Kubernetes High Availability: Core Concepts

1 July 2026KubernetesInfrastructureHigh Availability

A single-node Kubernetes cluster is convenient for development but is a single point of failure in production. This article walks through the key concepts for building an HA cluster.

Control plane redundancy

The control plane consists of the API server, etcd, the scheduler, and the controller manager. For HA, you need at least three control-plane nodes so that etcd — which uses the Raft consensus algorithm — can tolerate the loss of one node.

┌──────────────────────────────────────────────┐
│  Load balancer (e.g. HAProxy / kube-vip)     │
└───────────┬──────────────┬───────────────────┘
            │              │
  ┌─────────▼──────┐  ┌────▼───────────┐  ┌────────────────┐
  │  Control  [1]  │  │  Control  [2]  │  │  Control  [3]  │
  │  API / etcd    │  │  API / etcd    │  │  API / etcd    │
  └────────────────┘  └────────────────┘  └────────────────┘

Use kubeadm with --control-plane-endpoint pointing to the load-balancer VIP:

kubeadm init \
  --control-plane-endpoint "k8s-api.example.internal:6443" \
  --upload-certs \
  --pod-network-cidr 10.244.0.0/16

Additional control-plane nodes join with:

kubeadm join k8s-api.example.internal:6443 \
  --token <token> \
  --discovery-token-ca-cert-hash sha256:<hash> \
  --control-plane \
  --certificate-key <cert-key>

Pod disruption budgets

Prevent all replicas of a deployment from being evicted simultaneously during node maintenance or rolling updates.

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: api-pdb
spec:
  minAvailable: 2       # Always keep at least 2 pods running
  selector:
    matchLabels:
      app: api

Liveness and readiness probes

Kubernetes relies on probes to decide whether to route traffic to a pod and whether to restart it.

livenessProbe:
  httpGet:
    path: /healthz
    port: 8080
  initialDelaySeconds: 10
  periodSeconds: 15
  failureThreshold: 3

readinessProbe:
  httpGet:
    path: /ready
    port: 8080
  initialDelaySeconds: 5
  periodSeconds: 5

Liveness: if this fails, the pod is restarted.
Readiness: if this fails, the pod is removed from the Service endpoints (no traffic).

Node affinity and topology spread

Spread replicas across availability zones to survive zone failures:

topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: topology.kubernetes.io/zone
    whenUnsatisfiable: DoNotSchedule
    labelSelector:
      matchLabels:
        app: api

← Back to Blogs