Infrastructure & Operations › Kubernetes & Orchestration

Control Plane

The components that manage cluster state.

Also known as: control plane, master node, kubernetes control plane

The control plane is the set of components in a Kubernetes cluster that make decisions and hold state. When you kubectl apply, you’re talking to the control plane, which records what you want and works to make reality match.

Its main pieces:

• the API server — the front door. Every command and every component goes through it.
• etcd — the key-value store that holds the cluster’s entire state. If this is lost, the cluster forgets everything.
• the scheduler — decides which node each new pod should run on.
• the controller manager — runs controllers that watch for drift and correct it (a Deployment’s desired replicas, for example).

The nodes run a separate set of components — the kubelet, which starts pods, and the networking proxy — but they follow the control plane’s instructions. This split is why a node can fail without the cluster losing its memory of what should be running.

kubectl ──▶ API server ──▶ etcd (state)
                │
                ├──▶ scheduler ──▶ picks a node
                └──▶ controllers ──▶ reconcile desired vs actual

The classic mistakes:

• A single control-plane node. It’s a single point of failure: if it dies, you can’t change the cluster or schedule new pods, even if workloads keep running briefly. For high availability you need multiple control-plane nodes, and etcd needs a quorum (an odd number, typically three).
• Assuming the control plane runs your apps. It mostly doesn’t — your workloads run on worker nodes. Losing the control plane degrades management, not necessarily the running services, which is why HA setups focus on keeping it available.
• Ignoring etcd. It’s the source of truth and must be backed up. A cluster without a restorable etcd backup is one bad day from starting over.
• Managing it yourself by accident. Running your own control plane is real work; managed Kubernetes exists to take it off your plate.

When not to manage it: almost always, unless you have specific reasons. Providers run the control plane for you, including its HA and upgrades — see managed Kubernetes and container orchestration.