[ 
https://issues.apache.org/jira/browse/KAFKA-20115?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18056278#comment-18056278
 ] 

Sean Quah commented on KAFKA-20115:
-----------------------------------

I haven't checked the client code but I think they have to retry after the 
rebalance timeout and not the request timeout, since JOIN_GROUP requests can be 
blocked for longer than the request timeout. Once the timeout has elapsed, the 
client will recover. It will either re-send the request to the old broker and 
see NOT_COORDINATOR or issue a FIND_COORDINATOR directly and try the new 
broker, not sure which.

The main problem is the long timeout before clients recover.

> Group coordinator fails to unload metadata when no longer leader or follower
> ----------------------------------------------------------------------------
>
>                 Key: KAFKA-20115
>                 URL: https://issues.apache.org/jira/browse/KAFKA-20115
>             Project: Kafka
>          Issue Type: Bug
>          Components: group-coordinator
>    Affects Versions: 4.0.1, 4.1.0, 4.2.0, 4.1.1
>            Reporter: Sean Quah
>            Assignee: Kuan Po Tseng
>            Priority: Blocker
>
> KAFKA-19427 introduced logic in the coordinator runtime's 
> {{freeCurrentBatch}} to only recycle batch buffers when their size is less 
> than or equal to the partition's max message size.
>  
> KAFKA-19519 replaced the max message size with the 
> group.coordinator.cached.buffer.max.bytes and 
> share.coordinator.cached.buffer.max.bytes configs.
> {{PartitionWriter.config()}} throws a {{NOT_LEADER_OR_FOLLOWER}} exception 
> when the broker is no longer a leader or follower of the partition.
> When unloading partitions from the group coordinator, we try to free the 
> current batch. If the broker is no longer a leader or follower of the 
> {{__consumer_offsets}} partition, a {{NOT_LEADER_OR_FOLLOWER}} exception is 
> thrown and we do not complete the unload process. This leaves classic group 
> rebalance requests like JOIN_GROUP and SYNC_GROUP hanging.
> The bug affects versions after KAFKA-19427 and before KAFKA-19519. 



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

Reply via email to