Feature Request:

We request the implementation of a key features to simplify CPU management and enable reliable Live Migration in heterogeneous OpenShift Virtualization clusters.

Use Case:

We need a method to enable an automated Discovery and recommendation of the Lowest Common Denominator (CPU) in OpenShift Virtualization, especially in General Purpose Clusters with different CPU's. 

The system must scan all KubeVirt-enabled Worker Nodes, calculate the safest Lowest Common Denominator CPU Model (i.e., the most compatible QEMU model all nodes can support), and provide this information as an actionable recommendation.

• Output we expect: The Lowest Common Denominator model (e.g., Haswell-noTSX for a v3/v4 mix) should be displayed in the status of the HyperConverged CR or a dedicated cluster resource / host resource.
• Actionable UI: The OpenShift Console should display a prompt/recommendation when heterogeneous CPUs are detected, guiding the administrator to set the spec.defaultCPUModel to the suggested Lowest Common Denominator model value in the UI so that we can migrate virtual machine between different CPU Architectures in OCP+V

Existing Work:

The administrator must manually determine the oldest architecture (e.g., Haswell, and the right version) and set the global default to this model via the HCO manifest.

Means the current process requires manual identification of CPU architectures. With commands like "oc describe "worker-node" we have, manually to figure out which cpu-models & features are support in this very host (dozens of labels). After doing this with ALL worker nodes we have to compare everything with a public Intel/Wikipedia list (Like: https://en.wikipedia.org/wiki/List_of_Intel_CPU_microarchitectures) to match the appropriate CPU Architecture in our "oc...." output.

This is really error prone and a quite cumbersome process. With VMWARE e.g., you have that straight in the UI already during the installation - which is important since most clusters are growing, and we constantly have to manually adapt and adjust these settings it we want to live migrate VM's between worker nodes.

Hint from j.thadden@redhat.com:

What should be there, at least, from the very beginning is the "Node Feature Discovery Operator" - it should be installed per default when we install the "HCO" (HyperConvergedOperator).