In some instances the request serving node autoscaler helper fails to kick off a reconcile when there are pending pods.

    4.17

    sometimes

    1. Setup a mgmt cluster with size tagging
    2. Create a hosted cluster and get it scheduled
    3. Scale down the machinesets of the request serving nodes for the hosted cluster.
    4. Wait for the hosted cluster to recover

    A placeholder pod is created for the missing node of the hosted cluster, but does not cause a scale up of the corresponding machineset.

    The hosted cluster recovers by scaling up the corresponding machinesets.