As I had reported over a year ago, up to now it was not possible to use Pacemaker to properly manage resources that both resolve dependencies of other resources and can be migrated live to another node of your cluster. Pacemaker would simply stop the resource(s) that depend on the resource to be migrated, despite the fact that no interruption was required at all, technically.
But now there’s been a huge step forward, without warning, advance notice, nor threat of force, so to speak: As has been reported on the according ticket from clusterlab’s Bugzilla, David Vossel has checked in the according fixes to the source tree.
I have yet to pull and test the changes myself, but have no reason to believe that they will not provide the long-awaited solution. Once this hits the major enterprise distributions, this in my eyes will significantly boost the advantage of Pacemaker in clustered Xen/KVM environments!