[Pacemaker] migrate_from/migrate_to for Stateful RAs

[Andrew Beekhof]

On Tue, Sep 1, 2009 at 8:23 AM, Florian Haas<florian.haas at linbit.com> wrote:
> Andrew,
>
> does the CRM/PE allow for resource migration with
> migrate_to/migrate_from for Master/Slave RAs?

I think it does, but that wont help you here because...

> This would imply that at
> some point (during migration) multiple instances would be in the Master
> role, and master-max would be ignored (or implicitly incremented by 1)
> but while *not* migrating, master-max would be enforced.
>
> And in addition to that, what are the implications of initiating
> migration for a set of resources joined via colocation constraints, all
> the resources of which implement migrate_from and migrate_to and are set
> to "allow-migrate=1"?
>
> To sum up, would the CRM/PE allow something like this:
>
> - migration-capable drbd resource
> - VirtualDomain resource (already migration-capable)

only one layer of the stack is ever able to migrate.

a resource can only migrate if its location and ordering pre-requisits
are satisfied on the source and target nodes.
clearly this can't be true for drbd _and_ something running on top of
it (since drbd wouldn't be on the target yet).

this would likely change once we implement freeze/thaw... but thats
still a little way off.


> - colocation constraint making VirtualDomain dependent on drbd's Master role
> - "crm resource migrate <VirtualDomain resource>"
> - now do drbd migrate_to first, then VirtualDomain
> migrate_to/migrate_from, and *now* drbd migrate_from?
>
> I realize this probably sounds like I'm totally off my rocker, but would
> this be feasible in any way?
>
> Cheers,
> Florian
>
>
>
>
> _______________________________________________
> Pacemaker mailing list
> Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
>