[rabbitmq-discuss] Need help to interpret performance results

[Matthias Radestock]

On 07/12/11 03:18, tang qingxiong wrote:
> http://img522.imageshack.us/img522/4746/queuesrabbitmq1.jpg
>
> I noticed that the messages in 1 queue is increasing despite the
> fact that the incoming is still <= get/deliver.

Take a look at the details page for that queue. It may contain some
clues as to why you are seeing the above. In particular there should be
some (rather cryptic) information about the backing_queue_status that
contains the queue's internal rate calculations.

> one message could be up to 300kb as thats what I set the frame size
> to be (I assume that is the message size).

No, the frame_max setting controls the size of a single frame; messages 
automatically get fragmented across multiple frames when they exceed 
that size. And note that the frame_max includes the framing overhead. So 
if you set it to exactly your message size you'll actually be short of a 
few octets and the messages get split over two frames.

See 
http://www.rabbitmq.com/amqp-0-9-1-reference.html#connection.tune.frame-max 
for more info.

> Was wondering if it is a problem with the way I implement the
> clustering or code.
>
> my cluster config is as follows server 1 [
> {rabbit,[{vm_memory_high_watermark, 1.0},{frame_max,
> 0},{cluster_nodes,['rabbit at SERVER1','rabbit at SERVER2']}]} ]. server 2
>  [ {rabbit,[{vm_memory_high_watermark, 1.0},{frame_max,
> 0},{cluster_nodes,['rabbit at SERVER2','rabbit at SERVER1']}]} ]. . . . .

Nothing wrong with that except that setting the vm_memory_high_watermark 
to 1.0 is usually a bad idea. That's why your broker ran out of memory 
and crashed rather than paging messages to disk.

See http://www.rabbitmq.com/memory.html for more info.


Matthias.