[
https://issues.apache.org/jira/browse/OOZIE-1922?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15073124#comment-15073124
]
Rohini Palaniswamy commented on OOZIE-1922:
-------------------------------------------
bq. Sine we are in release function, which means that lock is acquired by
current thread. So, we don't have to check if lock is held by other thread or
not.
You have to. If it is read locks, more than one thread can hold the lock.
Currently only write locks are acquired in the code though, but still logic
needs to be correct as there are APIs to acquire read lock.
bq. I didn't remove getParticipantNodes because it was already part of the
condition and was not sure if it's break anything.
If you are really concern about performance, we can remove getParticipantNodes
from condition.
Yes. Performance and load on zookeeper is very important.
> MemoryLocksService fails if lock is acquired multiple times in same thread
> and released
> ---------------------------------------------------------------------------------------
>
> Key: OOZIE-1922
> URL: https://issues.apache.org/jira/browse/OOZIE-1922
> Project: Oozie
> Issue Type: Bug
> Reporter: Purshotam Shah
> Assignee: Purshotam Shah
> Attachments: OOZIE-1922-V1.patch, OOZIE-1922.1.patch,
> OOZIE-1922.2.patch, OOZIE-1922.3.patch
>
>
> ReentrantLock can be acquired multiple times in same thread. For multiple
> acquire call, ReentrantLock hold count is incremented by one.
> So if we acquire lock multiple time from same thread, all will be successful
> and hold count is increased for every call.
> But if we release lock, MemoryLocksService ignore the count and deletes the
> lock. Even if it's held by some command.
> Simple step can reproduce it.
> {code}
> service.getWriteLock("test", 5000); //writeHoldCount = 1
> MemoryLockToken lock = (MemoryLockToken)service.getWriteLock("test",
> 5000); //writeHoldCount = 2
> lock.release(); //writeHoldCount = 1
> lock = (MemoryLockToken)service.getWriteLock("test", 5000);
> //writeHoldCount = 1, it should be 2.
> {code}
> {code}
> @Override
> public void release() {
> int val = rwLock.getQueueLength();
> if (val == 0) {
> synchronized (locks) {
> locks.remove(resource);
> }
> }
> lock.unlock();
> }
> }
> {code}
> MemoryLocks should check hold count before removing lock.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
