Kubernetes Cost Optimization

Kubernetes Cost Optimization

Kubernetes clusters are notoriously over-provisioned. Studies show that 60-70% of allocated cluster resources are unused. This guide covers practical strategies to right-size your cluster and reduce costs by 30-50% without sacrificing reliability.

Understanding the Cost Problem

Kubernetes costs come from three sources:

• Compute (70%) — EC2/VM instances for worker nodes
• Storage (15%) — EBS volumes, EFS file systems
• Networking (15%) — Load balancers, data transfer, NAT gateways

The primary waste comes from over-provisioned resource requests. Developers set CPU/memory requests once and never revisit them.

Right-Sizing with VPA Recommendations

Deploy Vertical Pod Autoscaler in recommendation mode:

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: api-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: api-service
  updatePolicy:
    updateMode: "Off"

# Check recommendations
kubectl describe vpa api-vpa

# Typical output:
# Target: cpu=250m, memory=256Mi
# Lower Bound: cpu=100m, memory=128Mi
# Upper Bound: cpu=1000m, memory=512Mi

Spot Instances for Non-Critical Workloads

Use node selectors and tolerations to schedule workloads on Spot nodes:

spec:
  template:
    spec:
      nodeSelector:
        karpenter.sh/capacity-type: spot
      tolerations:
      - key: karpenter.sh/capacity-type
        operator: Equal
        value: spot
        effect: NoSchedule
      containers:
      - name: worker
        resources:
          requests:
            cpu: 500m
            memory: 512Mi

Spot instances save 60-90% on compute but can be interrupted. Use them for:

• Stateless web servers with multiple replicas
• Background job workers
• CI/CD build agents
• Development/staging environments

Kubecost for Visibility

Kubecost provides per-namespace, per-deployment, and per-label cost breakdowns:

helm install kubecost kubecost/cost-analyzer \
  --namespace kubecost \
  --create-namespace \
  --set kubecostToken="YOUR_TOKEN"

Key Kubecost insights:

• Idle costs — Resources allocated but not used
• Right-sizing recommendations — Based on actual usage
• Namespace cost allocation — Chargeback to teams
• Savings opportunities — Ranked by potential impact

Cluster Consolidation

Karpenter's consolidation feature automatically replaces underutilized nodes:

spec:
  disruption:
    consolidationPolicy: WhenUnderutilized
    consolidateAfter: 30s

When pods are removed and a node becomes underutilized, Karpenter will reschedule pods to other nodes and terminate the empty node.

Quick Wins Checklist

1. Deploy VPA in recommendation mode — Find over-provisioned workloads
2. Use Spot for stateless services — 60-90% compute savings
3. Right-size node instance types — Use Graviton for 40% better price/performance
4. Enable Karpenter consolidation — Automatic bin-packing
5. Schedule dev/staging shutdowns — Scale to zero outside business hours
6. Set resource requests accurately — Base on P95 usage, not guesses
7. Delete unused PVCs — Orphaned volumes cost money
8. Review load balancer count — Consolidate with shared Ingress

Eazy SaaS Tip: We run quarterly Kubernetes cost audits for our clients using Kubecost data combined with AWS Cost Explorer. The average client saves $2,000-5,000/month from right-sizing and Spot adoption alone.