Kubernetes Cost Optimization

Resource Optimization Spot Instances Committed Use Discounts Rightsizing

A comprehensive guide to Kubernetes cost optimization covering resource optimization, spot instances, committed use discounts, rightsizing, and practical implementation strategies for reducing cloud costs.

Resource Optimization Spot Instances Committed Use Discounts Rightsizing

Why Cost Optimization Matters

Kubernetes cost optimization is essential for managing cloud expenses while maintaining performance and reliability. Without optimization, costs can spiral due to:

Over-provisioning: Allocating more resources than needed
Idle Resources: Running unused or underutilized resources
Inefficient Workloads: Poor resource requests and limits
Lack of Visibility: No insight into cost drivers
Poor Scaling: Not scaling down when demand drops

Cost Optimization Principles:

Right-size: Match resources to actual needs
Scale Dynamically: Scale up and down with demand
Use Spot Instances: Leverage discounted capacity
Commit to Use: Use reserved instances for steady workloads
Monitor Continuously: Track and optimize regularly

Cost Optimization Strategies

Resource Rightsizing

Optimize resource requests and limits

Adjust CPU and memory requests/limits based on actual usage. Use VPA recommendations and monitoring data to find the optimal values.

Typical Savings: 20-40%

All workloads

Spot Instances

Use discounted spare capacity

Run fault-tolerant workloads on spot instances for up to 70-90% savings. Handle interruptions gracefully with node termination handlers.

Typical Savings: 60-90%

Stateless, fault-tolerant workloads

Committed Use Discounts

Reserve capacity for savings

Commit to using specific resources for 1-3 years for significant discounts (20-70%). Best for steady, predictable workloads.

Typical Savings: 20-70%

Steady-state workloads

Cluster Autoscaling

Dynamic node scaling

Automatically scale nodes up and down based on demand. Reduces cost during low-traffic periods by minimizing node count.

Typical Savings: 10-30%

Variable workloads

Storage Optimization

Optimize storage usage

Use appropriate storage classes, implement lifecycle policies, clean up unused volumes, and use compression for logs.

Typical Savings: 10-30%

Storage-heavy workloads

Efficient Networking

Optimize data transfer

Minimize cross-zone traffic, use caching, and implement content delivery networks for reduced bandwidth costs.

Typical Savings: 5-20%

Network-intensive workloads

Resource Rightsizing

Rightsizing involves adjusting resource requests and limits based on actual usage. It's one of the most effective cost optimization strategies.

                # Check current resource usage
kubectl top pods --all-namespaces
kubectl top nodes

# Identify over-provisioned pods
kubectl get pods -o json | jq '.items[] | {
  name: .metadata.name,
  namespace: .metadata.namespace,
  requests: .spec.containers[].resources.requests,
  limits: .spec.containers[].resources.limits
}'

# Use VPA for recommendations
apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: app-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: my-app
  updatePolicy:
    updateMode: "Off"  # Only recommend, don't apply
  resourcePolicy:
    containerPolicies:
    - containerName: '*'
      minAllowed:
        cpu: 50m
        memory: 64Mi
      maxAllowed:
        cpu: 2000m
        memory: 2Gi

# Get VPA recommendations
kubectl get vpa app-vpa -o yaml | grep -A 20 recommendation

# Manual rightsizing based on usage
# Original requests
requests:
  cpu: 500m
  memory: 512Mi

# After analysis (based on actual usage)
# Actual usage: CPU 150m, Memory 200Mi
requests:
  cpu: 200m   # Reduced from 500m
  memory: 256Mi  # Reduced from 512Mi

# Goldilocks tool for VPA recommendations
helm repo add fairwinds-stable https://charts.fairwinds.com/stable
helm repo update
helm install goldilocks fairwinds-stable/goldilocks --namespace goldilocks --create-namespace

# Goldilocks dashboard
kubectl port-forward -n goldilocks svc/goldilocks-dashboard 8080:80
            

Rightsizing Best Practices:

Use VPA recommendations as a starting point
Monitor actual usage over time (weeks)
Set requests based on steady-state usage
Set limits based on peak usage
Gradually adjust values to avoid instability
Review and adjust regularly (monthly)

Spot Instances for Cost Savings

Spot instances offer significant cost savings (60-90%) but can be interrupted. They're ideal for stateless, fault-tolerant workloads.

                # AWS Spot Node Group (eksctl)
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig
metadata:
  name: my-cluster
  region: us-east-1
nodeGroups:
- name: spot-workers
  instanceType: t3.medium
  desiredCapacity: 3
  spot: true  # Enable spot instances
  spotPrice: "0.02"
  labels:
    node-type: spot
  taints:
  - key: spot
    value: "true"
    effect: NoSchedule

# Pod toleration for spot nodes
apiVersion: v1
kind: Pod
metadata:
  name: spot-workload
spec:
  tolerations:
  - key: spot
    operator: Equal
    value: "true"
    effect: NoSchedule
  nodeSelector:
    node-type: spot
  containers:
  - name: app
    image: myapp:latest

# AWS Node Termination Handler
helm repo add aws https://aws.github.io/eks-charts
helm repo update
helm install aws-node-termination-handler aws/aws-node-termination-handler \
  --namespace kube-system \
  --set enableSpotInterruptionDraining=true \
  --set enableScheduledEventDraining=true

# GCP Preemptible VM
# Node pool with preemptible instances
gcloud container node-pools create spot-pool \
  --cluster=my-cluster \
  --zone=us-central1-a \
  --preemptible \
  --num-nodes=3 \
  --machine-type=n1-standard-2

# Azure Spot VM
az aks nodepool add \
  --resource-group my-rg \
  --cluster-name my-cluster \
  --name spotpool \
  --node-count 3 \
  --enable-spot \
  --spot-eviction-policy Delete
            

Spot Instance Considerations:

Use only for stateless, fault-tolerant workloads
Implement node termination handling
Set appropriate pod disruption budgets
Monitor spot interruption rates
Have fallback capacity (on-demand) for critical workloads
Test spot handling in staging

Committed Use Discounts

Committed use discounts (reserved instances) provide significant savings (20-70%) for predictable workloads by committing to use specific resources for 1-3 years.

                # AWS Reserved Instances
# Purchase standard or convertible RIs
# Standard RIs: Up to 72% savings, fixed instance type
# Convertible RIs: Up to 54% savings, flexible instance type

# AWS Compute Savings Plan
# Up to 66% savings, flexible across instance types and regions

# AWS EC2 Savings Plan
# Up to 72% savings, specific to EC2

# GCP Committed Use Discounts
# Up to 70% savings, 1-3 year commitment
gcloud compute commitments create \
  --region=us-central1 \
  --resources=vcpu=4,memory=16GB \
  --commitment-type=1-year

# Azure Reserved VM Instances
# Up to 72% savings, 1-3 year commitment
az reservation purchase \
  --reserved-resource-type VirtualMachines \
  --instance-flexibility Enabled \
  --quantity 3 \
  --term P1Y

# Kubernetes cluster with reservations
# Ensure reserved capacity matches your steady-state usage
# Monitor and adjust reservations based on usage
            

Committed Use Strategy:

Analyze historical usage to identify steady workloads
Purchase reservations for baseline capacity (e.g., 50-70% of usage)
Use spot/on-demand for variable capacity
Monitor utilization of reservations (aim for 80%+ utilization)
Use savings plans for flexibility
Review and adjust reservations regularly

Cost Monitoring and Visibility

                # Install Kubecost for cost visibility
helm repo add kubecost https://kubecost.github.io/cost-analyzer
helm repo update
helm install kubecost kubecost/cost-analyzer \
  --namespace kubecost \
  --create-namespace \
  --set prometheus.enabled=true \
  --set grafana.enabled=true

# Access Kubecost UI
kubectl port-forward -n kubecost svc/kubecost-cost-analyzer 9090:9090

# Check cost by namespace
kubectl get pods -n kubecost
# Access: http://localhost:9090

# Kubecost API for cost data
# Total monthly cost
curl http://localhost:9090/api/v1/allocation/aggregated?aggregate=namespace

# Cost by pod
curl http://localhost:9090/api/v1/allocation/aggregated?aggregate=pod

# Cloud provider cost monitoring
# AWS Cost Explorer
aws ce get-cost-and-usage --time-period Start=2025-01-01,End=2025-01-31

# GCP Cost Management
gcloud beta billing accounts describe

# Azure Cost Management
az consumption usage list --start-date 2025-01-01 --end-date 2025-01-31

# OpenCost (open-source alternative)
helm repo add opencost https://opencost.github.io/opencost-helm-chart
helm repo update
helm install opencost opencost/opencost --namespace opencost --create-namespace
            

Cost Monitoring Best Practices:

Implement cost visibility tools (Kubecost, OpenCost)
Monitor cost trends and anomalies
Set cost budgets and alerts
Track cost by namespace, team, and application
Regularly review cost reports
Share cost visibility with teams

Implementation Roadmap

1. Gain Visibility

Implement Kubecost or OpenCost. Understand where costs are coming from. Identify top cost drivers and trends.

2. Rightsize Resources

Use VPA recommendations and monitoring data. Adjust requests and limits. Aim for 60-80% utilization.

3. Implement Autoscaling

Configure HPA for pods and Cluster Autoscaler for nodes. Scale based on demand to avoid over-provisioning.

4. Adopt Spot Instances

Identify stateless workloads. Implement spot node groups with termination handling. Start with non-critical workloads.

5. Purchase Reservations

Analyze historical usage. Purchase reserved instances for steady workloads. Use savings plans for flexibility.

6. Optimize Storage

Review storage usage. Use appropriate storage classes. Implement lifecycle policies. Clean up unused volumes.

7. Monitor and Repeat

Continuous monitoring and optimization. Review costs monthly. Adjust strategies based on changing usage.

Cost Optimization Comparison

Strategy	Savings	Risk	Effort	Best For
Rightsizing	20-40%	Low	Medium	All workloads
Spot Instances	60-90%	High	High	Stateless, fault-tolerant
Committed Use	20-70%	Low	Low	Steady workloads
Cluster Autoscaling	10-30%	Low	Medium	Variable workloads
Storage Optimization	10-30%	Low	Medium	Storage-heavy workloads
Network Optimization	5-20%	Low	Medium	Network-intensive

Frequently Asked Questions

What is the most effective cost optimization strategy?

Rightsizing is the most effective starting point. It requires no infrastructure changes and provides significant savings. Combine with spot instances and committed use discounts for maximum savings.

What are the risks of using spot instances?

Spot instances can be interrupted with a 2-minute notice. Use them only for stateless, fault-tolerant workloads. Implement node termination handling and ensure pods can be rescheduled.

How do I know if I'm over-provisioned?

Check CPU and memory utilization. If usage is consistently below 60% of requests, you're over-provisioned. Use tools like Kubecost, Goldilocks, or VPA to identify over-provisioned pods.

What is the difference between spot and reserved instances?

Spot instances are spare capacity with high discounts but can be interrupted. Reserved instances are commitments with fixed discounts (20-70%). Use spot for flexible workloads, reserved for steady workloads.

How often should I review cost optimization?

Review monthly for trends and quarterly for major optimizations. Monitor cost dashboards continuously. Adjust rightsizing and reservations based on changing usage patterns.

What tools can help with Kubernetes cost optimization?

Kubecost (comprehensive), OpenCost (open-source), Goldilocks (VPA recommendations), CloudHealth, and cloud provider native tools (AWS Cost Explorer, GCP Cost Management).

How do I handle storage costs?

Use appropriate storage classes, implement lifecycle policies for automatic cleanup, use compression for logs, and monitor PVC usage. Consider using cheaper storage for backups and archival.

What is the recommended reserved instance coverage?

Aim for 50-70% of your steady-state workload covered by reservations. This provides significant savings while maintaining flexibility for variable workloads.

Previous: Node Maintenance Next: Troubleshooting Pods

Cost optimization is an ongoing process. Start with visibility, implement rightsizing, then gradually adopt spot instances and committed use discounts. Monitor continuously and adjust strategies based on changing usage patterns.