Infrastructure & Operations › Kubernetes & Orchestration

Managed Kubernetes (EKS, GKE, AKS)

Kubernetes with a cloud-run control plane.

Also known as: managed kubernetes, EKS, GKE, AKS

Managed Kubernetes is a Kubernetes service where the cloud provider runs the control plane for you. You create a cluster and get a working API server, etcd and its backups, and upgrades handled by the provider. Examples include Amazon EKS, Google GKE and Azure AKS; the underlying Kubernetes is the same, so your manifests and kubectl workflows carry over.

What the provider takes on:

• running and scaling the control plane, including high availability;
• patching and upgrading the Kubernetes version (often on your schedule);
• integrating with cloud networking, load balancers, storage and identity.

What you still own:

• the worker nodes (unless you use a serverless/autopilot mode) and their sizing;
• your workloads: requests and limits, scaling, and health;
• cluster add-ons, networking policy, and access control;
• cost.

provider: control plane, API server, etcd, upgrades
you:      nodes, workloads, configuration, security, cost

The classic mistakes:

• Assuming “managed” means “no ops”. It removes the hardest, least differentiating work — running etcd and the control plane — but node and workload operations are still yours. This is the shared responsibility model applied to Kubernetes.
• Letting the version drift. The provider upgrades the control plane, but your node pools and deprecated APIs are your concern. Stay close to a supported version to avoid forced, disruptive migrations.
• Ignoring cost. You pay for the control plane and for every node; idle overhead and overprovisioned pools add up (see cloud cost management).
• Accidental lock-in. Managed clusters integrate with provider networking, storage and identity. That’s convenient, but moving to another provider is real work (see vendor lock-in).

When not to use it: if you run on-premises, or need full control over the control plane, self-managed Kubernetes (or a simpler orchestrator) may fit. For most teams on a cloud, managed Kubernetes is the sensible default over running etcd yourself — the same build-versus-buy reasoning as managed vs self-hosted.