[
https://issues.apache.org/activemq/browse/AMQCPP-277?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=59190#action_59190
]
Gerald Kaas commented on AMQCPP-277:
------------------------------------
I am also having problems with ActiveMQ-CPP freezing on Windows when calling
setMessageListener. I have been spending a day or so trying to track it down
and there seems to be a race-condition deadlock occurring within the
Thread::join() function. You are not probably seeing this with pthreads since
you are just calling pthread_join. On the Windows platform you are calling
Thread::join( INFINITE, 0 ). The later uses a mutex to try to determine when a
thread is dead but as far as I can tell, the thread is still terminating and
thus some other non-mutex state variables are getting mixed up. I haven't tried
but I bet if you change Thread::join() to use Thread::join( INFINITE, 0 ) for
both Windows and Linux, you will see the problems on both platforms.
I was able to reproduce the problem very reliably by subscribing to various
topics in a loop. It would ALWAYS freeze in ActiveMQSessionExecutor::stop ->
taskRunner->shutdown -> this->thread->join. You may need a multiple core
machine to see the problem (I have 4 cores). The DedicatedTaskRunner thread
would already be terminated but the join would block.
I fixed the problem by replacing
Thread::join() ...
{
...
Thread::join ( INFINITE, 0 );
...
}
to
{
...
WaitForSingleObject(properties->handle, INFINITE);
...
}
to match the same logic in Linux.
Your threadWorker function may also be broken on Windows. On LINUX you called
thread_detach which doesn't actually terminate the thread and then sets the
state variable. On Windows you go ahead an terminate the thread and this set
the state variable. You might want to move that assignment.
Please let me know what you find out. As far as I know, this bug hasn't been
fixed in the trunk and is very reliable.
> Freeze when creating multiple Consumers
> ---------------------------------------
>
> Key: AMQCPP-277
> URL: https://issues.apache.org/activemq/browse/AMQCPP-277
> Project: ActiveMQ C++ Client
> Issue Type: Bug
> Affects Versions: 3.1
> Environment: Windows 7 x64, ActiveMQ Server 5.3, Java 6.0.17, Visual
> Studio 2008 SP1, Windows 7 SDK, APR 1.3.9
> Reporter: Eddie Fast
> Assignee: Timothy Bish
> Fix For: 3.2.0
>
>
> We create multiple MessageConsumers with different selectors. This can be
> upwards of 50 or so consumers per client.
> We recently upgraded to 3.1 from 2.2.2. We also upgraded our ActiveMQ
> Server to 5.3 from 5.2.
> After upgrading, we've been seeing random, but frequent freezing in our
> clients. I've narrowed it down to where it creates the MessageConsumer and
> calls setMessageListener();
> It seems to be stuck, waiting for a thread to join.
> I've modified the main.cpp to exhibit this behavior, but I've only been able
> to reproduce it randomly. It seems to happen more frequently if you stick a
> breakpoint on this line in Thread.cpp, Thread::join():
> Thread::join( INFINITE, 0 );
> Here's the callstack where it freezes, which is consistent in our clients as
> well as the main.cpp example:
> ntdll.dll!76fff871()
> [Frames below may be incorrect and/or missing, no symbols loaded for
> ntdll.dll]
> ntdll.dll!76fff871()
> KernelBase.dll!75e20816()
> kernel32.dll!755b1138()
>
> activemq-cppd.dll!decaf::internal::util::concurrent::ConditionImpl::wait(decaf::util::concurrent::ConditionHandle
> * condition=0x0239e0e0, __int64 mills=4294967295, __int64 nanos=0) Line 110
> + 0x10 bytes C++
> activemq-cppd.dll!decaf::util::concurrent::Mutex::wait(__int64
> millisecs=4294967295, int nanos=0) Line 124 + 0x20 bytes C++
> activemq-cppd.dll!decaf::lang::Thread::join(__int64
> millisecs=4294967295, unsigned int nanos=0) Line 464 + 0x36 bytes C++
> activemq-cppd.dll!decaf::lang::Thread::join() Line 421 C++
> > activemq-cppd.dll!activemq::threads::DedicatedTaskRunner::shutdown()
> > Line 83 C++
> activemq-cppd.dll!activemq::core::ActiveMQSessionExecutor::stop() Line
> 110 C++
> activemq-cppd.dll!activemq::core::ActiveMQSession::stop() Line 807
> C++
>
> activemq-cppd.dll!activemq::core::ActiveMQConsumer::setMessageListener(cms::MessageListener
> * listener=0x0018fe0c) Line 523 C++
> vs2005-activemq-example.exe!HelloWorldConsumer::run() Line 247 + 0x5b
> bytes C++
>
> activemq-cppd.dll!decaf::lang::ThreadProperties::runCallback(decaf::lang::ThreadProperties
> * properties=0x01df7f68) Line 133 + 0x11 bytes C++
> activemq-cppd.dll!`anonymous namespace'::threadWorker(void *
> arg=0x01df7f68) Line 204 + 0x9 bytes C++
> msvcr90d.dll!_callthreadstartex() Line 348 + 0xf bytes C
> msvcr90d.dll!_threadstartex(void * ptd=0x01dfafd8) Line 331 C
> To modify main.cpp, it's pretty simple. Instead of creating 1
> MessageConsumer in HelloWorldConsumer::run(), I've created 50:
> {quote}
> // put this line with the class variables
> std::vector<MessageConsumer*> consumers;
> for ( int i = 0; i < 50; i++ )
> {{
> // Create a MessageConsumer from the Session to the Topic or Queue
> consumers.push_back( session->createConsumer( destination ) );
> consumers.back()->setMessageListener( this );
> }}
> {quote}
> I'm using the standard .conf for the server. I've reverted to 2.2.2 and
> have no issues.
> Any ideas?
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
