You are viewing a plain text version of this content. The canonical link for it is here.

Posted to dev@kafka.apache.org by "Jay Kreps (JIRA)" <ji...@apache.org> on 2015/02/23 20:38:12 UTC

[jira] [Commented] (KAFKA-1979) LogCleaner delete file asynchronously and can cause inconsistency between log cleaner checkpoint and first dirty log segment file.

    [ https://issues.apache.org/jira/browse/KAFKA-1979?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14333703#comment-14333703 ] 

Jay Kreps commented on KAFKA-1979:
----------------------------------

Can you expand on this a bit. The delete protocol is to first rename the segment to .deleted so if we crash and recovery we should just delete it then. But I'm not sure if that covers the issue you see.

> LogCleaner delete file asynchronously and can cause inconsistency between log cleaner checkpoint and first dirty log segment file.
> ----------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: KAFKA-1979
>                 URL: https://issues.apache.org/jira/browse/KAFKA-1979
>             Project: Kafka
>          Issue Type: Bug
>            Reporter: Jiangjie Qin
>
> In LogCleaner we delete the log segment file asynchronously and then checkpoint the cleaned offsets. If broker exit after checkpoint is written but before log segment got deleted, when the broker start again, it will have the first dirty log offset different from the check-pointed offset.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)