[rabbitmq-discuss] rabbitmq cluster failover for persistent queues

[Arvind Jayaprakash]

On Jul 25, charles woerner wrote:
>Hello,
>
>Although this is my first post to the rabbitmq-discuss list, I've enjoyed
>reading the archives over the last few days.  We're evaluating rabbitmq for
>a system at work.  One apparent limitation that we've run into is that while
>exchange information (which node a queue lives on, etc.) is replicated each
>node itself seems to be a single point of failure in that if it goes down
>then a message whose routing key only matches bindings to queues on that
>node will be dropped.  That is, from what I understand it appears that
>although the contents of persistent queues are /recoverable/ in the event of
>a disaster I don't see a built-in way of having another node in the cluster
>"take over" a queue which previously lived on a node which has just gone
>down.
>
>I'd like to know whether a) I've missed something in my interpretation of
>how rabbitmq server handles persistent queues in a cluster and b) if this is
>indeed the current behavior what other rabbitmq users have done to provide
>fault tolerance.

Your description is correct and I had the same question myself a few
months ago. Many thanks to Alexis for providing this link:
http://skillsmatter.com/podcast/erlang/rabbitmq-internal-architecture-tony-garnock-jones