What were you trying to do that didn't work?

Attribute based colocation allows reference to a node-attribute in a colocation constraint basically allowing to use the content of that node-attribute to be used instead of a node-name.

When a cluster is e.g. split over multiple sites having multiple nodes on each site one would put the site name into a node-attribute.
When an instance of a resource cloned over the whole cluster would now be promoted we should now be able to have all instances of another cloned resource on the same site to be promoted - given that promoted-max allows enough of the instances to be promoted.

Current behavior is roughly as it would be without the node-attribute given (Just the instance on the same node is being promoted).

What is the impact of this issue to you?

Behavior is needed for a customer-project.

With the introduction of a dummy primitive the desired behavior can be achieved.
Dummy primitive follows promoted role of the first clone while promoted role of the 2nd clone is colocated (with node-attribute holding the site-name) with the dummy primitive.

This is of course a clumsy workaround and feature is expected to work as described.

Please provide the package NVR for which the bug is seen:

pacemaker-3.0.0-1.el10.1

How reproducible is this bug?:

100%

Steps to reproduce

1. Use 'crm_simulate Sx colocattr-role.xml' to make scheduler calculate a transition with the cib attached.
2.  
3.  

Expected results

The cib has a status-section where stubborn-clone is promoted on nodez01n01 while firewall-clone is promoted on nodez01n01 & nodez01n02 which do reside on the same site.

Scheduler should be content with the current situation.

Actual results

Scheduler would demote firewall-clone on nodez01n02 which is roughly the behavior without attribute-based-colocation.