On Mar 31, 2010, at 8:45 AM, Andy Gumbrecht wrote:
> On 31.03.2010 14:09, Kevan Miller wrote:
>> On Mar 30, 2010, at 11:59 AM, Andy Gumbrecht wrote:
>>
>>
>>> On 30.03.2010 16:18, Andy Gumbrecht wrote:
>>>
>>>> On 30.03.2010 16:01, Kevan Miller wrote:
>>>>
>>>>> On Mar 29, 2010, at 7:00 PM, David Blevins wrote:
>>>>>
>>>>>
>>>>>> So that occasional deadlock at startup we're seeing on some of the tests
>>>>>> is cause by the ActiveMQ 5 upgrade. Kevin had poked at it a bit, but
>>>>>> it's still unresolved.
>>>>>>
>>>>>> Just now noticing that the standalone server shutdown is also not
>>>>>> working, just hangs forever at "Stopping ResourceAdapter: My JMS
>>>>>> Resource Adapter"
>>>>>>
>>>>>> Couple release blocking issues if there's someone out there who is
>>>>>> looking for something to work on.
>>>>>>
>>>>> Are you running with Java 5 or Java 6? And I assume you are running on
>>>>> Mac OS...
>>>>>
>>>>> I have not seen the startup deadlock on Mac OS Java 6. IMO, it's a
>>>>> JVM/JSE issue. I haven't tested extensively. If we're seeing this issue
>>>>> on other platforms (not Java 5 Mac OS), then there's more motivation to
>>>>> fix...
>>>>>
>>>>> --kevan
>>>>>
>>>>>
>>>> So had a look at the ActiveMQ code and find this bit of code in RA.stop():
>>>>
>>>> while (endpointWorkers.size()> 0) {
>>>> ActiveMQEndpointActivationKey key =
>>>> endpointWorkers.keySet().iterator().next();
>>>> endpointDeactivation(key.getMessageEndpointFactory(),
>>>> key.getActivationSpec());
>>>> }
>>>>
>>>> It assumes that endpointDeactivation() will definitely stop 'a' worker -
>>>> no time-out.
>>>> With no concrete error I am just guessing that this worker fails to stop
>>>> for 'whatever' reason, and blocks indefinitely.
>>>>
>>>> It seems to be related to:
>>>> http://issues.apache.org/activemq/browse/AMQ-1447
>>>>
>>>> I am currently testing a possible solution, which basically gives up
>>>> waiting for RA.stop() after 5 seconds.
>>>>
>>>> Andy.
>>>>
>>>>
>>>>
>>> This is could be related to the 'Remote Shutdown' issue.
>>>
>>> The RAs were not getting 'stopped' if the the remote stop was called - do
>>> any tests call remote stop (without first having looked)?
>>>
>>> If ActiveMQ is not shut-down cleanly then I have seen it hang on the first
>>> restart (only the first restart, and not always) while it fixes it's data,
>>> this could be the observed 'occasional' deadlock!
>>>
>> This may be a problem, but is different from the deadlock problem that I
>> looked at. Is there a Jira open? Do we have information on the deadlock?
>>
>> The deadlock that I saw was a ClassLoader deadlock during a call to
>> ClassFileTransformer.transform(...)
>>
>> --kevan
>>
>>
>>
>>
> And in what way was it (the deadlock) related to ActiveMQ?
IMO, it was only "related" in that the deadlock occurred while running an
ActiveMQ test. ActiveMQ 5 would potentially exhibit different
threading/classloading behavior than ActiveMQ 4. In the stack trace that I saw,
I didn't see anything that indicated that ActiveMQ was doing something wrong.
The only platform that I've seen this error occur on is Mac OS Java 5. Unless
we're seeing the problem on reasonably current Java service levels, I would
ignore the Mac OS Java 5 problem.
--kevan
--kevan
