k0smotron as Cluster API provider

k0smotron can act as a Cluster API provider for both control planes and for Machine bootstrapping.

Note: Cluster API providers in k0smotron are experimental still. We're iterating fast on these but we'd definitely value your feedback how it behaves with various infrastructure providers.

Control Plane provider

When k0smotron acts as a control plane provider it will create and manage the cluster control plane within the management cluster, just as in the standalone case.

Bootstrap provider

k0smotron can also act as a bootstrap provider for worker nodes you want to manage via Cluster API. When k0smotron detects a new node that needs to be added to the cluster it will automatically create a new join token needed for the node and creates the provisioning cloud-init script for the node. Once Cluster API controllers sees the node initialization script in place (in a secret) the infrastructure provider will create the needed resources (usually VMs in cloud provider infrastructure) with the k0smotron created cloud-init script.

Cluster autoscaling

Cluster Autoscaler works with ClusterAPI. You need to deploy an "instance" of autoscaler per child cluster in order for it to work properly. If you deploy autoscaler via Helm, here's some values to look out:

Value	Why?
autodiscovery.clusterName	Select the child cluster you want to autoscale. E.g. my-cluster
cloudProvider	Set to clusterapi to make autoscaler work with ClusterAPI
clusterAPIKubeconfigSecret	Set to the kubeconfig secret created by CAPI. E.g. my-cluster-kubeconfig
clusterAPIMode	Set to kubeconfig-incluster. Essentially this tells that the child cluster API is accessed with the kubeconfig from the secret and management cluster via incluster kubeconfig

RBAC finetuning needed with Helm deployed autoscaler

The Helm chart does not take into account the need for autoscaler to access the implementation specific resources in infrastructure.cluster.x-k8s.io group. To fix that you need to modify* the ClusterRole to include e.g.

- verbs:
   - get
   - list
   - update
   - watch
 apiGroups:
   - infrastructure.cluster.x-k8s.io
 resources:
   - '*' # You can of course limit this to your specific infrastructure types only, e.g. `AWSMachineDeployment` etc.

*) Happy to get feedback whether there's a better workaround for this.