[ClusterLabs] migrating VirtualDomain starts migration of second VirtualDomain in reverse direction

[Ken Gaillot]

On Thu, 2018-09-20 at 17:34 +0200, Lentes, Bernd wrote:
> Hi,
> 
> i have a two-node cluster with several VirtualDomain resources.
> Scenario:
> Two VirtualDomains, running on different nodes.
> Migrating one VirtualDomain from node 1 to node 2 migrates the other
> VirtualDomain from node 2 to node 1.
> 
> These are the scores AFTER the migration:
> ...
> vm_mausdb               0         ha-idg-1   200        0
> vm_mausdb               INFINITY  ha-idg-2   200        0
> vm_sim                  0         ha-idg-1   200        0
> vm_sim                  0         ha-idg-2   200        0
> ...
> 
> The cluster creates a constraint for vm_mausdb that it should run on
> ha-idg-2. That's fine.
> The strange thing for me is that nothing concerning vm_sim changed
> during the migration. No change neither in the stickiness nor in the
> scores.
> Nevertheless vm_sim is migrated.
> I didn't find any explaining in the log from the DC.
> I know that the cluster tries to distribute the resources evenly. But
> i thought the stickiness (i set a default_stickiness of 200) prevents
> resources from migrating so easily.
> Or is the value too low ?
> 
> Bernd

I think you meant default-resource-stickiness ... and even that's
deprecated in 1.1 and gone in 2.0. :-)

The proper way is to set resource-stickiness in the rsc_defaults
section (however that's done using your tools of choice).
-- 
Ken Gaillot <kgaillot at redhat.com>