[rabbitmq-discuss] High availability questions

[Simon MacMullen]

On 07/12/12 14:36, Matthias Reik wrote:
> What would it take to allow a rolling update of the cluster? i.e. you
> take one node out of the cluster, upgrade it (ideally both Erlang as
> well as RabbitMQ) and bring it back into the cluster, wait until the
> nodes have fully synchronized and then continue with the next.
>     - database schema fixed?
>     - messages schema between nodes fixed?
>     - ...?

Yes.

Starting with 3.0.1 we intend to allow this for patch upgrades (e.g. 
3.0.x -> 3.0.y). Unfortunately we had to tweak 3.0.1 a bit versus 3.0.0 
to do this, so you won't be able to do a rolling upgrade of 3.0.0 to 3.0.1).

> According to the philosophy of Erlang it should even be possible to even
> change version of a running node (at least that was my understanding;
> upgrade your software with ongoing phone calls are not affected).

Erlang gives you the basic tools to do this, yes. But it would make 
every change we make to the broker really quite a lot more expensive.

Cheers, Simon

-- 
Simon MacMullen
RabbitMQ, VMware