You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Sam Tunnicliffe (JIRA)" <ji...@apache.org> on 2016/07/19 16:06:20 UTC

[jira] [Comment Edited] (CASSANDRA-12218) IndexSummaryManagerTest.testCancelIndex is flaky

    [ https://issues.apache.org/jira/browse/CASSANDRA-12218?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15384370#comment-15384370 ] 

Sam Tunnicliffe edited comment on CASSANDRA-12218 at 7/19/16 4:05 PM:
----------------------------------------------------------------------

The problem is that there is only a single point in {{IndexSummaryRedistribution::redistributeSummaries}} which checks the flag to see if it should halt. On occasion, this point is passed before the stop is actually requested, the redistribution completes normally and no {{CompactionInterruptedException}} is thrown. 

With the linked branch, I haven't been able to repro the failure in over 150 runs (without it I couldn't get above ~50 iterations without a failure). 

The patch may seem quite large, but it's really just a refactoring. The significant change is an extension of {{IndexSummaryRedistribution}} that pauses until signaled before running. Only {{testCancelIndex}} makes use of this, all other tests are unchanged.

||branch||testall||
|[12218-3.9|https://github.com/beobal/cassandra/tree/12218-3.9]|[testall|http://cassci.datastax.com/view/Dev/view/beobal/job/beobal-12218-3.9-testall]|



was (Author: beobal):
The problem is that there is only a single point in {{IndexSummaryRedistribution::redistributeSummaries}} which checks the flag to see if it should halt. On occasion, this point is passed before the stop is actually requested, the redistribution completes normally and no {{CompactionInterruptedException}} is thrown. 

With the linked branch, I haven't been able to repro the failure in over 150 runs (without it I couldn't get above ~50 iterations without a failure). 

The patch may seem quite large, but it's really just a refactoring. The significant change is an extension of {{IndexSummaryRedistribution}} that pauses until signaled before running. Only {{testCancelIndex}} makes use of this, all other tests are unchanged.

||branch||testall||dtest||
|[12218-3.9|https://github.com/beobal/cassandra/tree/12218-3.9]|[testall|http://cassci.datastax.com/view/Dev/view/beobal/job/beobal-12218-3.9-testall]|[dtest|http://cassci.datastax.com/view/Dev/view/beobal/job/beobal-12218-3.9-dtest]|


> IndexSummaryManagerTest.testCancelIndex is flaky
> ------------------------------------------------
>
>                 Key: CASSANDRA-12218
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-12218
>             Project: Cassandra
>          Issue Type: Test
>            Reporter: Joshua McKenzie
>            Assignee: Sam Tunnicliffe
>              Labels: unittest
>
> Test appears to be flaky given the history:
> [History|http://cassci.datastax.com/job/cassandra-3.9_testall/25/testReport/org.apache.cassandra.io.sstable/IndexSummaryManagerTest/testCancelIndex/history/]
> {noformat}
> Error Message
> Expected compaction interrupted exception
> Stacktrace
> junit.framework.AssertionFailedError: Expected compaction interrupted exception
> 	at org.apache.cassandra.io.sstable.IndexSummaryManagerTest.testCancelIndex(IndexSummaryManagerTest.java:655)
> {noformat}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)