[
https://issues.apache.org/jira/browse/CASSANDRA-13176?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16078207#comment-16078207
]
Alex Petrov commented on CASSANDRA-13176:
-----------------------------------------
I can confirm that the issue exists (dropping index won't automatically stop
the index rebuild). This is quite useful in cases when, for example, an
operator over a large cluster accidentally started building an index on a wrong
column, so they drop the index right away. Of course the issue has the obvious
workaround: stop index build manually.
My assumption on why this was not done on the first place is that right now
{{CompactionHolder}} that would be used for stopping the index build can now be
responsible for more than a single index build.
The middle ground could be to stop only the tasks that are rebuilding a single
index, not ones that are related to loaded sstables etc, but in this case it
feels a bit counter intuitive.
> DROP INDEX seemingly doesn't stop existing Index build
> ------------------------------------------------------
>
> Key: CASSANDRA-13176
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13176
> Project: Cassandra
> Issue Type: Bug
> Components: Compaction, CQL
> Environment: CentOS Linux, JRE 1.8
> Reporter: Soumya Sanyal
>
> There appears to be an edge case with secondary indexes (non SASI). I
> originally issued a CREATE INDEX on a column, and upon listening to advice
> from folks in the #cassandra room, decided against it, and issued a DROP
> INDEX.
> I didn't check the cluster overnight, but this morning, I found out that our
> cluster CPU usage was pegged around 80%. Looking at compaction stats, I saw
> that the index build was still ongoing. We had to restart the entire cluster
> for the changes to take effect.
> Version: 3.9
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
