[rabbitmq-discuss] RabbitMQ and Stomp topic behavior

[Steve Powell]

Massimo,
I've run the scenario on my 2-core 2Gb Ubuntu 32-bit VM (final changes, I removed vhost (default <<"/">>) from the config file).

The benchmark runs for about two lines of dots and then the rabbitmq_server gets a segmentation fault! and terminates.  The benchmark doesn't notice and ploughs on regardless.  Disappointingly I found no log information to explain the failure.

I added {vm_memory_high_watermark, 0.1} to the configuration, and switched the log file output to a local directory, and the benchmark and running broker (and benchmark) seem to continue fairly stably for ever!!  (I'm assuming the benchmark has no termination coded, it didn't seem to stop by itself, so I interrupt it after about half an hour.)

I notice several things:
rabbitmq_server process runs at full CPU (200% in my virtual machine) most of the time, with occasional lapses to 0 (zero), when the benchmark and the server appear to pause for breath.
the virtual memory requirements for rabbitmq_server seem to be below 0.5Gb, most of the time, but occasionally creep up to 0.7 or 0.8Gb, and then drop to about 0.2Gb.  By comparison, the java stomp-benchmark process runs at 723Mb all the time.
the memory (as opposed to virtual memory) requirements for rabbit_server appear to run between 0.54 to 0.75 of the virtual requirements.
I suspect that, when set to 0.4 (the default vm_memory_high_watermark ratio) the vMem occasionally creeps above the maximum allowed on this machine, which is about 1.6/1.7Gb, and that's when we get the segmentation fault.  This is worth investigation.  Also the CPU glitches look interesting; this may be throttling, or some sort of Erlang process garbage collection of which I am ignorant.

So where are the benchmark results put?  I see nothing produced (save a truncated rabbitmq_topic file) when I interrupt it.  Is this another problem with my sbt/workbench setup?

Steve Powell  (a happy bunny)
----------some definitions from the SPD----------
Rigatoni n. A prime-ministerial grin.
Nigella n. The next bottle-size up from a Nebuchadnezzar.
Homily adv. Rather like a frenchman.

On 16 Aug 2011, at 11:04, Steve Powell wrote:

> Massimo,
> I got a response from stomp-benchmark on github; apparently the sbt moved to xsbt github is not compatible with the benchmark, so I managed to get a 0.7.7 version which apparently works.  now running your scenario with your configuration, against rabbitmq 2.5.1.
> 
> Minor changes, --host localhost --port 61223 are compatible with your scenario, I believe.
> 
> Currently monitoring memory during run (at 1Gb in a 2Gb mem machine); will get back to you here.
> 
> Steve Powell
> steve at rabbitmq.com
> [wrk: +44-2380-111-528] [mob: +44-7815-838-558]
> 
> On 15 Aug 2011, at 17:37, Steve Powell wrote:
>> http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2011-August/014555.html
> 
> _______________________________________________
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
> https://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20110816/6ca65fbf/attachment.htm>