[rabbitmq-discuss] suggestions for failover

[Shane]

Am I correct in that anything connected to the broker needs to
reconnect (via new hostname) after the failover takes place? So the
sequence needs to be something like the following:

Producers, Consumers connected to Broker 1
Failover occurs: Broker 1 goes down, Broker 2 comes up
Producers, Consumers reconnect to Broker 2

In general, it would be nice to have documentation that outlines what
needs to occur in a failover scenario as opposed to outlining steps of
a specific implementation (the Pacemaker guide).

Thanks,
Shane


On Sep 28, 10:24 am, David Wragg <da... at rabbitmq.com> wrote:
> Hi Shane,
>
> Shane <sh... at digitalsanctum.com> writes:
> > I'd like suggestions on automating failover of brokers and the
> > consumers that are connected to them. The failover doesn't have to be
> > active/active and may take up to a half hour. Ideally, I'd like to
> > have at least two brokers located on different hosts. The hosts will
> > have access to the same file system (NFS). The number of messages will
> > be relatively small (<5K/day) and the number of queues no more than 10
> > so one broker running at a time should be sufficient.
>
> Note that while RabbitMQ persistence is intended to work on top of NFS,
> we haven't done performance testing of that configuration.  With the
> message rates you mention, it seems unlikely that you'll run into
> problems in that regards, but it's something to be aware of.
>
> > I've looked at the clustering guide and it's purpose seems to be for
> > scaling but not failover.
>
> That's correct.  What we have traditionally termed clustering is
> intended for scaling, not availability.
>
> > I've also looked at the Pacemaker guide but it seems to be an involved
> > and more complicated setup than what I need.
>
> > What steps are others doing for failover? What are my options beyond
> > the Pacemaker guide?
>
> The pacemaker guide is intended to be illustrative, as well as giving a
> concrete guide on one way to implement failover with RabbitMQ.  There
> are alternative technologies, both open source and proprietary, that can
> be used instead of pacemaker, drbd, etc.  We'd welcome reports from
> users that have implemented failover using other products.
>
> If you have configured rabbit to store its data on NFS, then you won't
> need the drbd component.  And if you already have appropriate monitoring
> and alerting arrangements in place, then you might well be able to
> achieve a 30 minute failover time without using pacemaker or some other
> cluster resource manager.  Even a manual failover process might be
> acceptable in such a context.
>
> David
>
> --
> David Wragg
> Staff Engineer, RabbitMQ
> SpringSource, a division of VMware
> _______________________________________________
> rabbitmq-discuss mailing list
> rabbitmq-disc... at lists.rabbitmq.comhttps://lists.rabbitmq.com/cgi-bin/mailman/listinfo/rabbitmq-discuss