knative-serving-ingress ServiceMeshControlPlane has

    ownerReferences:
    - apiVersion: serving.knative.dev/v1alpha1
      blockOwnerDeletion: true
      controller: true
      kind: KnativeServing
      name: knative-serving
      uid: 222b1c2a-0610-11ea-bdbe-069014e085

which is invalid, as it doesn't exist in the same namespace.

This causes the SMCP and SMMR to be deleted by the k8s garbage collector whenever it runs.

This causes:

• Existing ksvc to stop working (Ingress cease to exist, so the routes starts returning 404s, even if the ksvc are still being reported as "ready")
• The istio-ingressgateway LoadBalancer is recreated when SMCP gets recreated, so its hostname may change
• Namespaces already added to SMMR will not be added there when the new one is generated.

To reproduce:

• Find the kube-controller-manager pod with the lease in openshift-kube-controller-manager
• kill it, e.g.: (oc delete pod doesn't work on those)
  • oc rsh kube-controller-manager-ip-10-0-137-180.us-west-1.compute.internal /bin/sh -c "kill 1"