[rabbitmq-discuss] use of an NFS for broker failover

[Matthew Sackman]

On Fri, Oct 01, 2010 at 09:48:58AM -0400, Adam Nelson wrote:
> > > What else should I be considering to achieve failover?
> >
> > That sounds good. NFS may not be the best tool for this job, but
> > it might just work. Do make sure that at no point two Rabbits
> > are using the same data. You may also find the "reconnect"
> > bit hard, how will your clients know which Rabbit to connect to?

Sticking a load balancer in front is one idea, mobile IPs is another
idea. Heck, pacemaker can even move NIC MACs between different machines
if you want to continue to use DHCP. Even DNS round robin will work -
the client just needs to keep on trying until it gets a connection
again.

> I agree with Marek.  NFS seems like it will introduce more problems than it
> solves.

Most SANs or NASs which export filesystems rather than block-level
storage support NFS or CIFS. The former is much simpler and more
battle tested. NFS is a perfectly good solution provided you ensure that
you don't have both Rabbits trying to access the store at the same time
(as they'll stomp over each other's files and then get very upset and
angry with each other).

Matthew