Hi there

AFAIK Machines are considered immutable.
So whenever a spec in the MachineDeployment changes, a new MachineSet gets created which creates new Machines and therefore new Kubernetes Nodes.

What are your opinions on excluding Node Labels & Annotations?
So whenever these change in the MachineDeployment spec, the machine-controller updates those on the existing Kubernetes Nodes instead of recreating the Machines via MachineSet.

In our environment, especially Node Labels can change quite frequently.
We have multiple Nodepools per Cluster and use nodeSelectors and/or nodeAffinity to decide where Pods will be scheduled.

Also I would argue, that Labels & Annotations of the Kubernetes Node are metadata (as the parent key implies) and are not directly part of a Machine like CPU, RAM, SSH Keys etc.

What are your thoughts?
Would you be open to implement this or accept a contribution?
Could also be made, opt-in..