We should get metrics from the cluster. The Prometheus operator is probably the best way (with Grafana).