You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Ajay Patel (Jira)" <ji...@apache.org> on 2021/03/29 20:52:00 UTC

[jira] [Commented] (KAFKA-10127) kafka cluster not recovering - Shrinking ISR continously

    [ https://issues.apache.org/jira/browse/KAFKA-10127?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17310982#comment-17310982 ] 

Ajay Patel commented on KAFKA-10127:
------------------------------------

You might find luck in [KAFKA-2729|https://issues.apache.org/jira/browse/KAFKA-2729] as that issue sounds identical and has some more traffic.

> kafka cluster not recovering - Shrinking ISR  continously
> ---------------------------------------------------------
>
>                 Key: KAFKA-10127
>                 URL: https://issues.apache.org/jira/browse/KAFKA-10127
>             Project: Kafka
>          Issue Type: Bug
>          Components: replication, zkclient
>    Affects Versions: 2.4.1
>         Environment: using kafka version 2.4.1 and zookeeper version 3.5.7
>            Reporter: Youssef BOUZAIENNE
>            Priority: Major
>
> We are actually facing issue from time to time where our kafka cluster goes into a weird state. We see the following log repeating
> [2020-06-06 08:35:48,117] INFO [Partition test broker=1002] Cached zkVersion 620 not equal to that in zookeeper, skip updating ISR (kafka.cluster.Partition)
>  [2020-06-06 08:35:48,117] INFO [Partition test broker=1002] Shrinking ISR from 1006,1002 to 1002. Leader: (highWatermark: 3222733572, endOffset: 3222741893). Out of sync replicas: (brokerId: 1006, endOffset: 3222733572). (kafka.cluster.Partition)
>  
> Just before that our zookeeper session expired which lead us to that state.
>  
> After we increased this two values below we encounter the issue less frequently but it still appears from time to time and the only solution is restart of kafka service on all brokers to recover.
> zookeeper.session.timeout.ms=18000
> replica.lag.time.max.ms=30000
>  
> Any thoughts on that please  



--
This message was sent by Atlassian Jira
(v8.3.4#803005)