[
https://issues.apache.org/jira/browse/KAFKA-6902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16474825#comment-16474825
]
Ismael Juma commented on KAFKA-6902:
------------------------------------
The issue is that the StopReplica request will not succeed if the broker is
down. If the broker comes back and the data is still on disk, then it won't be
deleted. It's a bit tricky to fix this safely. Just relying on the LeaderAndIsr
request could cause data to be deleted in cases where it should not be.
> During replica reassignment if source broker is down the partition directory
> may not be removed from broker log.dirs
> --------------------------------------------------------------------------------------------------------------------
>
> Key: KAFKA-6902
> URL: https://issues.apache.org/jira/browse/KAFKA-6902
> Project: Kafka
> Issue Type: Bug
> Components: core
> Affects Versions: 0.11.0.2
> Reporter: Koelli Mungee
> Priority: Minor
>
> During replica reassignment if source broker is down the partition directory
> may not be removed from the source broker's log.dirs location. This can lead
> to the partition directory still being there with zero length log data. This
> in turn leads to this partition showing up in JMX metric for the original
> broker: kafka.log.LogEndOffset even though this partition is no longer a part
> of the ISR list in that broker.
> It would be good to have this cleaned up so the metric can be accurate at all
> times.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
