[ 
https://issues.apache.org/jira/browse/KAFKA-21467?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

David Jacot resolved KAFKA-21467.
---------------------------------
    Fix Version/s: 4.5.0
                   4.4.1
                   4.3.3
       Resolution: Fixed

> Classic to consumer group conversion fails when no protocol is selected or a 
> member does not support it
> -------------------------------------------------------------------------------------------------------
>
>                 Key: KAFKA-21467
>                 URL: https://issues.apache.org/jira/browse/KAFKA-21467
>             Project: Kafka
>          Issue Type: Bug
>    Affects Versions: 4.0.0
>            Reporter: David Jacot
>            Assignee: David Jacot
>            Priority: Major
>             Fix For: 4.5.0, 4.4.1, 4.3.3
>
>
> When the first member using the consumer rebalance protocol joins a classic 
> group, the group coordinator converts the group to a consumer group in 
> {{ConsumerGroup.fromClassicGroup}}. The conversion parses the subscription of 
> every classic member with the selected protocol of the classic group. This 
> fails in two states that the classic group can be in while it rebalances:
> * No protocol is selected yet, for instance because the first rebalance of 
> the group has not completed. {{classicGroup.protocolName().get()}} throws 
> {{NoSuchElementException}}.
> * A member that does not support the selected protocol has joined the 
> rebalance. The next generation would select another protocol, but the member 
> has no metadata for the current one, so {{ClassicGroupMember.metadata}} 
> throws {{IllegalArgumentException}}.
> In both cases, the ConsumerGroupHeartbeat request that triggers the 
> conversion fails with UNKNOWN_SERVER_ERROR. The consumer treats this error as 
> fatal, so the upgraded consumer fails instead of joining the group. Every 
> upgraded consumer that joins fails the same way until the classic rebalance 
> completes.
> Steps to reproduce the first state, with 
> group.consumer.migration.policy=upgrade:
> # A classic consumer joins the group and waits for the initial rebalance to 
> complete (group.initial.rebalance.delay.ms).
> # Before it completes, a consumer with group.protocol=consumer joins the same 
> group.
> # The heartbeat of the second consumer fails with UNKNOWN_SERVER_ERROR.
> Steps to reproduce the second state:
> # A classic consumer that supports the range and cooperative-sticky assignors 
> joins the group, which becomes stable with range.
> # A classic consumer that only supports cooperative-sticky joins the group, 
> which starts a rebalance.
> # Before the rebalance completes, a consumer with group.protocol=consumer 
> joins the same group. Its heartbeat fails with UNKNOWN_SERVER_ERROR.
> The fix converts the members with the selected protocol when every member 
> supports it, and otherwise with the protocol that the next generation would 
> select.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to