GKE Autopilot

GKE Autopilot is a fully managed mode of operation where Google manages the entire cluster infrastructure — including nodes, node pools, scaling, security, and networking configuration. You only define your workloads, and Autopilot provisions the optimal resources to run them.

---

What Is Autopilot?

Autopilot is designed to reduce the operational overhead of running Kubernetes. In Standard mode, you manage node pools, machine types, scaling policies, and node-level security. In Autopilot, Google handles all of this automatically. You interact only with the Kubernetes API to deploy pods, services, and other workload resources.

Autopilot vs Standard

Aspect	Standard	Autopilot
Node management	You create and manage node pools	Google manages nodes automatically
Billing model	Per node (Compute Engine VMs)	Per pod (CPU, memory, storage requested)
Scaling	Configure cluster autoscaler per pool	Automatic — no configuration needed
Security hardening	You configure (optional)	Enforced by default
SSH to nodes	Yes	No
DaemonSets	Yes	Limited (Google-approved partners only)
Privileged containers	Allowed	Not allowed
Host network/PID	Allowed	Not allowed
Node affinity	Full control	Limited — no node pool management
GPU support	Yes	Yes (with specific configurations)
Windows containers	Yes	Not supported

---

Creating an Autopilot Cluster

# Create an Autopilot cluster
gcloud container clusters create-auto my-autopilot-cluster \
  --region europe-west2

# Get credentials
gcloud container clusters get-credentials my-autopilot-cluster \
  --region europe-west2

# Deploy a workload — Autopilot handles the rest
kubectl create deployment nginx --image=nginx:latest --replicas=3

---

How Billing Works

In Autopilot, you pay for the resources your pods request — CPU, memory, and ephemeral storage. You do not pay for idle node capacity. This makes costs more predictable and eliminates waste from over-provisioned nodes.

Minimum Resource Requirements

Autopilot enforces minimum resource requests per pod:

Resource	Minimum
CPU	250m (0.25 vCPU)
Memory	512 MiB
Ephemeral storage	10 MiB

If you do not specify resource requests, Autopilot applies defaults:

Resource	Default
CPU	500m
Memory	2 GiB

Billing Example

If you deploy 10 pods, each requesting 500m CPU and 1 GiB memory, you pay for:

• 5 vCPUs (10 x 500m)
• 10 GiB memory
• Regardless of how many nodes Autopilot provisions to run them

---

Security Enforced by Default

Autopilot enforces several security best practices that are optional in Standard mode: