[jira] [Resolved] (KAFKA-18200) New coordinator can flush an empty batch if append fails

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

     [ https://issues.apache.org/jira/browse/KAFKA-18200?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

David Jacot resolved KAFKA-18200.
---------------------------------
    Resolution: Fixed

> New coordinator can flush an empty batch if append fails
> --------------------------------------------------------
>
>                 Key: KAFKA-18200
>                 URL: https://issues.apache.org/jira/browse/KAFKA-18200
>             Project: Kafka
>          Issue Type: Bug
>          Components: group-coordinator
>            Reporter: Sean Quah
>            Assignee: David Jacot
>            Priority: Blocker
>             Fix For: 4.0.0
>
>
> When serialization of records fails in {{{}CoordinatorRuntime.append{}}}, the exception is not caught and we can leave an empty batch around. After the linger time elapses, we try to flush it and {{PartitionWriter.append()}} returns 0 instead of the usual offset. {{SnapshottableCoordinator}} throws an exception when we try to update the last written offset.
> {code:java}
> java.lang.IllegalStateException: New last written offset 0 of __consumer_offsets-1 must be greater than 5.
> 	at org.apache.kafka.coordinator.common.runtime.SnapshottableCoordinator.updateLastWrittenOffset(SnapshottableCoordinator.java:143) {code}
>  
> Luckily this error is handled and the empty batch is failed and removed, so the coordinator does not get stuck.



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