Improve config for multi node scenario

[fabriziopandini]

as per #117 (comment), one of the step for implementing multi node support is to improve Kind config in order to support the concept of topology: how many control-plane nodes should be in the cluster? how many worker nodes? should I add an external-load balancer? add an external-etcd?

A first high level idea for the new config design is something in the middle between current Kind config and cluster api. This imply the config will be composed by a combination of following objects:

• a Cluster object for "cluster wide settings"
• a Nodeobject (or Container/Machine to avoid confusion with Kubernets Nodeobject) for "node specific settings"
• eventually a NodeSet (or ContainerSet/MachineSet) object for defining groups of machines starting from a template

The new config api should be ready for supporting following use cases:

• different CRI/different node-image (and mixed clusters)
• version skew (between control-plane and kubelet or between nodes)
• volume mounts
• kubeadm config patching (should apply only to the bootstrap control-plane node?)
• LifecycleHooks (in which forms, considering different types of nodes?)
• Adding/Removing nodes to the cluster
• ...

Opinions about the proposed api design?
Preferences on object names?
Are there other use cases to be supported (in the first release)? Are there use cases not necessary (in the first release)?

/assign

[fabriziopandini]

/lifecycle active