[rabbitmq-discuss] Feature request - automatic recovery from network partitions in simple case
simon at rabbitmq.com
Thu Mar 14 10:36:27 GMT 2013
You'll be pleased to hear that we are already working on improving the
handling of partitions for 3.1, broadly along these lines. I'm not quite
sure how we would define the "active" node in the general case though.
Mind you, maybe you could just put something in the config to tell us.
On 14/03/13 02:16, Jesse Young wrote:
> I run a simple cluster of two RabbitMQ 3.0.x nodes with mirrored
> high-availability queues. Only one of the two nodes is actively used at
> any given time; the other node is a "hot spare" that can be ready in
> case the first node goes offline.
> Occasionally, my cloud hosting provider has network interruptions within
> their internal network. RabbitMQ then detects a network partition, and
> stops mirroring queues until I manually restart the nodes. In order to
> avoid losing all queued messages, I have to manually stop the spare
> node, then restart the active node, then start the spare node again.
> I understand the reasoning given
> in http://www.rabbitmq.com/partitions.html for why recovery doesn't
> happen automatically in the general case.
> However, in my simple case, only one of the nodes has any activity
> during the network partition, so there isn't any ambiguity about what
> should happen when the network starts working again: The "spare" node
> should simply restore its state from the "active" node without me having
> to manually restart them in the correct order.
> I don't know how feasible this request is, but it would be great if
> RabbitMQ's partition recovery could be slightly more intelligent so that
> having a spare RabbitMQ node doesn't require as much handholding.
> Apologies in advance if this has already been discussed or if this is
> the wrong forum. I'd also like to thank the developers for such a great
> rabbitmq-discuss mailing list
> rabbitmq-discuss at lists.rabbitmq.com
More information about the rabbitmq-discuss