You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Edoardo Comar (Jira)" <ji...@apache.org> on 2023/01/28 09:26:00 UTC

[jira] [Assigned] (KAFKA-14657) Admin.fenceProducers fails when Producer has ongoing transaction - but Producer gets fenced

     [ https://issues.apache.org/jira/browse/KAFKA-14657?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Edoardo Comar reassigned KAFKA-14657:
-------------------------------------

    Assignee: Edoardo Comar

> Admin.fenceProducers fails when Producer has ongoing transaction - but Producer gets fenced
> -------------------------------------------------------------------------------------------
>
>                 Key: KAFKA-14657
>                 URL: https://issues.apache.org/jira/browse/KAFKA-14657
>             Project: Kafka
>          Issue Type: Bug
>          Components: admin
>            Reporter: Edoardo Comar
>            Assignee: Edoardo Comar
>            Priority: Major
>         Attachments: FenceProducerDuringTx.java, FenceProducerOutsideTx.java
>
>
> Admin.fenceProducers() 
> fails with a ConcurrentTransactionsException if invoked when a Producer has a transaction ongoing.
> However, further attempts by that producer to produce fail with InvalidProducerEpochException and the producer is not re-usable, 
> cannot abort/commit as it is fenced.
> An InvalidProducerEpochException is also logged as error on the broker
> [2023-01-27 17:16:32,220] ERROR [ReplicaManager broker=1] Error processing append operation on partition topic-0 (kafka.server.ReplicaManager)
> org.apache.kafka.common.errors.InvalidProducerEpochException: Epoch of producer 1062 at offset 84 in topic-0 is 0, which is smaller than the last seen epoch
>  
> Conversely, if Admin.fenceProducers() 
> is invoked while there is no open transaction, the call succeeds and further attempts by that producer to produce fail with ProducerFenced.
> see attached snippets
> As the caller of Admin.fenceProducers() is likely unaware of the producers state, the call should succeed regardless



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