In the "request-serving" deployment model for HCPs, the request-serving nodes are being memory starved by the k8s API server. This has an observed impact on limiting the number of nodes a guest HCP cluster can provision, especially during upgrade events.

This is a spike card to investing setting the API Priority and Fairness [1] configuration, and exactly what configuration would be necessary to set.

[1] https://kubernetes.io/docs/concepts/cluster-administration/flow-control/