I ran the following scenario

• br-ex defined using nmstate
• cluster configured as dual-stack
• nodes configured at boot as dual-stack
• machineconfig with nmstate configuration as ipv4-only

As a result, nodeip-configuration detected 2 IP addresses from each stack respectively. It passed it down to kubelet. At the moment when kubelet started, nmstate configuration was already applied so system was single stack.

I would expect something to fail in an easy to debug way. In reality, kubelet was stuck but not failed. It was pretty tricky to discover what happened in reality.

As an admin installing a cluster I would like to see a clear error message somewhere.