[rabbitmq-discuss] HA queues

[Matthias Radestock]

Michael,

Michael Vierling wrote:
> We're using RabbitMQ for several internal projects.  However there have
> been requests for HA Queues.  HA queues would leverage a design similar
> to MasterSlave, i.e. http://activemq.apache.org/pure-master-slave.html
> However, unlike ActiveMQ, the feature could operate at the queue level,
> instead of the Broker level.  Currently when a RabbitMQ queue is
> declared it has a primary node content owner.  If the node hardware
> fails or is shutdown, clients will get a 404 error if they attempt to
> subscribe or publish to this queue, on other cluster nodes.
>  
> We're considering designing an Erlang patch to RabbitMQ to handle the
> 404 case, and support a MasterSlave queue failover.
>  
> Thoughts, comments?  Is this something already planned or prototyped?

We have thought about this. A lot. And yes, we are planning to introduce 
queues-that-survive-node-failures at some point in the not too distant 
future, but it's not something we are working on right now, so patches 
are welcome.

Doing this properly, such that the observable behaviour is consistent 
with the queue never having been disturbed, is hard. Matthew has been 
looking into this most recently and may be able to provide some pointers 
on potential pitfalls - he's on holiday atm but will be back next week.


Regards,

Matthias.