fix: Ensure ClusterAutoscaler can write status CM in workload cluster (#864)

Previously workload ClusterAutoscaler deployments were configured to
write the status configmap to the same namesapce as the workload cluster
is in on the management cluster. However, ClusterAutoscaler actually
writes this configmap and acquires a lock in the specified namespace in
the workload cluster. If that namespace does not exist then CA will fail
to run correctly, repeatedly throwing the error `error initially
creating leader
election record: namespaces "my-new-ns" not found`.

This commit updates the target namespace to use the default
`kube-system` which always exists on all clusters. This is the CA
default and as such is completely safe to use.

Author: jimmidyson

charts/cluster-api-runtime-extensions-nutanix/templates/cluster-autoscaler/manifests/cluster-autoscaler-configmap.yaml

@@ -168,7 +168,7 @@ data:
           - command:
             - ./cluster-autoscaler
             - --cloud-provider=clusterapi
-            - --namespace='{{ `{{ .Cluster.Namespace }}` }}'
+            - --namespace=kube-system
             - --node-group-auto-discovery=clusterapi:clusterName='{{ `{{ .Cluster.Name }}` }}',namespace='{{ `{{ .Cluster.Namespace }}` }}'
             - --kubeconfig=/cluster/kubeconfig
             - --clusterapi-cloud-config-authoritative

charts/cluster-api-runtime-extensions-nutanix/templates/cluster-autoscaler/manifests/helm-addon-installation.yaml

@@ -29,7 +29,6 @@ data:
       labels:
         - namespace: "{{ `{{ .Cluster.Namespace }}` }}"
 
-    clusterAPIConfigMapsNamespace: "{{ `{{ .Cluster.Namespace }}` }}"
     # For workload clusters it is not possible to use the in-cluster client.
     # To simplify the configuration, use the admin kubeconfig generated by CAPI for all clusters.
     clusterAPIMode: kubeconfig-incluster

hack/addons/kustomize/cluster-autoscaler/helm-values.yaml

@@ -25,7 +25,6 @@ autoDiscovery:
   labels:
     - namespace: tmpl-clusternamespace-tmpl
 
-clusterAPIConfigMapsNamespace: tmpl-clusternamespace-tmpl
 # For workload clusters it is not possible to use the in-cluster client.
 # To simplify the configuration, use the admin kubeconfig generated by CAPI for all clusters.
 clusterAPIMode: kubeconfig-incluster