--- title: "What is the Kubernetes control plane?" canonical: https://enum.co/learn/what-is-kubernetes-control-plane locale: en topic: Kubernetes publishedAt: 2026-10-16 updatedAt: 2026-10-16 author: "Max Heyer" summary: "The Kubernetes control plane manages a cluster's state through the API server, etcd, the scheduler and the controller managers. Worker nodes run pods; the control plane decides what runs where." --- # What is the Kubernetes control plane? The Kubernetes control plane manages a cluster's state through the API server, etcd, the scheduler and the controller managers. Worker nodes run pods; the control plane decides what runs where. After reading this you will be able to: - Name the control plane components and what breaks without each - Explain why running pods keep serving when the control plane is down - Understand why etcd runs with 3 or 5 members ## What is the Kubernetes control plane? The Kubernetes control plane is the part of a cluster that stores and manages its state. It consists of the API server, etcd, the scheduler and the controller managers. Worker nodes run your containers, while the control plane accepts changes, places pods on nodes and corrects the cluster until it matches what you asked for. Think of a thermostat. You set the temperature you want, and the thermostat keeps checking the room and switching the heating on or off until the two match. Every deploy, from kubectl or from a CI pipeline, goes through the control plane. It is also usually the layer a managed Kubernetes SLA measures. ## What are the control plane components? Four core components do the work. Each has one job, and all of them read and change cluster state through the API server. ### API server kube-apiserver exposes the Kubernetes HTTP API and validates the objects you send it. It is the frontend to the cluster's shared state, so without it there are no deploys, no scaling and no kubectl. ### etcd etcd is a consistent, highly available key-value store that holds all API server data. If that data is lost, the API server cannot start, and its state has to be recovered or recreated by hand. ### Scheduler kube-scheduler finds pods that are not yet bound to a node and assigns each one to a suitable node. Without it, new pods are not placed on any node. ### Controller managers kube-controller-manager runs the built-in controllers, such as the Deployment and Job controllers. These are the thermostats: each one compares the current state with the desired state and asks the API server to create or remove objects until they match. An optional cloud-controller-manager links the cluster to a cloud provider's API. It handles Node objects, routes and load balancers for Services. ## What happens when the control plane goes down? Running pods keep running, but you cannot change anything: you can't stop, update or start pods. Existing pods and services keep working unless they depend on the Kubernetes API. In practice: - Traffic to existing workloads usually keeps flowing. - Deploys, rollbacks and scaling fail, including autoscalers. - Operators and controllers you installed can break, because they depend on the Kubernetes API. - Pods on a node that fails during the outage are not replaced. So an outage rarely takes a website offline by itself, but it leaves you unable to react to anything else going wrong. ### Why etcd runs with 3 or 5 members etcd accepts an update only when a majority of its members, called a quorum, agree. Three members need two to agree and survive one failure. Five members need three and survive two failures. An even count adds nothing, since four members tolerate one failure, just like three. Going above seven hurts write performance, so most clusters run three members, or five to survive two failures at once. A highly available control plane runs several copies of each component behind a load balancer. Losing one machine then does not stop the API. ## How enum handles the control plane Every enum Kubernetes Engine cluster gets its own highly available control plane, never shared with another customer and included in the price. The replicas run in separate failure domains in one zone, so they survive a server failure but not the loss of the zone. Details and pricing are on the [Kubernetes Engine page](/kubernetes-engine). ## FAQ **Do my pods stop if the Kubernetes control plane goes down?** No. Existing pods keep running and serving traffic unless they depend on the Kubernetes API. You cannot deploy, scale or replace failed pods until the control plane is back. **Is etcd part of the control plane?** Yes. etcd is the key-value store for all API server data. It can run on the control plane nodes, called stacked etcd, or on separate hosts, called external etcd. **Why not run etcd with 2 or 4 members?** An even member count tolerates no more failures than the odd count below it. Two members tolerate zero failures, the same as one, and four tolerate one, the same as three.