[
https://issues.apache.org/jira/browse/HBASE-14222?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14702746#comment-14702746
]
Heng Chen commented on HBASE-14222:
-----------------------------------
The problem is below:
{code}
1. if (getValue(oldValAndFlags) == 1) {
2. synchronized (this) { this.notifyAll(); }
3. }
{code}
{code}
1. if (getValue(oldValAndFlags) == 1) return;
2. synchronized (this) { this.wait(); }
{code}
there will be problem when wait Thread run to line 2 but lock was acquired by
notify thread, the result is wait thread will wait forever.
So we add condition check in wait thread sync block.
{code}
while(getValue(oldValAndFlags) != 1) {
this.wait();
}
{code}
when wait thread acquired lock if will check condition first, because notify
has run, the condition {{getValue(oldValAndFlags) != 1}} will be false.
and wait thread will not wait.
> Improve DrainBarrier
> --------------------
>
> Key: HBASE-14222
> URL: https://issues.apache.org/jira/browse/HBASE-14222
> Project: HBase
> Issue Type: Bug
> Components: util
> Reporter: Hiroshi Ikeda
> Assignee: Hiroshi Ikeda
> Priority: Minor
> Attachments: HBASE-14222-V2.patch, HBASE-14222.patch
>
>
> 1. {{DrainBarrier.stopAndDrainOps}} may wait forever if
> {{DrainBarrier.endOp}} changes its state and calls {{Object.notifyAll}} just
> before {{stopAndDrainOps}} enters the synchronized block to call
> {{Object.wait}}. Moreover, {{Object.wait}} may wake up false-positively, and
> {{stopAndDrainOps}} may break the block before outstanding operations are
> complete.
> 2. Some tests for {{DrainBarrier}} catch and ignore {{AssertionError}}
> explicitly thrown JUnit's {{fail}} method.
> The implementation of {{DrainBarrier}} is a little complex, and I'll fix and
> refactor it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
