> I’d like to conclude by mentioning that RabbitMQ does documentation right. Most of the distributed systems I research say nothing about failure modes–but Rabbit’s docs have an entire section devoted to partition tolerance, reliable delivery, and truly comprehensive descriptions of the various failure modes in the API and distribution layer. I wish more distributed systems shared Rabbit’s integrity and attention to detail.
I would love to see someone build out a system using RabbitMQ's suggestions and see how resilient it is.
I was running a cluster of 3 RabbitMQ servers on 3 servers, all in the same data center, but different racks The page https://www.rabbitmq.com/clustering.html highlights that "RabbitMQ clustering does not tolerate network partitions well". Very true, I had this problem too often. The error recovering is manual and a nightmare. I didn't try the shovel or federation plugins as suggested. Because my needs for queuing were very simple, I decided to build a queue on top of Redis (cluster of 3 servers with redis-sentinel).
