How to fix failed to get cpu utilization: unable to get metrics in Kubernetes

KubernetesINTERMEDIATEMEDIUM

The HPA (Horizontal Pod Autoscaler) error about unable to get CPU metrics indicates the Metrics Server is not installed, misconfigured, or pods lack resource requests. Without metrics, HPA cannot determine if pods need more replicas.

What this error means

HPA relies on the Metrics Server to collect CPU and memory metrics from the kubelet on each node. These metrics are then used to determine if pods need scaling. The "unable to get metrics" error means either: 1. Metrics Server is not installed in the cluster 2. Pods lack CPU resource requests (HPA uses requests to calculate percentages) 3. Metrics API is unreachable or misconfigured 4. Pods are not in Ready state (metrics only available for running pods) Without metrics, HPA cannot make scaling decisions and remains in "Waiting for metrics" state indefinitely.

How to fix "failed to get cpu utilization: unable to get metrics"

1Check if Metrics Server is installed

Verify Metrics Server is running:

bash

kubectl get deployment -n kube-system metrics-server
kubectl get pods -n kube-system | grep metrics-server

If not found, install it:

bash

kubectl apply -f https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml

Wait 30-60 seconds for the pod to start, then verify:

bash

kubectl get pods -n kube-system metrics-server -o wide
kubectl logs -n kube-system deployment/metrics-server

2Add resource requests to pod specifications

HPA requires resource requests to calculate CPU utilization. Update your deployment:

yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: myapp
spec:
  template:
    spec:
      containers:
      - name: app
        image: myapp:latest
        resources:
          requests:
            cpu: 100m        # Required for HPA
            memory: 128Mi    # Optional but recommended
          limits:
            cpu: 500m
            memory: 512Mi

Apply:

bash

kubectl apply -f deployment.yaml

Metrics should be available 15-60 seconds after pod startup.

3Verify pods are in Ready state

HPA only metrics for Ready pods. Check status:

bash

kubectl get pods
kubectl describe pod <pod-name>

If pods are not Ready, fix the underlying issues first (image pull errors, health check failures, resource constraints). Once pods are Running and Ready:

bash

kubectl top pods  # Should show CPU/memory now

4Test metrics API directly

Verify the Metrics API is working:

bash

kubectl get --raw /apis/metrics.k8s.io/v1beta1/nodes
kubectl get --raw /apis/metrics.k8s.io/v1beta1/namespaces/default/pods

If this returns valid JSON with metrics, the API is working. If it returns errors, check Metrics Server logs:

bash

kubectl logs -n kube-system deployment/metrics-server

Common errors:
- "Unauthorized" or "Forbidden": RBAC permissions issue
- "Connection refused": Metrics Server not running or API unavailable

5Check Metrics Server logs for errors

Get detailed error information:

bash

kubectl logs -n kube-system deployment/metrics-server -f
kubectl describe pod -n kube-system -l k8s-app=metrics-server

Common issues:
- "Unable to connect to Kubelet": Firewall blocking port 10250, or Kubelet not running
- "Forbidden": RBAC permissions missing for Metrics Server
- "CrashLoopBackOff": Pod startup failure (check earlier logs)

If Metrics Server is crashing, check available resources and restart it:

bash

kubectl rollout restart deployment -n kube-system metrics-server

6Verify Kubelet metrics endpoint is accessible

From the control plane, test kubelet metrics port:

bash

kubectl get nodes -o wide  # Get node IPs
curl -k --cert /var/run/kubernetes/client-admin.crt \
  --key /var/run/kubernetes/client-admin.key \
  https://<node-ip>:10250/metrics

If this fails with connection refused, kubelet isn't exposing metrics. Verify kubelet is running on the node:

bash

ssh <node-ip>
sudo systemctl status kubelet
netstat -tlnp | grep 10250

7Create a test HPA and verify it scales

Once metrics are available, test with a simple HPA:

bash

# Create a test deployment with metrics
kubectl create deployment hpa-test --image=nginx --replicas=1 -- sleep 3600
kubectl set resources deployment hpa-test --requests=cpu=100m

# Create HPA
kubectl autoscale deployment hpa-test --min=1 --max=3 --cpu-percent=50

# Check HPA status
kubectl get hpa -w
kubectl describe hpa hpa-test

Once metrics become available (15-60 seconds), HPA should show CPU percentage instead of "unknown".

8Wait for metrics stabilization

Metrics are not immediately available after pod creation:

bash

# Right after pod start:
kubectl top pod <pod-name>  # May return "no metrics"

# Wait 15-60 seconds:
watch kubectl top pod <pod-name>

Metrics become available once the kubelet has collected data from the container runtime. Be patient—metrics typically appear within 1-2 minutes for healthy pods.

9Verify RBAC permissions for Metrics Server (if using custom RBAC)

If Metrics Server cannot access Kubelet metrics, RBAC may be too strict. Verify:

bash

kubectl get clusterrole system:metrics-server -o yaml
kubectl get clusterrolebinding system:metrics-server -o yaml

Metrics Server needs permissions to:
- Get nodes and node metrics
- Access kubelet stats endpoints
- List pods in all namespaces

If using restrictive RBAC, ensure the Metrics Server ClusterRole includes these permissions.

How to fix failed to get cpu utilization: unable to get metrics in Kubernetes

What this error means

Typical symptoms

Common causes

How to fix "failed to get cpu utilization: unable to get metrics"

Advanced notes

Related errors

Official resources & further reading