Best approach to expose variables to KCPT?

[pierrecdn]

Hello there,

Trying to wrap my head around CAPI in a kamaji/kubevirt context.
I'd like to define minimalist clusters, such as what is shown in https://cluster-api.sigs.k8s.io/tasks/experimental-features/cluster-class/write-clusterclass#basic-clusterclass.
Typically it seems that duplicating KamajiControlPlaneTemplate doesn't make much sense. Still, for specific reasons, I have at least the need to propagate the cluster name. For simplicity let's consider that I need to annotate the Control Plane service with the cluster name:

kind: KamajiControlPlaneTemplate
apiVersion: controlplane.cluster.x-k8s.io/v1alpha1
metadata:
  name: generic-control-plane
(...)
spec:
  template:
    spec:
      network:
        serviceAnnotations:
          mynamespace.tld/cluster-name: <CLUSTER_NAME>

I'm reading that in the ClusterClass context, using {{ .cluster.name }} is supported, as well as patches.
But is it the case within a KamajiControlPlaneTemplate? My early experiments doesn't seem that successful.

Thanks,

[prometherion]

Hey 👋🏻 @bsctl just released a new Kamaji documentation regarding ClusterClass support.

From the CAPI perspective, a template is just a template; if it's a Kubeadm or a Kamaji one, it doesn't matter.

At first sight, it seems you're facing an issue with CAPI itself, rather than Kamaji: templates are supported only for JSON patches and if combined with valueFrom.template structs.

[pierrecdn]

Hey 👋,
Many thanks. I figured out that I missed one or two point of CAPI ClusterClass architecture regarding patches and variables 😄 .
In the end it works fine, and your documentation updates made it quite clear for not-yet-advanced users like me.