Comment 4 for bug 1894173

Upstream Kubernetes doesn't really know how to handle "isolated" CPUs separately from "normal" CPUs. Because of this sysinv will reserve the "application-isolated" CPUs from use by Kubernetes when using the "static" CPU Manager policy.

If you set the kube-cpu-mgr-policy host label to "none" (via Horizon or the "sysinv" CLI) Kubernetes will use the default policy and should include the "application-isolated" CPUs as part of the "Allocatable" cpu count. You'll have to handle CPU affinity of containerized processes explicitly though to ensure that they don't interfere with each other.