[
https://issues.apache.org/jira/browse/QPID-1391?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12648614#action_12648614
]
Martin Ritchie commented on QPID-1391:
--------------------------------------
Having taken a look at the stack trace the problem seems to be comming from the
BerkeleyDB MessageStore module.
Thead 'pool-1-thread-13' is blocked trying to add a new message to the channel
Unacknowledged Map.
The thread is blocked waiting for a lock held by thread 'pool-1-thread-27'
which is acknowledging a message locking the map.
Thread 27 is currently waiting in the BDBStore code for the completion of the
commit. Question is why is it taking so long, as in it is not completing at
all. Hours have passed and the code is still sitting at wait();
When (if) this wait-ing thread returns then that will release the locks for all
the currently waiting threads.
So points for further discussion:
1) [Slightly off Apache Qpid] BDBMessageStore L1804 synchronizes on 'this', IMO
this is a poor design as you cannot tell if the BDB code is also going to lock
on that object.
2) UnacknowledgeMessageMapImpl L:141 acknowledgeMessage: This is synchronizing
around the whole acknowledge method per TranscationalContext. This seems
unnecessary as we pass in the UMMI (this) to the method which then uses the
visitors to safely access the map in the NonTransactionalContext and the
LoclaTransactionalContext does not actually update the map so should not need
to lock at all.
> Reliability tests fail, broker is unable to process connections
> ---------------------------------------------------------------
>
> Key: QPID-1391
> URL: https://issues.apache.org/jira/browse/QPID-1391
> Project: Qpid
> Issue Type: Bug
> Components: Java Broker
> Affects Versions: M4
> Reporter: Aidan Skinner
> Assignee: Aidan Skinner
> Fix For: M4
>
> Attachments: stack.txt
>
>
> The reliability tests eventually cause the broker to lock up, it's still up
> but all threads a waiting on either a BDB lock or a senderLock. This is bad.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
