[rabbitmq-discuss] Distributing clients across a cluster

[Bill Moseley]

On Fri, Jan 14, 2011 at 6:58 AM, Mark J. Reed <markjreed at gmail.com> wrote:

> What do you all do to distribute activity across a cluster?  That is,
> given a cluster of rabbitmq brokers and a group of client applications
> that each expect to be configured to talk to a single broker, what do
> I do?
>
> 1. Stand up TCP load-balancer in front of the cluster?
> 2. Assign a cluster node to each client pseudo-randomly?
> 3. Set up each client host as an additional rabbitmq broker and point
> the applications at localhost?
>

I've been spending a lot of time looking for the same answers.  Is there
documentation on actual hardware configuration and management of servers in
a production environment?

To me, it is important that producers (web processes in my case) can always
talk to a broker.  So, I've been looking at #1 with two HAProxy servers (one
as fail-over with Heartbeat) in front of a cluster of nodes.  Only partial
success with that so far.

On the same topic, I'm also curious how clients are suppose to behave.  If I
have workers listening on a queue how should they handle a network failure?
 Do any of the clients manage reconnecting on error?

BTW -- I also looked at the HA docs at
http://www.rabbitmq.com/pacemaker.html -- but that's replicating queue data,
correct? Is there a need for DRDB if willing to accept loss of messages?




-- 
Bill Moseley
moseley at hank.org
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20110114/3d00d066/attachment.htm>