[rabbitmq-discuss] Rolling upgrade of a cluster

[Matthew Sackman]

On Tue, Mar 29, 2011 at 03:15:37PM -0600, Jason J. W. Williams wrote:
> That's good to know. I suppose the amount of downtime needed for the
> automatic upgrade to complete will vary by the number of queues and metadata
> to be migrated?

Depends on the nature of the upgrade steps to be performed. Worst case
is linear on the number of msgs on disk, but upgrades like that only
happen when we change the on-disk format, which doesn't happen very
often.

Matthew