You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Yuki Morishita (JIRA)" <ji...@apache.org> on 2014/08/12 06:03:11 UTC
[jira] [Comment Edited] (CASSANDRA-7745) Background LCS compactions
stall with pending compactions remaining
[ https://issues.apache.org/jira/browse/CASSANDRA-7745?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14093700#comment-14093700 ]
Yuki Morishita edited comment on CASSANDRA-7745 at 8/12/14 4:01 AM:
--------------------------------------------------------------------
I think we have concurrency bug in CompactionManager that compactingCF count remains 1 when there is no active compaction.
{{submitBackground}} adds compacting CF to compactingCF and {{BackgroundCompactionTask}} removes the same CF after compaction.
Currently this can happen in either order, and when remove happens first, compacting count remains 1.
Attaching patch to fix this.
edit: Though it should continue to compact if further flushing happens and there is compaction thread available.
was (Author: yukim):
I think we have concurrency bug in CompactionManager that compactingCF count remains 1 when there is no active compaction.
{{submitBackground}} adds compacting CF to compactingCF and {{BackgroundCompactionTask}} removes the same CF after compaction.
Currently this can happen in either order, and when remove happens first, compacting count remains 1.
Attaching patch to fix this.
> Background LCS compactions stall with pending compactions remaining
> -------------------------------------------------------------------
>
> Key: CASSANDRA-7745
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7745
> Project: Cassandra
> Issue Type: Bug
> Reporter: J.B. Langston
> Assignee: Yuki Morishita
> Fix For: 1.2.19, 2.0.11, 2.1 rc6
>
> Attachments: 7745-1.2.txt
>
>
> We've hit a scenario where background LCS compactions will stall. compactionstats output shows hundreds of pending compactions but none active. The thread dumps show no CompactionExecutor threads running, and no compaction activity is being logged to system.log. This seems to happen when there are no writes to the node. There are no flushes logged either, and when writes resume, compactions seem to resume as well, but still don't ever get to 0.
--
This message was sent by Atlassian JIRA
(v6.2#6252)