[rabbitmq-discuss] LATEST.LOG growth on clustered ram node

[Matthias Radestock]

Chris,

On 28/06/11 02:55, Chris Madden wrote:
> I've whipped up a pair of python scripts that demonstrate the issue.
> [...] the LATEST.LOG grows decently fast, maybe 15 meg an hour.

I can reproduce this.

> If there is something pathological in my scripts, I'd love to hear it
> (beyond the aforementioned anti-pattern)...

There is nothing pathological in the scripts, though they are an 
exercise in AMQP anti-patterns ;) (you really shouldn't be using a topic 
exchange; a direct exchange, in fact the default exchange, should do 
just fine; take a look at 
http://www.rabbitmq.com/tutorials/tutorial-six-python.html when you have 
a spare moment).

I've filed a bug. My investigation so far indicates that this may well 
be a problem in mnesia. We have a somewhat unusual configuration in that 
that our 'ram' nodes aren't really ram nodes in the mnesia sense - they 
are fully-fledged disk nodes which happen to have no local disk copies 
of tables. It is conceivable that this induces a faulty code path in the 
log handling code.

Until we fix that... I am pretty sure that in your setup the disk usage 
growth rate will be proportional to the queue creation rate. So if you 
eliminated the "one queue per request" anti-pattern that would probably 
give you plenty of headroom before bumping into any disk space limits.

Thanks for reporting this problem and taking the time constructing a 
reproducible test. Much appreciated.


Regards,

Matthias.