Kubernetes Cost Optimization: FinOps Strategies That Actually Work

\n## 💰 Stop Overpaying for Kubernetes

Most organizations overprovision Kubernetes by 60-70%. That’s not a guess — it’s what I consistently find in cost audits. Here are the strategies that actually reduce your bill.

1. Right-Size Your Workloads

The biggest savings come from matching requests to actual usage:

# Install Goldilocks for right-sizing recommendations
helm install goldilocks fairwinds-stable/goldilocks -n goldilocks --create-namespace

# Enable VPA recommendations for a namespace
kubectl label namespace production goldilocks.fairwinds.com/enabled=true

# Check recommendations after 24-48 hours
kubectl get vpa -n production -o yaml

Common findings:

• Java apps requesting 4Gi memory but using 1.5Gi
• CPU requests at 1000m but averaging 100m
• Every pod requesting the same resources regardless of actual needs

2. Use Spot/Preemptible Instances

For fault-tolerant workloads, spot instances save 60-90%:

apiVersion: karpenter.sh/v1
kind: NodePool
metadata:
  name: spot-pool
spec:
  template:
    spec:
      requirements:
      - key: karpenter.sh/capacity-type
        operator: In
        values: ["spot"]
      - key: node.kubernetes.io/instance-type
        operator: In
        values: ["m5.xlarge", "m5a.xlarge", "m6i.xlarge", "m6a.xlarge"]
  disruption:
    consolidationPolicy: WhenEmptyOrUnderutilized
    consolidateAfter: 30s

Good for spot: Stateless web servers, batch jobs, CI/CD runners, dev/staging environments. Bad for spot: Databases, stateful services, long-running training jobs.

3. Namespace Cost Allocation with OpenCost

helm install opencost opencost/opencost -n opencost --create-namespace

# Query costs per namespace
curl -s http://opencost.opencost:9003/allocation/compute?window=7d&aggregate=namespace | \
  python3 -m json.tool

Show teams what they’re spending. Cost visibility drives behavior change faster than any policy.

4. Autoscaling That Works

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: api-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: api
  minReplicas: 2
  maxReplicas: 20
  behavior:
    scaleDown:
      stabilizationWindowSeconds: 300  # Wait 5 min before scaling down
      policies:
      - type: Percent
        value: 25
        periodSeconds: 60  # Scale down max 25% per minute
    scaleUp:
      policies:
      - type: Percent
        value: 100
        periodSeconds: 30  # Scale up fast
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 70

5. Quick Wins Checklist

• Delete unused PVCs (check with kubectl get pvc --all-namespaces | grep -v Bound)
• Remove unused LoadBalancer services (each costs $15-20/month on cloud)
• Set resource quotas per namespace
• Schedule dev/staging to scale down after hours
• Use ephemeral-storage limits to prevent log-bombs
• Compress container images (multi-stage builds, distroless base)

Average savings from a thorough FinOps review: 40-60% reduction in Kubernetes spend.

Need a Kubernetes cost audit? I help organizations optimize their cloud spend without sacrificing reliability. Get in touch.\n