[rabbitmq-discuss] behavior when erlang processes exhausted

[Matthew Sackman]

On Thu, May 26, 2011 at 08:25:20AM -0400, Aaron Westendorf wrote:
> Thanks guys, I expected that. For what it's worth, Rabbit hung in
> there for several hours before finally breaching said horizon.

The problem really amounts to the same as why our memory monitoring is
very very conservative: we just have no idea what the rest of the Erlang
stdlibs and kernel gets up to, so monitoring our own usage of processes
is never going to be a reliable metric as for some reason the kernel
could suddenly decide to start using hundreds more processes. In
reality, that's not especially likely to happen, but it's still
complexity with thorns that we've yet to grasp at.

> We're going to adjust our settings and add some monitoring for how
> close we are to the limit.

Sounds sensible.

Matthew