[rabbitmq-discuss] Stability/Reliability of a cluster with mirrored queue

[Mirosław Nagaś]

Hi,

I would like to know your opinion about stability/reliability of a 
cluster with mirrored queues. I'm considering to use it and even tried 
with 3.2.0. I guess I don't have to point out how huge disappointment 
was it. And what's worse, if you skim recent releases notes, it's really 
hard to find a release without some bug fixes related to clustering 
and/or HA.

What I'm asking is, is a cluster setup with mirrored queues stable and 
reliable and past bugs/problems appeared really rarely in anomalous use 
cases/conditions or the reality is that it has some 
stability/reliability issues that affects every robust use cases at some 
point?

We have been running a single node without interruption on production 
(cloud) for about 1,5 year. Having in mind all past bugs, what I'm 
worried about is that, in a year from now I will come to a conclusion 
that because of a cluster with mirrored queue my overall reliability 
and/or stability was much worse than eg. just two separate nodes behind 
LB (maintainability is not a big deal here; we use persistence and can 
afford restoring messages from a failed node manually in case of a cloud 
machine failure).

Maybe someone can share his experience from production run?


/mn