[rabbitmq-discuss] errors from management plugin appear to stop messages delivery

[Simon MacMullen]

On 12/04/11 23:16, Matthias Radestock wrote:
> Ian Ragsdale wrote:
>> It isn't currently happening, but I'll reenable the plugin and see
>> what happens. One other note is that when I did a 'rabbitmqctl
>> rotate_logs' the numbers would update again and data would flow for a
>> while, but then stop.
>
> It's possible that looking at 4000+ queues in the management plug-in,
> let alone with rabbitmqctl, puts quite a bit of load on the system,
> depending on how frequently you are doing that.

On my machine, viewing the queue list, refreshing every 5 seconds, with 
6000 queues takes on average 35% of one core assuming no errors. So 
that's not great, but should be tolerable. And it doesn't cause anything 
to lock up.

> Also, writing the reported error to the logs will be expensive and will
> block acceptance of new connections while that is taking place.

This however is very true - the entire data structure gets pretty 
printed and written to the logs when this error occurs. In my testing I 
wasn't able to make Rabbit *completely* unresponsive doing this but a 
small number of HTTP clients polling certainly made it very slow to respond.

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