[rabbitmq-discuss] Topic routing

Wed Oct 22 18:01:13 BST 2008

Brian,

On Wed, Oct 22, 2008 at 5:19 PM, Brian Sullivan <bsullivan at lindenlab.com> wrote:
> Is "N" the number of subscriptions?

n is the number of bindings attached to an exchange. Whether or not
somebody is actually consuming something from any of the queues, or
whether an immediate flag causes undeliverable messages to be dropped,
is irrevalent. What counts is the runtime complexity of deciding where
you have to route to, not what happens to a message after you have
decided where it is going to go.

If so, I think we're fine, since my
> number of consumers will be relatively low.  If it's the number of topics,
> then that becomes a little more troublesome if I want to get clever with my
> routing keys to help distribute the consumption load across nodes.

When you say "number of topics", do you mean:

1. The number of topic exhanges
2. The number of bindings attached to a single topic exchange

?

The former would not be a problem, because this would represent
natural application partitioning.

The latter is what I think you mean, and is, as indicated above, what
the O(n) cost is based on for a topic exchnage, because wildcards
force a linear scan.

HTH,

Ben