we fix a bug to allow
`ClientQuotaCallback#updateClusterMetadata` to be supported in KRaft mode. But
the cluster object generated in MetadataCache could cause NPE. See
[https://github.com/strimzi/kafka-quotas-plugin/issues/71] for more info.
I think the root cause is that in toCluster method, we filter out the fenced
brokers to build brokerToNodes map
[here|https://github.com/apache/kafka/blob/4.3.1/metadata/src/main/java/org/apache/kafka/metadata/MetadataCache.java#L151-L154].
But later, we want to get all the replicas info, which include the fenced
replicas, via the filtered brokerToNodes map. So it causes the NPE. That's my
assumption because from the original reported bug, we don't have other info.
But it's a possible way to cause NPE. We should also think if there is other
possibility to cause NPE here.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
