GCP GKE Deployment

This guide covers deploying Greenfield Cluster on Google Kubernetes Engine (GKE) with ARM Tau T2A instances for optimal price/performance.

Overview

Google Kubernetes Engine provides managed Kubernetes clusters on GCP, offering:

Managed Control Plane: Google handles the Kubernetes control plane
ARM Tau T2A Support: Up to 50% price-performance advantage
GCP Integration: Native integration with Cloud services
Auto-scaling: Node auto-provisioning and cluster autoscaler
Autopilot Mode: Fully managed Kubernetes (optional)

Prerequisites

GCP account with billing enabled
gcloud CLI configured
Terraform >= 1.0 (for IaC)
kubectl v1.24+
GCP project with necessary APIs enabled

Deployment Options

Option 1: Terraform (Recommended)

Use our Terraform configuration for automated, reproducible deployments.

Quick Start

# Navigate to GCP infrastructure directory
cd infrastructure/gcp/

# Login and set project
gcloud auth login
gcloud auth application-default login
gcloud config set project YOUR_PROJECT_ID

# Enable required APIs
gcloud services enable container.googleapis.com compute.googleapis.com

# Initialize Terraform
terraform init

# Review planned changes
terraform plan

# Create cluster (takes 5-10 minutes)
terraform apply

# Configure kubectl
gcloud container clusters get-credentials greenfield-cluster \
  --zone us-central1-a

# Verify cluster
kubectl get nodes

Custom Configuration

Create terraform.tfvars:

project_id      = "your-gcp-project"
cluster_name    = "greenfield-cluster"
region          = "us-central1"
zone            = "us-central1-a"
environment     = "dev"
use_arm         = true           # Use ARM Tau T2A instances
regional        = false          # Zonal cluster (cheaper)
node_count      = 3
arm_machine_type = "t2a-standard-2"  # 2 vCPU, 8GB RAM

Apply configuration:

terraform apply

See the complete GCP Infrastructure Guide for all options.

Option 2: gcloud CLI

Quick cluster creation using gcloud command-line tool.

Basic Cluster with ARM

gcloud container clusters create greenfield-cluster \
  --zone us-central1-a \
  --machine-type t2a-standard-2 \
  --num-nodes 3 \
  --enable-autoscaling \
  --min-nodes 2 \
  --max-nodes 6 \
  --disk-size 50 \
  --disk-type pd-standard \
  --enable-autorepair \
  --enable-autoupgrade \
  --workload-pool=YOUR_PROJECT_ID.svc.id.goog

Regional Cluster (High Availability)

gcloud container clusters create greenfield-cluster \
  --region us-central1 \
  --machine-type t2a-standard-2 \
  --num-nodes 1 \
  --node-locations us-central1-a,us-central1-b,us-central1-c \
  --enable-autoscaling \
  --min-nodes 3 \
  --max-nodes 9 \
  --enable-autorepair \
  --enable-autoupgrade

Option 3: GCP Console

Manual creation through GCP Console:

Navigate: Console → Kubernetes Engine → Clusters
Create: Click "Create"
Mode: Standard or Autopilot
Configure:
Name: greenfield-cluster
Location type: Zonal (us-central1-a) or Regional
Kubernetes version: 1.28 or later
Node Pools:
Machine type: t2a-standard-2 (ARM)
Number of nodes: 3
Autoscaling: Min 2, Max 6
Disk: 50 GB standard persistent disk
Create: Wait 5-10 minutes

Architecture

Default Configuration

Component	Configuration
Region	us-central1 (configurable)
Zone	us-central1-a (or regional)
K8s Version	1.28+ (auto-upgrade enabled)
Control Plane	Managed by Google
Machine Type	t2a-standard-2 (ARM Tau T2A)
Node Count	2-6 (3 desired, auto-scaling)
Node Storage	50 GB pd-standard
Release Channel	Regular (balanced updates)

ARM vs x86 Instance Options

ARM (Tau T2A) - Default

Advantages: - Up to 50% better price-performance - Latest ARM architecture (Ampere Altra) - Energy efficient

Machine Types: - t2a-standard-2: 2 vCPU, 8 GB RAM (~$0.0536/hr) - t2a-standard-4: 4 vCPU, 16 GB RAM (~$0.1072/hr) - t2a-standard-8: 8 vCPU, 32 GB RAM (~$0.2144/hr)

x86 (N2/E2) - Fallback

Machine Types: - e2-standard-2: 2 vCPU, 8 GB RAM (~$0.0670/hr) - n2-standard-2: 2 vCPU, 8 GB RAM (~$0.0971/hr) - e2-standard-4: 4 vCPU, 16 GB RAM (~$0.1340/hr)

Cluster Types

Zonal Cluster (Default)

Cost: Lower (single zone)
Availability: Single zone
Use case: Development, testing, cost-sensitive workloads

--zone us-central1-a

Regional Cluster

Cost: Higher (~3x node costs)
Availability: Multi-zone (3 zones)
Use case: Production, high availability

--region us-central1 \
--node-locations us-central1-a,us-central1-b,us-central1-c

Deploying Greenfield Cluster

After creating your GKE cluster:

1. Configure kubectl

# Get credentials
gcloud container clusters get-credentials greenfield-cluster \
  --zone us-central1-a \
  --project YOUR_PROJECT_ID

# Verify connection
kubectl cluster-info
kubectl get nodes

2. Deploy Greenfield

Using Kustomize:

# Deploy base configuration
kubectl apply -k kustomize/base/

# Or use environment overlay
kubectl apply -k kustomize/overlays/prod/

Using Helm:

helm install greenfield helm/greenfield-cluster \
  --namespace greenfield \
  --create-namespace \
  --values helm/greenfield-cluster/values-prod.yaml

3. Configure Ingress (Optional)

GKE Ingress (HTTP(S) Load Balancer):

apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: greenfield-ingress
  annotations:
    kubernetes.io/ingress.class: "gce"
    kubernetes.io/ingress.global-static-ip-name: "greenfield-ip"
spec:
  rules:
    - host: api.example.com
      http:
        paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: fastapi-app
                port:
                  number: 8000

NGINX Ingress Controller:

# Install NGINX
helm repo add ingress-nginx https://kubernetes.github.io/ingress-nginx
helm install ingress-nginx ingress-nginx/ingress-nginx \
  --namespace ingress-nginx \
  --create-namespace

Cost Optimization

Estimated Monthly Costs

Development (t2a-standard-2, 3 nodes, zonal): - Control Plane: Free (zonal) - Worker Nodes: ~$117 (3 × $0.0536/hr × 730hr) - Persistent Disks: ~$15 (300GB × $0.04/GB-month) - Total: ~$132/month

Production (t2a-standard-4, 5 nodes, regional): - Control Plane: ~$73 (regional cluster fee) - Worker Nodes: ~$1,566 (15 nodes across 3 zones × $0.1072/hr × 730hr) - Persistent Disks: ~$50 (500GB) - Load Balancer: ~$18 - Total: ~$1,707/month (or ~$570/month for zonal)

Cost Reduction Tips

Use ARM Tau T2A: Up to 50% savings
Zonal vs Regional: Regional costs 3x more for nodes
Preemptible/Spot VMs: Up to 91% discount
Committed Use Discounts: 37-57% savings with 1-3 year commitments
Autopilot: Pay only for running pods (good for variable workloads)
Right-sizing: Use smaller machine types with autoscaling
Standard Persistent Disk: Use pd-standard instead of pd-ssd when possible

Enable Preemptible Nodes

For fault-tolerant workloads:

gcloud container node-pools create spot-pool \
  --cluster greenfield-cluster \
  --zone us-central1-a \
  --machine-type t2a-standard-2 \
  --spot \
  --num-nodes 2 \
  --enable-autoscaling \
  --min-nodes 0 \
  --max-nodes 5

GKE Autopilot

Fully managed option (pay per pod):

gcloud container clusters create-auto greenfield-cluster \
  --region us-central1 \
  --release-channel regular

Monitoring and Operations

Cloud Monitoring (formerly Stackdriver)

Enabled by default on GKE. View metrics:

# View in console
# Console → Monitoring → GKE

# Or use gcloud
gcloud container clusters describe greenfield-cluster \
  --zone us-central1-a \
  --format="value(monitoringConfig)"

Cluster Autoscaler

Already enabled if created with --enable-autoscaling. Configure:

gcloud container clusters update greenfield-cluster \
  --zone us-central1-a \
  --enable-autoscaling \
  --min-nodes 2 \
  --max-nodes 10 \
  --node-pool default-pool

Vertical Pod Autoscaler

Automatically adjust pod resources:

# Enable VPA
gcloud container clusters update greenfield-cluster \
  --zone us-central1-a \
  --enable-vertical-pod-autoscaling

Node Auto-Provisioning

Automatically create new node pools:

gcloud container clusters update greenfield-cluster \
  --zone us-central1-a \
  --enable-autoprovisioning \
  --min-cpu 1 \
  --max-cpu 100 \
  --min-memory 1 \
  --max-memory 1000

Security Best Practices

Workload Identity: Use GCP IAM for pod authentication
Private Clusters: Control plane on private IP
Binary Authorization: Ensure only trusted images run
Shielded GKE Nodes: Secure boot and vTPM
Network Policies: Restrict pod-to-pod communication
GKE Sandbox: Container isolation with gVisor
Regular Updates: Enable auto-upgrade

Enable Workload Identity

# Already enabled in Terraform config
gcloud container clusters update greenfield-cluster \
  --zone us-central1-a \
  --workload-pool=YOUR_PROJECT_ID.svc.id.goog

# Create service account binding
kubectl annotate serviceaccount default \
  iam.gke.io/gcp-service-account=GSA_NAME@PROJECT_ID.iam.gserviceaccount.com \
  --namespace greenfield

Private Cluster

gcloud container clusters create greenfield-cluster \
  --zone us-central1-a \
  --enable-private-nodes \
  --enable-private-endpoint \
  --master-ipv4-cidr 172.16.0.0/28

Troubleshooting

Nodes Not Ready

# Check node status
kubectl get nodes
kubectl describe node NODE_NAME

# Check node pool
gcloud container node-pools list \
  --cluster greenfield-cluster \
  --zone us-central1-a

# View logs
gcloud logging read "resource.type=k8s_node"

Pods Stuck in Pending

# Check events
kubectl get events --sort-by='.lastTimestamp'

# Check node resources
kubectl top nodes

# Scale node pool
gcloud container clusters resize greenfield-cluster \
  --zone us-central1-a \
  --num-nodes 5 \
  --node-pool default-pool

Persistent Volume Issues

# Check storage classes
kubectl get storageclass

# Check PVC status
kubectl get pvc -n greenfield

# View persistent disk
gcloud compute disks list

Quota Issues

# Check quotas
gcloud compute project-info describe --project YOUR_PROJECT_ID

# Request quota increase
# Console → IAM & Admin → Quotas

Cleanup

Terraform

# Delete Greenfield resources first
kubectl delete -k kustomize/base/

# Destroy GKE cluster
cd infrastructure/gcp/
terraform destroy

gcloud

# Delete cluster
gcloud container clusters delete greenfield-cluster \
  --zone us-central1-a

# Clean up persistent disks (if any remain)
gcloud compute disks list --filter="name~greenfield"
gcloud compute disks delete DISK_NAME --zone us-central1-a

GKE vs Alternatives

Feature	GKE Standard	GKE Autopilot	Competitor
Control Plane	Managed	Managed	Varies
Nodes	You manage	Google manages	You manage
Cost Model	Per node-hour	Per pod-resources	Per node-hour
Flexibility	High	Medium	High
Ops Burden	Medium	Low	Medium-High
Best For	Custom needs	Simple deployments	Varies