[rabbitmq-discuss] aliveness-test issue

[Simon MacMullen]

On 11/04/11 18:24, Byron Albert wrote:
> I am using the aliveness-test to test via zenoss and after a few weeks
> the test stops responding. If I run it on a different vhost it works.
> This is getting tested every 5 min so wondering if there is some counter
> that is overflowing or something…

Hmm. No, there's no counter. I just ran jmeter against the 
aliveness-test URI for about 20k iterations, which should be equivalent 
to several weeks at one request every 5 min and it was fine.

Looking at the code though, I think it could get wedged if the aliveness 
test ever blows up after declaring the consumer but before cancelling it 
- you'd have a stale consumer that stuck around and would probably 
consume the test message first. I still can't figure out how that would 
actually happen though. I don't suppose you have logs that show the 
response the first time it got stuck (or rather, just before that)?

That would explain why switching to a different vhost fixes it - the new 
queue would not have a stale consumer. You could also fix it by deleting 
the queue - it'll get recreated when needed.

Anyway, I'll file a bug to ensure we always tear down the consumer here. 
Thanks for the report.

Cheers, Simon
-- 
Simon MacMullen
Staff Engineer, RabbitMQ
SpringSource, a division of VMware