This is a tested example.
Not all clusters will have the metrics API, but this will function for ones that do.
The deployment uses the bitnami/kubectl image.
It runs "get node metrics" in a shell while loop that sleeps.
The Deployment's Pod-template uses a ServiceAccount that is ClusterRoleBound to a ClusterRole permitting the list and get verbs for the Node kind in the metrics.k8s.io API group.
Here's how to reproduce: