When HighNodeUtilization strategy is enabled for descheduler, if all the worker nodes are underutilized the strategy does not work fine, meaning pods tries to evict to master nodes and the pods could not be scheduled on it as the taints does not match.
    

    [knarra@knarra-thinkpadx1carbon7th openshift-tests-private]$ oc get csv -n openshift-kube-descheduler-operator
NAME                                    DISPLAY                     VERSION   REPLACES                                PHASE
clusterkubedescheduleroperator.v5.1.0   Kube Descheduler Operator   5.1.0     clusterkubedescheduleroperator.v5.0.1   Succeeded

    

     Always
    

    1.  Install OCP latest version
    2.  Install KDO 5.1.0 Version
    3.  Enabled HighNodeUtilization profile and adjust the devHighNodeUtilizationThreshold in such a way that all workers will become underUtilized nodes.
     4. Try creating pods with ownerRef (like replicaSets or deployments)
    

    Pod evictions gets skipped and HighNodeUtilization strategy does not work well
    

    HighNodeUtilization strategy should work well
    

    Upstream issue: https://github.com/kubernetes-sigs/descheduler/issues/725