Cluster FieldbookGKE platform architect
GKE · Platform ArchitectureView Markdown source

Autopilot or Standard

GKE offers two operating modes. The best default is the least operational control that still meets the requirement.

Autopilot

Google manages nodes, node provisioning, scaling, many security settings, and resource defaults. Workloads request CPU, memory, accelerators, and placement; GKE provisions capacity. Autopilot is optimized for most production workloads and is Google's recommended streamlined experience.

Choose it when the team wants Kubernetes APIs without operating node pools. Expect opinionated security constraints, automatic upgrades, and workload-based pricing. Verify compatibility for privileged agents, host access, specialized networking, or unsupported features.

Standard

You select node pools, machine types, operating systems, scaling, upgrades, taints, accelerators, and many low-level settings. Choose Standard when a proven requirement needs node-level control, a privileged DaemonSet, specialized hardware behavior, or a feature unavailable in Autopilot.

Decision table

Requirement Start with
Most web, API, worker, and batch containers Autopilot
Minimal platform operations Autopilot
Direct node configuration or privileged host agent Standard
Exact node purchase/capacity strategy Standard
Kubernetes learning lab needing node access Standard or local Kubernetes

Common error

“Standard gives more control” is not enough. Control has a carrying cost: capacity planning, upgrades, node security, repair, bin packing, and on-call knowledge. Record the exact capability that pays for that cost.

Official source: GKE Autopilot overview.

Cluster FieldbookIndependent study material · verify production details in Google Cloud documentation