[
https://issues.apache.org/jira/browse/CASSANDRA-20180?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17936649#comment-17936649
]
Andy Tolbert commented on CASSANDRA-20180:
------------------------------------------
Attached with test results [^CASSANDRA-20180-5_ci_summary.html] , almost
entirely clean except some ongoing flakes that aren't related to the change.
> AutoRepair scheduler should attempt to avoid selecting nodes when their
> replicas are already scheduling repairs
> ---------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-20180
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20180
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Consistency/Repair
> Reporter: Andy Tolbert
> Assignee: Andy Tolbert
> Priority: Normal
> Attachments: CASSANDRA-20180-5_ci_summary.html
>
>
> Currently AutoRepair scheduling does not factor in whether replicas of a node
> are running their auto repair schedule.
> Ideally, the scheduler would attempt to have repairs issued in a replica
> disjoint way if it is possible.
> e.g. assume you have nodes 1, 2, 3, 4, ..., 12 and replication is in
> sequential order with RF 3
> e.g. 5 shares ranges with 3, 4, 6, 7
> If 5 has an ongoing repair, we could avoid having nodes 3, 4, 6, 7 as being
> chosen in favor of other nodes.
> Anecdotally, I had a 12 node cluster with {{parallel_repair_count}} of 3, and
> I observed that at a given time, 3 replicas of one another were scheduling
> incremental repairs, which could lead to conflicts if ranges overlap sstables.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
