Metrics Server

Metrics Server collects container resource metrics from kubelets and exposes the Kubernetes Metrics API for HPA, VPA, and kubectl top workflows.

Key Features

• Official registry.k8s.io/metrics-server/metrics-server image pinned to v0.8.1
• Metrics API APIService, ServiceAccount, ClusterRole, ClusterRoleBinding, and delegated authentication RoleBinding
• Typed kubelet flag modeling instead of only raw args
• Optional k3d and kind-compatible kubelet insecure TLS mode for local clusters
• Optional HA replicas, PDB, topology spread, and hostNetwork mode
• Dual-stack Service fields, NetworkPolicy, ServiceMonitor, and readiness Helm test

Installation

helm repo add helmforge https://repo.helmforge.dev
helm repo update
helm install metrics-server helmforge/metrics-server --namespace kube-system

helm install metrics-server oci://ghcr.io/helmforgedev/helm/metrics-server --namespace kube-system

Local Clusters

Many local clusters expose kubelet certificates without the address SANs Metrics Server validates. For k3d or kind smoke tests:

metricsServer:
  kubelet:
    insecureTLS: true

Do not enable insecure kubelet TLS in production unless your platform explicitly accepts that risk.

Operations

Metrics Server 0.8.x targets Kubernetes 1.31+. For HA, use at least two replicas and configure aggregator routing on the API server where supported.

Architecture

Metrics Server is a cluster add-on. The chart renders the Deployment, Service, RBAC, and optional APIService that backs metrics.k8s.io/v1beta1. The pod scrapes kubelets, aggregates CPU and memory samples, and serves them through the Kubernetes aggregation layer for kubectl top, HPA, and VPA integrations.

Traffic flow:

1. Metrics Server connects to kubelets on the configured secure port.
2. The Kubernetes API aggregation layer proxies metrics.k8s.io requests to the Metrics Server Service.
3. Controllers such as HPA read pod and node metrics from the aggregated API.
4. Optional ServiceMonitor resources scrape Metrics Server’s own process metrics.

Production Values

Use two replicas, a PDB, and topology spread in production clusters:

replicaCount: 2

pdb:
  enabled: true
  maxUnavailable: 1
  unhealthyPodEvictionPolicy: AlwaysAllow

topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: kubernetes.io/hostname
    whenUnsatisfiable: ScheduleAnyway
    labelSelector:
      matchLabels:
        app.kubernetes.io/name: metrics-server

For local k3d or self-signed kubelet serving certificates, use the dedicated local override:

metricsServer:
  kubelet:
    insecureTLS: true

Do not carry insecureTLS: true into production unless the platform team has explicitly accepted the TLS validation tradeoff.

APIService And RBAC

Keep apiService.create=true when the cluster does not already provide Metrics Server. The APIService must be Available=True before HPA and kubectl top can work:

kubectl get apiservice v1beta1.metrics.k8s.io
kubectl describe apiservice v1beta1.metrics.k8s.io

The chart can also run without creating RBAC or the APIService for distributions that already manage those resources, but in that mode ownership must be clear to avoid reconciliation conflicts. When the cluster already provides v1beta1.metrics.k8s.io or system:aggregated-metrics-reader, the chart skips those cluster-scoped resources instead of attempting to adopt platform-owned objects into the Helm release.

Networking

Metrics Server usually needs egress from the pod to every node kubelet. If NetworkPolicy enforcement is enabled, allow node/kubelet traffic and DNS:

networkPolicy:
  enabled: true
  egress:
    enabled: true
    allowDNS: true
    extraEgress: []

Host networking is opt-in for platforms where pod-to-kubelet routing requires it:

hostNetwork:
  enabled: true
dnsPolicy: ClusterFirstWithHostNet

Observability

Enable the ServiceMonitor when Prometheus Operator is installed:

serviceMonitor:
  enabled: true
  labels:
    release: prometheus

Use Metrics Server process metrics to detect scrape failures, API aggregation errors, and high request latency. Use the Metrics API itself for autoscaling signals rather than scraping application metrics from Metrics Server.

Validation

After deployment:

helm test metrics-server -n kube-system
kubectl get pods -n kube-system -l app.kubernetes.io/name=metrics-server
kubectl top nodes
kubectl top pods -A
kubectl get events -n kube-system --sort-by=.lastTimestamp

If kubectl top returns no data immediately after install, wait for at least one metric resolution interval and check the APIService condition.

Common Issues

Symptom	Likely Cause	Fix
kubectl top returns Metrics API not available	APIService is unavailable	Describe v1beta1.metrics.k8s.io and check Service endpoints.
Logs show kubelet certificate errors	Kubelet serving certs are not trusted	Configure platform CA trust or use insecureTLS only for local clusters.
HPA never scales	Metrics API has no pod samples	Check pod readiness, Metrics Server logs, and kubelet connectivity.
Pods cannot scrape kubelets	NetworkPolicy blocks node egress	Allow kubelet ports and DNS in egress policy.

Values

Parameter	Default	Description
image.repository	registry.k8s.io/metrics-server/metrics-server	Official image repository.
image.tag	v0.8.1	Metrics Server image tag.
replicaCount	1	Deployment replicas.
apiService.create	true	Create v1beta1.metrics.k8s.io APIService.
rbac.create	true	Create required RBAC resources.
metricsServer.metricResolution	15s	Metrics collection interval.
metricsServer.kubelet.insecureTLS	false	Disable kubelet TLS verification for local clusters.
hostNetwork.enabled	false	Use host networking.
service.ipFamilyPolicy	null	Optional Service dual-stack policy.
networkPolicy.egress.extraEgress	[]	Append full custom NetworkPolicy egress rules.
serviceMonitor.enabled	false	Render ServiceMonitor.

Links

• Metrics Server
• Upstream chart
• Chart source