[
https://issues.apache.org/jira/browse/CASSANDRA-10377?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vovodroid updated CASSANDRA-10377:
----------------------------------
Attachment: 10377.patch
It seems I found reason for issue. Function *recycleSegment* is called from
different threads, then it removes segment from activeSegments queue and then
it queues segment deletion task, even if segment not found in activeSegments
queue.
Due to racing it possible that *recycleSegment* will be called twice and thus
two identical deletion task will be queued.
Proposed patch check remove from activeSegments queue result and delete segment
only if it found in queue.
> AssertionError: attempted to delete non-existing file CommitLog
> ---------------------------------------------------------------
>
> Key: CASSANDRA-10377
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10377
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Environment: CentOS 7.1/x64
> Reporter: Vovodroid
> Priority: Critical
> Attachments: 10377.patch
>
>
> After several hours of script tests (create and drop users, keyspaces and
> tables) exception is thrown:
> {code}
> ERROR 02:58:39 Failed managing commit log segments. Commit disk failure
> policy is stop; terminating thread
> java.lang.AssertionError: attempted to delete non-existing file
> CommitLog-5-1442599226756.log
> at
> org.apache.cassandra.io.util.FileUtils.deleteWithConfirm(FileUtils.java:122)
> ~[main/:na]
> at
> org.apache.cassandra.io.util.FileUtils.deleteWithConfirm(FileUtils.java:149)
> ~[main/:na]
> at
> org.apache.cassandra.db.commitlog.CommitLogSegment.discard(CommitLogSegment.java:314)
> ~[main/:na]
> at
> org.apache.cassandra.db.commitlog.CommitLogSegmentManager$2.run(CommitLogSegmentManager.java:374)
> ~[main/:na]
> at
> org.apache.cassandra.db.commitlog.CommitLogSegmentManager$1.runMayThrow(CommitLogSegmentManager.java:155)
> ~[main/:na]
> at
> org.apache.cassandra.utils.WrappedRunnable.run(WrappedRunnable.java:28)
> [main/:na]
> at java.lang.Thread.run(Thread.java:745) [na:1.8.0_45]
> {code}
> I added some logs to *deleteWithConfirm* and it showed that this file really
> was deleted by previous delete action, i.e. it was second attempt to delete
> the same log. Commit log with next number exists in the same time, so log was
> switched.
> I disabled assert and it seems to have no no bad effect.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
