CKS and the metrics-server
- CKS clusters created after July 7, 2025 do not include metrics-server; if your workloads need it, install it yourself as described on this page.
- Clusters created before July 7, 2025 include a CoreWeave-managed metrics-server in the
kube-systemnamespace, so you don’t need to install your own. See the July 7, 2025 release notes for details about this change.
kubectl top node and kubectl top pod and provides real-time metrics to the Horizontal Pod Autoscaler (HPA).
Metrics-server only affects kubectl top and HPA. Metrics shown in CoreWeave Grafana are collected through a separate pipeline and don’t depend on metrics-server. To collect custom application metrics, see Forward application metrics.
Prerequisites
Before you begin, make sure you have:- At least one CPU Node in your cluster to run the metrics-server Pods.
kubectlaccess to your CKS cluster with permission to create resources in thekube-systemnamespace. See Obtain access with a kubeconfig file.- Helm version 3 or later installed on your workstation.
Install metrics-server
1
Install the chart
Install the upstream metrics-server Helm chart into the Confirm the Deployment is ready:You should see
kube-system namespace:1/1 in the READY column:Example output
2
Verify the Metrics API
Query Node metrics:If metrics-server is working correctly, the command returns CPU and memory usage for each Node:Metrics can take a minute or two to appear after installation.
Example output