[Pacemaker] "Follow and then push off"

[Andrew Beekhof]

On Wed, Oct 28, 2009 at 12:26 PM, Lars Marowsky-Bree <lmb at suse.de> wrote:
> Hi,
>
> sitting in the Open-HA tutorial, they implemented their support for
> replicated resources in an interesting fashion; not via a complex m/s
> resource type, but they mapped it to dependencies.
>
> They have a distinction between weak (optional) and strong collocation
> dependencies (similar to score="0" and score="infinity" in our world).
>
> Imagine we have two resources, DB_Active and DB_Replica. DB_Active is
> assigned higher priority than DB_Replica. One would want to have a)
> DB_Active and DB_Replica to be running on different nodes, of course;
> and b) if DB_Active fails, it should move to the node where DB_Replica
> was running.
>
> To achieve a), they create a strong negative collocation dependency from
> DB_Replica to DB_Active (implying that DB_Replica must not run on the
> same node as DB_Active).
>
> For b), they create a weak positive collocation dependency from
> DB_Active to DB_Replica (implying that DB_Active should be placed on the
> node where DB_Replica is running, if it exists).
>
>
> In case of fail-over, this causes DB_Active to be moved to the node
> where DB_Replica is still running, and then DB_Replica is stopped
> because it mustn't be running there. Basically a "follow and push off"
> strategy.
>
> I sort of like the approach, and wonder if we can do something similar?

Its interesting, but sounds like a nightmare to configure.
And how does it scale?  (ie. clone-max > 2 and/or master-max > 1)