[
https://issues.apache.org/jira/browse/CASSANDRA-9288?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14539138#comment-14539138
]
Stefania commented on CASSANDRA-9288:
-------------------------------------
Patch for trunk is available here:
https://github.com/stef1927/cassandra/commits/9288
Aside from the class cast exception introduced by WrappingCompactionStrategy,
there was something fundamentally wrong in testParallelLeveledCompaction() in
that CASSANDRA-7272 changed LCS.getMaximalTask() to become a major compaction
task, so it did not make much sense to test this in parallel.
Prior to CASSANDRA-7272, LCS.getMaximalTask() was returning the first
background task, same as getNextBackgroundTask(). Comparing the source code
indicates that getNextBackgroundTask() is the old getMaximalTask(). So I
changed the test to call getNextBackgroundTask() instead of getMaximalTask().
I also picked the unrepaired compaction strategy since the repaired one did not
return any tasks.
Finally this code did not make much sense to me:
{code}
for (SSTableReader sstable : store.getSSTables())
{
assert sstable.getSSTableLevel() == sstable.getSSTableLevel();
}
{code}
So I removed it in favor of an assertion against the level an sstable is
reported to belong to, in the for loop just above.
[~krummas] I think you would be the best person to review this and to determine
if this test is sufficient or it can be improved further (e.g. repaired vs
unrepaired)?
> LongLeveledCompactionStrategyTest is failing
> --------------------------------------------
>
> Key: CASSANDRA-9288
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9288
> Project: Cassandra
> Issue Type: Test
> Reporter: Ariel Weisberg
> Assignee: Stefania
> Fix For: 3.x
>
>
> Pretty straightforward bad cast followed by some code rot where things return
> null that didn't used to.
> I suspect this has been broken since WrappingCompactionStrategy was
> introduced.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
