Redundant anti-colocation chains can cause resources to flip-flop in a Pacemaker cluster

Issue

• My cluster has two or more colocation constraint sets, each containing negative colocations for three or more resources. Every time the scheduler runs a new transition, resources move back and forth. For example:

Colocation Constraints:
  Resource Sets:
    set dummy1 dummy2 dummy3 setoptions score=-INFINITY
    set dummy1 dummy2 dummy3 setoptions score=-INFINITY

Mar 26 17:58:22 fastvm-rhel-8-0-24 pacemaker-controld[530541]: notice: State transition S_IDLE -> S_POLICY_ENGINE
Mar 26 17:58:22 fastvm-rhel-8-0-24 pacemaker-schedulerd[530540]: notice:  * Move       dummy1             (         node2 -> node1 )
Mar 26 17:58:22 fastvm-rhel-8-0-24 pacemaker-schedulerd[530540]: notice:  * Move       dummy3             (         node1 -> node2 )
...
Mar 26 18:13:22 fastvm-rhel-8-0-24 pacemaker-controld[530541]: notice: State transition S_IDLE -> S_POLICY_ENGINE
Mar 26 18:13:22 fastvm-rhel-8-0-24 pacemaker-schedulerd[530540]: notice:  * Move       dummy1             (         node1 -> node2 )
Mar 26 18:13:22 fastvm-rhel-8-0-24 pacemaker-schedulerd[530540]: notice:  * Move       dummy3             (         node2 -> node1 )