[KAFKA-14833] ZK migration cannot start if a ZK controller was not previously elected

• Type: Sub-task
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 3.4.0
• Component/s: kraft
• Assignee: HaiyuanZhao
• Reporter: David Arthur
• Created: 22/Mar/23 14:51
• Updated: 09/Aug/24 22:26
• Resolved: 09/Aug/24 22:26

Description

During the BecomeZkController event in KRaftMigrationDriver, an error occurs if there was not data in the /controller_epoch znode.

[2023-03-22 14:33:06,082] ERROR Had an exception in BecomeZkControllerEvent (org.apache.kafka.metadata.migration.KRaftMigrationDriver)
java.lang.IllegalStateException: Cannot register KRaft controller 3001 as the active controller since there is no ZK controller epoch present.
        at kafka.zk.KafkaZkClient.tryRegisterKRaftControllerAsActiveController(KafkaZkClient.scala:201)
        at kafka.zk.ZkMigrationClient.claimControllerLeadership(ZkMigrationClient.scala:60)
        at org.apache.kafka.metadata.migration.KRaftMigrationDriver.apply(KRaftMigrationDriver.java:143)
        at org.apache.kafka.metadata.migration.KRaftMigrationDriver.access$800(KRaftMigrationDriver.java:57)
        at org.apache.kafka.metadata.migration.KRaftMigrationDriver$BecomeZkControllerEvent.run(KRaftMigrationDriver.java:357)
        at org.apache.kafka.queue.KafkaEventQueue$EventContext.run(KafkaEventQueue.java:121)
        at org.apache.kafka.queue.KafkaEventQueue$EventHandler.handleEvents(KafkaEventQueue.java:201)
        at org.apache.kafka.queue.KafkaEventQueue$EventHandler.run(KafkaEventQueue.java:175)
        at java.lang.Thread.run(Thread.java:750)

This mainly affects tests when we start a KRaft quorum first, and then the (new) ZK cluster. The migration driver will kick off when the ZK brokers register but it will race with KafkaController on the /controller_epoch creation.