[rabbitmq-discuss] Cluster and memory usage

[Matthias Radestock]

On 11/11/11 14:36, Cheremukhin Andrey wrote:
> node1 and node2 have approximately count of publishers and messages. I
> am sure.  UI managment shows than queue don't have messages.

The backlog isn't in the queues, it's in the various buffers prior to 
that. And publishing a message to node1 when it needs to go into a queue 
on node2 is more expensive than publishing the same message to node2.

> Every node is receiving approximately 4K messages per second.

That may simply be too high a rate in your setup. If this is just a 
test, try reducing the rate to, say 1k/s and check whether the memory 
usage is more stable.

Note that nothing really bad should happen in any case. As I explained 
in my previous reply, eventually the memory alarm will trigger and 
rabbit will pause producers to allow the backlog to shrink.

Regards,

Matthias.