[
https://issues.apache.org/jira/browse/CASSANDRA-9271?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14549110#comment-14549110
]
Benedict commented on CASSANDRA-9271:
-------------------------------------
Actually [~tjake] is the author; I just committed it along with the fix(es).
From my POV we should *both* 1) move it to a long test; 2) make it more robust
I'm pretty sure performing an actual major compaction is unnecessary, and this
is largely where the problem comes from (with the indeterminate sleeping as we
wait for cessation of other active tasks). What we need is for the instances of
reader to change as we try marking the files for redistribution, and we should
probably just do this in a direct tight loop so that both code sections are
spinning in direct contention. This both increases the likelihood of the
failure scenario, but also ensures that if we don't encounter it we can stop
promptly (by just spinning for a fixed interval, and calling it a day if we
haven't failed).
> IndexSummaryManagerTest.testCompactionRace times out periodically
> -----------------------------------------------------------------
>
> Key: CASSANDRA-9271
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9271
> Project: Cassandra
> Issue Type: Test
> Reporter: Ariel Weisberg
> Assignee: Joshua McKenzie
>
> The issue is that the amount of time the test takes is highly variable to it
> being biased towards creating a condition where the test has to retry the
> compaction it is attempting.
> Solution is to decrease the bias by having
> https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/db/ColumnFamilyStore.java#L2522
> check every millisecond instead of every 100 milliseconds.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
