[jira] [Commented] (KAFKA-14697) KRaft controller should not send LISR to KRaft brokers during migration

["David Arthur (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/KAFKA-14697?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17686590#comment-17686590 ] 

David Arthur commented on KAFKA-14697:
--------------------------------------

cc [~showuon] [~mimaison]

> KRaft controller should not send LISR to KRaft brokers during migration
> -----------------------------------------------------------------------
>
>                 Key: KAFKA-14697
>                 URL: https://issues.apache.org/jira/browse/KAFKA-14697
>             Project: Kafka
>          Issue Type: Bug
>          Components: kraft
>    Affects Versions: 3.4.0
>            Reporter: David Arthur
>            Priority: Major
>             Fix For: 3.5.0, 3.4.1
>
>         Attachments: test_online_migration.tar.gz
>
>
> During a migration, after a broker is restarted in KRaft mode the KRaft controller is still sending it LISR as if it were a ZK broker. This results in errors like:
> {code}
> [2023-02-09 14:53:26,892] ERROR Closing socket for 172.19.0.4:9092-172.19.0.9:32876-0 because of error (kafka.network.Processor)
> org.apache.kafka.common.errors.InvalidRequestException: Received request api key LEADER_AND_ISR which is not enabled
> {code}
> This can be observed by running the kafkatest.tests.core.zookeeper_migration_test.TestMigration.test_online_migration ducktape test. Attached are logs from a local run of this test with --debug set.
> It seems these errors do not affect the consistency of metadata during the migration, but the excessive LISR sent out will impact performance on large clusters. 



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