Today, Rancher runs the cluster-api controllers in an embedded fashion, see: https://github.com/rancher/rancher/blob/8bee06f75426e17c2ef7bacf11a0554a4754efb2/pkg/provisioningv2/capi/capi.go for a reference. Unfortunately, this makes maintenance of cluster-api versions a nightmare. For example, we can't bump from v1.2 to v1.3 or v1.4 because of K8s and controller-runtime dependencies.

We should move to running the cluster-api controllers in a more traditional fashion, i.e. run the controllers in a pod managed by the cluster-api operator.

cc: @richardcase this is going to be important for you folks.