[rabbitmq-discuss] about durable queues in a cluster

[Steve]

于 2013-2-27 17:21, Tim Watson 写道:
> The disk/ram node distinction refers to where definitions (queue, exchange, etc) are stored. Messages are handled separately and are always stored on disk if they're marked persistent (and sometimes even if they're not, e.g., if memory is running out). You should note however, that just because you have a cluster doesn't mean that messages will be present on all nodes! If the nodes are clustered then the definition will exist on all nodes but the messages will only end up on the node where the queue was declared and is running. If you want messages to be mirrored across nodes then you'll need to make it a mirrored queue by setting the ha policy to ensure messages are replicated across nodes in the cluster. HTH.


Thanks. That make things clear.