[rabbitmq-discuss] Queue Failover and restore

[Michael Klishin]

On 25 March 2014 at 11:33:27, Arun Rao (arunrao.seattle at gmail.com) wrote:
> > These 300 Gigs are split as 75 gigs per node and process messages  
> accordingly. If there is a node failure, this increases to 100  
> Gigs per node. This is a substantial increase during peak periods.  

Understandable.

> My only option now is to stop all the nodes from taking traffic,  
> delete the queues and apply HA again.

Have you considered sticking a proxy in front of the cluster?
It may be quite effective (and HAproxy supports several load balancing
modes, including “least connected” and client IP hashing).

This would still require clients to reconnect after node recovery
but sounds a lot less intrusive than shutting down the entire system. 
--  
MK  

Software Engineer, Pivotal/RabbitMQ