[rabbitmq-discuss] HA policies [was: RabbitMQ 2.6.0 is released]

[Brad Knowles]

On Sep 2, 2011, at 11:25 PM, Matthias Radestock wrote:

>> The "all nodes" setting really isn't a solution to that.
> 
> Why is that?
> 
> I would have thought that 'all' is a sensible option for the majority of
> users that want HA. And it does not require argument changes when the topology evolves.

Mirroring the queue across all nodes means that you don't have any scalability for that queue -- you can't load-balance or round-robin or distribute the work by the routing key or anything, because all nodes will see all traffic for that queue.

Putting yourself in a roomful of mirrors with no exit doesn't actually give you hundreds or millions of more copies of yourself to do more work, just more images that all look more or less the same.

-- 
Brad Knowles <bknowles at ihiji.com>