[Pacemaker] advanced clone operation serialization and requirements

[David Vossel]

----- Original Message -----
> From: "Brad Jorgensen" <bjorgensen at sydcon.com>
> To: pacemaker at clusterlabs.org
> Sent: Tuesday, August 20, 2013 12:04:30 PM
> Subject: [Pacemaker] advanced clone operation serialization and requirements
> 
> I'm using pacemaker (version 1.1.9-1512.el6-2a917dd with the cman stack) and
> I want to add my MySQL cluster (actually MariaDB galera cluster) as a
> managed resource.
> 
> The problem I have is that the MySQL nodes must connect to an already
> established cluster.  This means that at least one node must be up in order
> for others to join.  The first node must bootstrap the cluster so it creates
> a new cluster to which other nodes connect.  If multiple servers bootstrap a
> cluster, they create their own cluster, which creates split brains.
> 
> I'd like pacemaker to block the resource on a monitor fail if no other
> clones are running but restart if at least one other clone is running.

If the monitor fails, there is only a single action that can be associated with that failure, and there is no way to express some sort of conditional logic that determines what that action is. (see http://clusterlabs.org/doc/en-US/Pacemaker/1.1-pcs/html-single/Pacemaker_Explained/index.html#_monitoring_resources_for_failure)

This probably isn't exactly what you're looking for, but what if you created a single MySQL instance that wasn't a clone, and left the other MySQL instances as a clone.  You could make some sort of ordering constraint that wouldn't allow the clone instances to start until single MySQl instance exists.

-- Vossel

> It
> would be even better if pacemaker could bootstrap one server if all of the
> clones are down.
> 
> Writing this out, I'm realizing how different this is from normal pacemaker
> functionality and I'm less sure that there is an easy solution.  I am going
> to look into modifying the ocf scripts to implement this functionality, too.
>  Does anyone know if what I'm looking for is possible?
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org
>