You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "A. Sophie Blee-Goldman (Jira)" <ji...@apache.org> on 2021/06/26 00:36:00 UTC

[jira] [Commented] (KAFKA-9177) Pause completed partitions on restore consumer

    [ https://issues.apache.org/jira/browse/KAFKA-9177?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17369771#comment-17369771 ] 

A. Sophie Blee-Goldman commented on KAFKA-9177:
-----------------------------------------------

[~apolyakov] this ticket was specifically about pausing those completed changelogs in the restore consumer's assignment so that it doesn't continue to fetch records beyond the "end" of the changelog. The log message you're seeing doesn't indicate that this wasn't done, it's just a debug message to confirm we've finished restoring all changelogs when we run through the no-op restore phase after restoration is complete.

I wouldn't worry about it.

 

> Pause completed partitions on restore consumer
> ----------------------------------------------
>
>                 Key: KAFKA-9177
>                 URL: https://issues.apache.org/jira/browse/KAFKA-9177
>             Project: Kafka
>          Issue Type: Improvement
>          Components: streams
>            Reporter: A. Sophie Blee-Goldman
>            Assignee: Guozhang Wang
>            Priority: Major
>             Fix For: 2.6.0
>
>
> The StoreChangelogReader is responsible for tracking and restoring active tasks, but once a store has finished restoring it will continue polling for records on that partition.
> Ordinarily this doesn't make a difference as a store is not completely restored until its entire changelog has been read, so there are no more records for poll to return anyway. But if the restoring state is actually an optimized source KTable, the changelog is just the source topic and poll will keep returning records for that partition until all stores have been restored.
> Note that this isn't a correctness issue since it's just the restore consumer, but it is wasteful to be polling for records and throwing them away. We should pause completed partitions in StoreChangelogReader so we don't slow down the restore consumer in reading from the unfinished changelog topics, and avoid wasted network.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)