Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-54414

descheduler namespace filtering not creating exclusions

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Major Major
    • None
    • 4.17.z, 4.18
    • descheduler
    • Quality / Stability / Reliability
    • False
    • Hide

      None

      Show
      None
    • 2
    • None
    • None
    • None
    • None
    • Workloads Sprint 269, Workloads Sprint 270
    • 2
    • In Progress
    • Bug Fix
    • Hide
      Prior to KDO 5.1.2, it was possible to disable the protection that prevented eviction of pods in the openshift-* namespaces by modifying the list of included namespaces for the LowNodeUtilization descheduler plugin. Starting with the 5.1.2 release, this protection has been reinstated.
      Show
      Prior to KDO 5.1.2, it was possible to disable the protection that prevented eviction of pods in the openshift-* namespaces by modifying the list of included namespaces for the LowNodeUtilization descheduler plugin. Starting with the 5.1.2 release, this protection has been reinstated.
    • None
    • None
    • None
    • None

      I am attempting to configure descheduler for virtual machines only by configuring namespace filtering (full KubeDescheduler definition attached).

      namespaces:
        included:
          - virt-vms

      I'm observing that pods from openshift-* namespaces are also being evicted despite the requested filter. Expecting only workloads from included namespace to be evicted. In addition, CRD suggests that all openshift-* namespaces are part of default exclusion.

      Namespaces overrides included and excluded namespaces while keeping the default exclusion of all openshift-*, kube-system and hypershift namespaces.

      Namespace exclusion is not being rendered in the cluster configmap.

      Attached are KubeDescheduler yaml, rendered configmap and logs from descheduler pod showing evictions.

      In addition to this, when VM evictions do take place, there is no audit record in the logs. It shows that 1 pod can be evicted but does not report which workload is moved.

       

        1. descheduler-84f686c4c5-74q65-openshift-descheduler.log
          63 kB
          Patrick Harrison
        2. descheduler.yaml
          0.4 kB
          Patrick Harrison
        3. cluster-configmap
          2 kB
          Patrick Harrison

              jchaloup@redhat.com Jan Chaloupka
              pharriso@redhat.com Patrick Harrison
              None
              None
              None
              None
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

                Created:
                Updated: