[rabbitmq-discuss] Order of publication to queues and HA nodes

Tue Jan 28 15:07:56 GMT 2014

On 28/01/14 14:21, Duncan McIntyre wrote:
> Let's say I have a fanout exchange bound to a set of HA queues across 3
> nodes.
>
> When a message is published to the exchange, is there any deterministic
> ordering when the messages become available on the queues?

No.

RabbitMQ gives ordering guarantees for any given (publishing channel) -> 
(queue) -> (consuming channel) route. There aren't any ordering 
guarantees that make sense when more than one queue is involved, since 
the queues run in parallel.

> Some
> customers care that there should be no advantage to one consumer over
> another in terms of when such a message becomes available.

Depends what you mean. It's obviously the case that the queues cannot 
all receive the message simultaneously and transmit to all consumers 
simultaneously too; even if we have more cores than consumers we 
probably have fewer network cables...

But if you just mean "is there a systemic bias"? - the answer is 
probably not. The publishing channel will dispatch the message to 
multiple queues in a predictable order - but it's sending the message 
asynchronously. Which queues receive the message first is then in the 
hands of the scheduler - which is likely to impart enough randomness to 
things for most purposes.

Having said which, I haven't tried to test this. Have you?

Cheers, Simon

-- 
Simon MacMullen
RabbitMQ, Pivotal