[
https://issues.apache.org/jira/browse/CASSANDRA-12359?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15403526#comment-15403526
]
Stefania commented on CASSANDRA-12359:
--------------------------------------
This can be reproduced reliably by fixing the seed to the same seed that failed
on Jenkins: 2080431860597L. I've also reproduced it another couple of times
locally, it seems on average we have one failure every 100 to 150 runs.
For seed 2080431860597L, despite inserting 50 corrupt bytes at position 112 of
sstable no. 24, the sstable still gets compacted. This is because we read
through the corrupted bytes without complaining: vint encoding assume a
positive byte is a valid int, flags don't particularly raise problems, neither
do ascii types.
I've replaced Ascii types with Long types, which are fixed size types and
should have a better chance at detecting corrupted bytes, and I've doubled the
corruption size. I've already run the test locally 100 times and I've launched
another multiplex of 250 times on Jenkins:
[3.9 patch|https://github.com/stef1927/cassandra/commits/12359-3.9]
[multiplexed
results|http://cassci.datastax.com/view/Dev/view/stef1927/job/stef1927-testall-multiplex/16/]
Alternatively we could fix the corrupted bytes to zero or -1, and only leave
the corruption position as random.
> BlacklistingCompactionsTest.testBlacklistingWithSizeTieredCompactionStrategy
> is flaky
> -------------------------------------------------------------------------------------
>
> Key: CASSANDRA-12359
> URL: https://issues.apache.org/jira/browse/CASSANDRA-12359
> Project: Cassandra
> Issue Type: Bug
> Reporter: Joel Knighton
> Assignee: Stefania
>
> example failure:
>
> [http://cassci.datastax.com/job/cassandra-3.9_testall/50/testReport/junit/org.apache.cassandra.db.compaction/BlacklistingCompactionsTest/testBlacklistingWithSizeTieredCompactionStrategy/]
> Google suggests that this test is somewhat flaky historically, but we don't
> have logs from any of the older failures any more.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
