[
https://issues.apache.org/jira/browse/KAFKA-20254?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18064387#comment-18064387
]
ASF GitHub Bot commented on KAFKA-20254:
----------------------------------------
lucasbru opened a new pull request, #826:
URL: https://github.com/apache/kafka-site/pull/826
The original warning about KAFKA-20254 said users should "upgrade their
brokers to a later release", which implies such a release already exists. This
updates the wording to say users should "wait for a future release that
includes the fix" and clarifies that the warning applies specifically to
migrating a classic group to the new protocol.
> Streams group replay fails after log compaction removes classic group
> tombstone
> -------------------------------------------------------------------------------
>
> Key: KAFKA-20254
> URL: https://issues.apache.org/jira/browse/KAFKA-20254
> Project: Kafka
> Issue Type: Task
> Components: group-coordinator, streams
> Affects Versions: 4.2.0
> Reporter: Lucas Brutschy
> Assignee: Lucas Brutschy
> Priority: Blocker
> Fix For: 4.3.0, 4.2.1
>
>
> After offline migration from classic group to streams group, the group
> coordinator fails with "Group X is not a streams group" after a coordinator
> restart or partition reassignment.
> The offline migration (classic → streams) writes a GroupMetadataKey tombstone
> to delete the classic group. However, the OffsetCommitKey records for that
> group remain in __consumer_offsets. After log compaction removes the
> tombstone (past delete.retention.ms), the following happens during
> coordinator replay:
> 1. OffsetMetadataManager.replay encounters offset commit records without an
> existing group
> 2. It calls getOrMaybeCreateClassicGroup(groupId, true), which creates a
> simple classic group (empty, no protocol type)
> 3. When streams group records are replayed next,
> getOrMaybeCreatePersistedStreamsGroup finds a classic group already exists
> and throws "Group X is not a streams group"
>
> Trigger conditions:
> - Classic group converted to streams group via offline migration
> - Enough time passes for log compaction to clean up the GroupMetadataKey
> tombstone
> - Coordinator restarts or partition is reassigned, triggering a full replay
>
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
