[
https://issues.apache.org/jira/browse/CASSANDRA-9151?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14522067#comment-14522067
]
sankalp kohli commented on CASSANDRA-9151:
------------------------------------------
I am +1 on the patch which I applied to 2.1. While looking at the 2.1 branch, I
found one more potential issue not related to the patch.
We could potentially block this stage again since many methods are synchronized
in ActiveRepairService.
Methods prepareForRepair(could block for 1 hour for prepare message response)
and finishParentSession(this one block for anticompaction to finish) are
synchronized and could hold on to the lock for a long time.
In RepairMessageVerbHandler.doVerb, if there is an exception for another
repair, removeParentRepairSession(also synchronized) will block.
I am not sure whether we can take this in another JIRA as this is blocking lot
of people.
> Anti-compaction is blocking ANTI_ENTROPY stage
> -----------------------------------------------
>
> Key: CASSANDRA-9151
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9151
> Project: Cassandra
> Issue Type: Bug
> Reporter: sankalp kohli
> Assignee: Yuki Morishita
> Fix For: 3.x, 2.1.x
>
> Attachments:
> 0001-Do-not-block-stage-when-performing-anticompaction.patch
>
>
> In RepairMessageVerbHandler.doVerb case ANTICOMPACTION_REQUEST
> We block on ANTI_ENTROPY stage. This is not good as no other message could be
> processed till any anticompaction is running. This can cause snapshots for
> new repairs to fail.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
