[rabbitmq-discuss] Single Point of Failure

[Alexis Richardson]

James

It's good to hear that your colleague had good experiences with RabbitMQ.

I'm confused by the statement "couldn't be placed behind a
load-balancer".  That's not the case for RabbitMQ which *can* be used
with a load balancer.  Here's Josh from the Nokia team talking about
their set-up: http://www.joshdevins.net/2010/04/16/rabbitmq-ha-testing-with-haproxy/

See also our HA page here: http://www.rabbitmq.com/pacemaker.html
This an 'active/passive' set-up.

I think that perhaps your colleague meant that RabbitMQ lacks an
'active/active' style HA set-up.  In this set-up there can be two
'live' copies of the broker state, so that if one fails, the other is
available immediately.  It is correct that RabbitMQ cannot do this
today, but we have a solution in the works which has been discussed on
this list recently.  If you can't find it in the archives let us know.

BTW, if you look at Josh's blog post above you can also click through
the the Beetle project from Xing.  This provides an active/active dual
redundant type set-up by running two brokers and sending all messages
to both.  It's an application layer solution, specific to Ruby and
using the consumers to de-dup.  One reason many people shy away from
such a set up is that you need to have a thorough understanding of
what can go wrong.  By contrast the new Rabbit HA will provide the
capability within the broker itself, which should make it easier to
manage and generally more predictable...

In the new HA there *will* be a concept of a master.  This is
described in the draft documentation which Matthew posted to this list
recently.

alexis



On Thu, Jun 23, 2011 at 3:27 AM, James Carr <james.r.carr at gmail.com> wrote:
> Hi All,
>
> Had a half day of architecture meetings where we discussed rabbitMQ as
> an essential component to a new large scale project and one of the
> attendees mentioned that they had used RabbitMQ at his previous job
> awhile back. He said it worked great but one concern was that it had a
> single point of failure and it couldn't be placed behind a load
> balancer. He didn't know if that had changed but was just curious as
> it was something that had concerned him.
>
> However, he also added that they had been using it for a year by the
> time he left and no problems had cropped up. :)
>
> So has this changed any? Do clusters rely on a master node to be running?
>
>
> Thanks,
> James
> _______________________________________________
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
> https://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss
>