A possible browsing concurrency issue in org.apache.activemq.broker.region.Queue
--------------------------------------------------------------------------------
Key: AMQ-3319
URL: https://issues.apache.org/jira/browse/AMQ-3319
Project: ActiveMQ
Issue Type: Bug
Reporter: Przemek Bruski
Some time ago clients of our ActiveMQ instance locked up while browsing.
Analysis of the log files showed a large amount of:
{code}
java.util.NoSuchElementException
at java.util.LinkedList.remove(LinkedList.java:788)
at java.util.LinkedList.removeFirst(LinkedList.java:134)
at
org.apache.activemq.broker.region.Queue.getNextBrowserDispatch(Queue.java:1341)
at org.apache.activemq.broker.region.Queue.iterate(Queue.java:1463)
at
org.apache.activemq.thread.PooledTaskRunner.runTask(PooledTaskRunner.java:122)
at
org.apache.activemq.thread.PooledTaskRunner$1.run(PooledTaskRunner.java:43)
at
java.util.concurrent.ThreadPoolExecutor$Worker.runTask(ThreadPoolExecutor.java:886)
at
java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:908)
at java.lang.Thread.run(Thread.java:637)
{code}
Happening before lockup.
It seems there is a problem in Queue class, which uses non thread safe
LinkedList collection. Additions and removals to/from this collection are
wrapped by a shared readLock, which means there is no guard against concurrent
modification and there is also a possibility of a race condition between
isEmpty and removeFirst call during concurrent usages of getNextBrowserDispatch
(if they are possible).
I think the easiest fix is to switch from LinkedList to ConcurrentLinkedQueue
and make use of Queue methods to access the collection (because they allow
single step isEmpty/remove call). I am attaching a patch that does it. I've
left the readLocks in case they are used to block writes someplace else, but
they are not needed anymore for concurrency control over the new collection.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
