[rabbitmq-discuss] rabbitmq dying

[Ben Hood]

On Thu, Jun 19, 2008 at 9:06 AM,  <David.Corcoran at edftrading.com> wrote:
> I'll try to reproduce it with a small test myself. The message "Mnesia is
> overloaded" seems to hit sometimes when the message queue length is only
> 200,000 or so. Is this normal? Or perhaps the better questions is when does
> Mnesia become overloaded? Does it need more disk or maybe more memory? Btw,
> our queues are 'non-durable'.

The messages themselves are not actually stored in mnesia. Mnesia just
maintains the existence of a queue with a unique name in a cluster.

The mnesia overloaded warning may indicate that there is a lot of
t-log activity caused by pending activities.

A good diagnostic for this is mnesia:info(). If we can get the remote
shell going, then the output of this would be good.

Furthermore having a script so that we can reproduce it here would be
good as well.

I also wondering why so many messages have been backed up. Can they
got get consumed at a fast rate?

>
>> There is a *nicer* way to shut a client down, but at the end of the
>> day, the server *should* be able to handle any type of behaviour from
>> a client.
>
> Yeah, we could be polite to our clients and let them shutdown nicely but
> because they're stateless we've always done it this way. If the patch isn't
> available in time I can have a look into it.

The patch I was talking about neatens out the server side handling of
clients just going AWOL, but I don't think that is going to solve your
problem, which is what we need to get to the bottom of.

Ben