[
https://issues.apache.org/jira/browse/HBASE-14977?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15060704#comment-15060704
]
Enis Soztutar commented on HBASE-14977:
---------------------------------------
bq. That is because of a deadlock. ScheduledChore is heavily synchronized, even
getName() is synchronized and toString(). I attached small addendum which fixes
this issue.
Yeah, I was afraid of a possible deadlock in introducing an extra condition.
But it seemed (from my brief look at the code) that we do not hold any locks in
the thread calling shutdown().
ScheduledChore.getName() is synchronized? That does not make sense.
Is the deadlock between ScheduledChore.cancel() calling choreService.cancel()
while holding chore lock, and ChoreService.shutdown() calling toString() while
holding lock for the service.
I think a better addendum would be to declared name, period, etc final, and
un-synchronize the getName() and getPeriod() methods.
> ChoreService.shutdown may result in ConcurrentModificationException
> -------------------------------------------------------------------
>
> Key: HBASE-14977
> URL: https://issues.apache.org/jira/browse/HBASE-14977
> Project: HBase
> Issue Type: Bug
> Reporter: Vladimir Rodionov
> Assignee: Vladimir Rodionov
> Priority: Minor
> Fix For: 2.0.0, 1.2.0, 1.3.0, 1.1.4
>
> Attachments: HBASE-14977-add.patch, HBASE-14977-v1.patch
>
>
> As seen in this test:
> https://builds.apache.org/job/HBase-1.3/jdk=latest1.8,label=Hadoop/425/artifact/hbase-server/target/surefire-reports/org.apache.hadoop.hbase.regionserver.compactions.TestFIFOCompactionPolicy-output.txt
> We need to make shutdown method synchronized to avoid this issue.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
