When hosted clusters are delayed in deleting, their dedicated request serving nodes may have already been removed, but the configmap indicating that the node pair label is in use remains. Placeholder pods are currently getting scheduled on new nodes that have these pair labels. When the scheduler tries to use these new nodes, it says it can't because there is a configmap associating the pair label with a cluster that is in the process of deleting.

    4.19

    sometimes

    1. In a size tagging dedicated request serving architecture, create hosted cluster(s).
    2. Place an arbitrary finalizer on the hosted cluster(s) so it cannot be deleted.
    3. Delete the hosted clusters
    4. Look at placeholder pods in hypershift-request-serving-node-placeholders      

some placeholder pods are scheduled on nodes that correspond to fleet manager pairs taken up by the deleting clusters    

    no placeholder pods are scheduled on nodes that correspond to hosted clusters.