[rabbitmq-discuss] socket_error

[Emile Joubert]

Hi,

On 20/03/13 16:02, Elizabeth Liao wrote:
> We are currently running on Fedora 11 and have not been able to
> figure out how to reproduce the issue.  The actual problem that we're
> chasing down is that our hard disks are getting hammered but we can't
> find the root cause.  We're essentially wondering if the errors are
> cause or effect.

The {socket_error,enomem} error relates to networking rather than disk.
I would be surprised if this was the cause of disk use.

I've not come across the same error, but it would seem to indicate
resource exhaustion of some kind in the network stack.

If you are not using the latest version of Erlang (R16B at the time of
writing) then try upgrading.

You could also check with a network analyser whether the broker is
exposed to network traffic with unusually large TCP packets.

The only other suggestion is to make sure all your TCP/IP kernel
parameters are set to sane values, or to try more conservative values.
You will need to refer to your OS documentation.

If you do discover a way of reproducing the problem then please get back
to us.



-Emile