GKE Cost Optimization Guide

Complete guide to optimizing costs in Google Kubernetes Engine (GKE), covering resource management, Spot VMs, committed use discounts, and best practices.

Cost Optimization Overview

Cost Factors in GKE

Factor	Impact	Optimization
Compute	60-80% of cost	Right-size, Spot VMs, CUDs
Storage	10-20% of cost	Right-size PVs, lifecycle policies
Networking	5-15% of cost	Optimize egress, use internal LBs
Management	Fixed per cluster	Consolidate workloads

---

Right-Sizing Resources

Resource Requests vs Usage

Use VPA Recommendations

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: my-app-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: my-app
  updatePolicy:
    updateMode: "Off"  # Recommendations only

# View recommendations
kubectl describe vpa my-app-vpa

# Example output:
# Recommendation:
#   Container Recommendations:
#     Container Name: app
#     Lower Bound:  Cpu: 25m, Memory: 64Mi
#     Target:       Cpu: 50m, Memory: 128Mi
#     Upper Bound:  Cpu: 100m, Memory: 256Mi

Resource Request Guidelines

Resource	Recommendation
CPU Request	P95 usage + 20% buffer
Memory Request	P99 usage + 10% buffer
CPU Limit	2-4x request or unset
Memory Limit	Equal to request

---

Spot VMs

Up to 91% discount for interruptible workloads.

Create Spot VM Pool

gcloud container node-pools create spot-pool \
    --cluster=my-cluster \
    --spot \
    --enable-autoscaling \
    --min-nodes=0 \
    --max-nodes=100 \
    --machine-type=e2-standard-4

Schedule on Spot VMs

spec:
  affinity:
    nodeAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 100
        preference:
          matchExpressions:
          - key: cloud.google.com/gke-spot
            operator: In
            values:
            - "true"
  tolerations:
  - key: cloud.google.com/gke-spot
    operator: Equal
    value: "true"
    effect: NoSchedule

Spot VM Best Practices

Practice	Description
Use for batch jobs	Jobs can handle interruption
Set Pod disruption budgets	Graceful handling
Enable preemption handling	Handle SIGTERM
Mix with on-demand	Fallback for critical workloads

---

Committed Use Discounts (CUDs)

Save up to 57% with 1-3 year commitments.

CUD Recommendations

Scenario	Recommendation
Stable baseline	Commit to baseline usage
Variable workloads	Combine CUD + Spot VMs
Multi-region	Regional CUDs for flexibility

Calculate CUD Coverage

# View current usage
gcloud compute resource-policies list

# Recommended: Commit to 60-70% of baseline
# Use Spot VMs for variable demand

---

Machine Type Selection

Cost-Effective Machine Types

Machine Type Comparison

Series	Cost	Best For
E2	Lowest	General purpose, dev/test
N2	Medium	Production workloads
N2D	Medium	AMD-based, cost-effective
C2	Higher	Compute-intensive
C2D	Higher	High single-thread performance

Shared-Core Options

Type	vCPU	Memory	Use Case
e2-micro	0.25	1 GB	Tiny services
e2-small	0.5	2 GB	Light workloads
e2-medium	1	4 GB	Small apps

---

Autoscaling for Cost

Scale to Zero

Configure HPA for Cost

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: my-app-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: my-app
  minReplicas: 0  # Scale to zero!
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 70
  behavior:
    scaleDown:
      stabilizationWindowSeconds: 300
      policies:
      - type: Percent
        value: 50
        periodSeconds: 60

Cluster Autoscaler Profile

# Aggressive scale-down for cost
gcloud container clusters update my-cluster \
    --autoscaling-profile=optimize-utilization

---

Storage Cost Optimization

Storage Type Selection

Storage Best Practices

Practice	Savings
Use pd-standard for cold data	60-75% vs SSD
Right-size PVCs	Pay only for what you use
Delete unused PVCs	Eliminate waste
Use regional PD selectively	2x cost for HA

Clean Up Unused Storage

# Find unbound PVCs
kubectl get pvc -A | grep -v Bound

# Find orphaned PVs
kubectl get pv | grep Released

# Delete unused PVC
kubectl delete pvc unused-pvc -n namespace

---

Network Cost Optimization

Egress Costs

Minimize Egress Costs

Strategy	Implementation
Co-locate services	Same zone node affinity
Use internal LBs	For internal traffic
Cache at edge	Cloud CDN for static content
Compress data	Reduce transfer size

Pod Anti-Affinity for Zone Spread

spec:
  affinity:
    podAntiAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 100
        podAffinityTerm:
          topologyKey: topology.kubernetes.io/zone

---

Cost Visibility

GKE Cost Allocation

Enable Cost Allocation

# Enable GKE usage metering
gcloud container clusters update my-cluster \
    --resource-usage-bigquery-dataset=gke_usage

# Add labels for cost tracking
gcloud container node-pools update my-pool \
    --cluster=my-cluster \
    --node-labels=cost-center=engineering,environment=production

Label Standards

Label	Purpose	Example
cost-center	Department allocation	engineering
environment	Dev/staging/prod	production
team	Team ownership	platform
app	Application name	api-gateway

---

Cost Optimization Checklist

Immediate Actions

Action	Potential Savings
Right-size resource requests	20-40%
Use Spot VMs for batch	60-91%
Delete idle resources	Variable
Use E2 machine types	10-30%

Medium-Term Actions

Action	Potential Savings
Implement autoscaling	30-50%
Committed use discounts	37-57%
Optimize storage classes	20-40%
Network optimization	10-20%

Ongoing Optimization

---

Quick Reference

Cost Commands

# View node pool machine types
gcloud container node-pools describe POOL --cluster=CLUSTER \
    --format="value(config.machineType)"

# List Spot VM nodes
kubectl get nodes -l cloud.google.com/gke-spot=true

# View resource usage
kubectl top pods -A --sort-by=cpu
kubectl top nodes

# Find over-provisioned pods
kubectl describe vpa -A | grep -A5 "Container Recommendations"

Cost Estimation

Resource	Monthly Cost (approx)
e2-standard-4 (on-demand)	~$100
e2-standard-4 (Spot)	~$30
e2-standard-4 (1yr CUD)	~$63
pd-balanced 100GB	~$10
pd-standard 100GB	~$4

---

Last updated: December 2025