[rabbitmq-discuss] Clustering

[Jason J. W. Williams]

Hi Ben,

While replication would be helpful, an alternate stop gap would be the
ability to use the message queue log off the dead node and populate
the messages back into the cluster on command. More often than not, we
can stand a bit of a delay in the processing of the messages in the
event of the loss of a node (while we get the node back up). However,
if it looks like the node is going to be down for awhile (and
providing the message log is on some sort of network storage like NFS
or iSCSI), we may need to get the messages re-populated back into the
cluster while the dead node gets rebuilt.

Thank you again!

-J

On Mon, Jan 12, 2009 at 2:53 PM, Ben Hood <0x6e6562 at gmail.com> wrote:
> Jason,
>
> On Mon, Jan 12, 2009 at 9:01 PM, Jason J. W. Williams
> <jasonjwwilliams at gmail.com> wrote:
>> It appears that on a fail-over the contents of queues are not
>> available on the take-over server, but the queues and exchanges
>> themselves are available. Is this accurate? Thank you in advance.
>
> Yes, this is true. Broker entities (exchanges, queues and bindings)
> are maintained in a coherent fashion accross a cluster, but messages
> are not. This is due to the cost of replicating messages - it would be
> possible to do this, but it is quite expensive and not that many
> people really need it (yet). Having said this, it is a subject of
> investigation and may appear at some time in the future. Read the
> manual for more details: http://www.rabbitmq.com/clustering.html
>
> Ben
>