GKE

GKE is Google Cloud’s managed Kubernetes service. Google operates the control plane, and you choose how much of the node layer you also hand over, through two modes of operation.

How it works

In Autopilot mode, GKE manages the cluster infrastructure including nodes and scaling, and it provisions compute based on your workload specifications. General-purpose Pods are billed under a Pod-based model, while Pods that request specific hardware are billed on that hardware plus a node management premium. Autopilot clusters use a hardened configuration by default.

In Standard mode you control the infrastructure: you configure node pools, node settings and scaling, and you manage how Pods are placed on nodes. This is closer to the shape of EKS and AKS with self-managed or managed node groups.

A practical example

A team starts a new stateless API on Google Cloud. Using Autopilot, they create a cluster and apply a Deployment with CPU and memory requests. Capacity follows the Pod specifications, so the team does not size node pools. If the same team later needs a privileged agent or a node configuration that Autopilot does not allow, they would move that workload to a Standard cluster.

Common mistakes

• Choosing Standard by habit and then under-investing in node upgrades, node pool sizing and capacity planning.
• Leaving out resource requests. On Autopilot the requests drive capacity and cost, so they should reflect real usage; see resource requests and limits.
• Assuming a workload will run unchanged. Autopilot clusters enforce restrictions, so check the privileges your containers need before migrating.
• Not planning release channels and upgrade windows for the cluster.

When to use it and when not to

Google’s own guidance is to use Autopilot for most workloads, unless the application needs privileges or configuration options that conflict with Autopilot constraints, or you need the granular control of Standard. GKE is a natural choice when the rest of the platform is already on Google Cloud. If the estate is mostly on another provider, the operational cost of a second cloud usually outweighs any feature difference. Our Kubernetes consulting service covers cluster design across providers, and the GKE technology page gives more detail.

See also

• Kubernetes
• EKS
• AKS
• Horizontal Pod Autoscaler