My preliminary tests show that I cannot reproduce it using seda flow. I still need to confirm that with my QA guys, as they've been able to produce it more reliably than I.

I tried putting Activemq in debug earlier but the output is so verbose it doesn't mean much to me. I need to throw a whole bunch of messages at it at the same time to produce the issue, so you can imagine what that logging looks like. Can you suggest what my log4j categories should be to get some meaningful output?

I can use seda for a while, but I really wanted to use the cool clustering that jms flow provides :-)

Thanks,

BJ


On Feb 28, 2007, at 1:15 PM, Guillaume Nodet wrote:

Have you tried (if possible) to run your tests with a SEDA flow to
see if the problem comes from the jms flow or elsewhere ?

On 2/28/07, William Blackburn <[EMAIL PROTECTED]> wrote:
My application uses the servicemix-3.1 release version and leverages
servicemix 'pojo' components heavily. Everything has been working
fine up until now, however the app is currently undergoing some
stress testing. During these tests, I lose message-exchanges
routinely. About 1 out of every 10 exchanges generated by the
components in my app never make it to their destination.

I am using JMSflow with an embedded activemq broker. I have tried
several Activemq persistence strategies, form no-persistence to kaha
and journalled jdbc, all with the same result. I have extensive
logging and there is never an error or warning logged. I have
reveiwed my onMessageExchange implementations in the components to
check that I am not silently consuming an exception case, I am not.

The issue appears to only be reproducible on higher-end machines with
two or more cpus when the app is under heavy load (ie: lots of
concurrent exchanges are firing between components.

Has anyone seen this before? With no exceptions or warnings and a
seemingly random pattern of success vs. failure, I am at a loss.



--
Cheers,
Guillaume Nodet
------------------------
Architect, LogicBlaze (http://www.logicblaze.com/)
Blog: http://gnodet.blogspot.com/

Reply via email to