[rabbitmq-discuss] Rabbitmq 2.8.2 crash using node-amqp client

Fri Jun 1 12:25:17 BST 2012

On 01/06/12 12:20, Andrei Vereha wrote:
> On Fri, 01 Jun 2012 11:45:39 +0100
> Tim Watson<tim at rabbitmq.com>  wrote:
>
>> On 01/06/12 11:44, Andrei Vereha wrote:
>>> Hello,
>>>
>>> Rabbitmq is crashing with one of these messages:
>>> Slogan: binary_alloc: Cannot allocate 245000040 bytes of memory (of
>>> type "binary").
>>> in /var/lib/rabbitmq/erl_crash.dump
>>> or killed by OOM killer:
>>> Out of memory: Kill process 29165 (beam) score 889 or sacrifice
>>> child
>>>
>>> I attached the script I used to stress it with. If it doesn't crash,
>>> try to start 2 instances of this script. The script is wrote in
>>> node.js and uses the node-amqp library.
>> Well it *is* possible to crash the vm by exhausting all available
>> memory and/or triggering the OOM Killer on Linux. Take a read of
>> http://www.rabbitmq.com/memory.html for more details on how to manage
>> this.
> I read the page above, and it seems that the vm_memory_high_watermark
> has no effect in this case.
> Configuration:
>
> [{rabbit, [{vm_memory_high_watermark, 0.2}]}].
>
> The log file says:
>
> =INFO REPORT==== 1-Jun-2012::14:01:45 ===
> vm_memory_high_watermark set. Memory used:105251880 allowed:103223296
>
> =INFO REPORT==== 1-Jun-2012::14:01:45 ===
>      alarm_handler:
> {set,{{resource_limit,memory,'rabbit at rabbitmq-03'},[]}}
>
> When I look into the management interface the connections are in
> "running" state, though the memory alarm is triggered(the memory
> section is red in the overview interface). And the memory usage is just
> growing.
Your memory limit is too low for this.
>
> When I try to send a big message from python using Pika, I cannot
> reproduce the crash. That's why I attached the example client in
> node.js.

A message's size should not be greater than a 10th (or in extremis 1/4) 
of the available RAM. Where is your Pika code?

> Thank you,
>   Andrei
>
>