<div id="readability-page-1" class="page"><div><img src="https://devstarsj.github.io/assets/https://images.unsplash.com/photo-1667372393119-3d4c48d07fc9?w=1200"><h2 id="kubernetes-cost-optimization-7-proven-strategies-to-cut-cloud-spending-by-50">Kubernetes Cost Optimization: 7 Proven Strategies to Cut Cloud Spending by 50%</h2>Running Kubernetes in production is powerful—but expensive. The average organization wastes 30-40% of their cloud budget on idle or over-provisioned resources. Here’s how to fix that.<img src="https://images.unsplash.com/photo-1667372393119-3d4c48d07fc9?w=800" alt="Cloud Infrastructure"> Photo by <a href="https://unsplash.com/@growtika">Growtika</a> on UnsplashBefore diving into solutions, let’s understand where money disappears:<ul><li>Over-provisioned pods: Requesting 2GB RAM when using 200MB</li><li>Idle resources: Dev clusters running 24/7</li><li>Unused persistent volumes: Orphaned PVCs eating storage costs</li><li>Inefficient autoscaling: Scaling too aggressively or too slowly</li></ul><h2 id="strategy-1-right-size-your-workloads">Strategy 1: Right-Size Your Workloads</h2>The biggest win. Most teams set resource requests based on guesses, not data.<h3 id="use-vertical-pod-autoscaler-vpa-for-recommendations">Use Vertical Pod Autoscaler (VPA) for Recommendations</h3><div><pre><code>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" # Just get recommendations first
</code></pre></div>Check recommendations:<div><pre><code>kubectl describe vpa my-app-vpa
</code></pre></div><h3 id="tools-for-right-sizing">Tools for Right-Sizing</h3><ul><li>Kubecost: Free tier available, shows cost by namespace/deployment</li><li>Goldilocks: Open-source VPA dashboard</li><li>AWS Compute Optimizer: If running EKS</li></ul><h2 id="strategy-2-leverage-spotpreemptible-instances">Strategy 2: Leverage Spot/Preemptible Instances</h2>Spot instances cost 60-90% less than on-demand. Perfect for:<ul><li>Batch processing</li><li>CI/CD runners</li><li>Stateless workloads</li><li>Dev/staging environments</li></ul><h3 id="implementation-with-node-affinity">Implementation with Node Affinity</h3><div><pre><code>affinity:
 nodeAffinity:
 preferredDuringSchedulingIgnoredDuringExecution:
 - weight: 100
 preference:
 matchExpressions:
 - key: kubernetes.io/os
 operator: In
 values:
 - linux
 - key: node.kubernetes.io/lifecycle
 operator: In
 values:
 - spot
</code></pre></div><h3 id="handle-spot-interruptions">Handle Spot Interruptions</h3>Use AWS Node Termination Handler or Karpenter to gracefully drain nodes before termination.<img src="https://images.unsplash.com/photo-1558494949-ef010cbdcc31?w=800" alt="Server Infrastructure"> Photo by <a href="https://unsplash.com/@tvick">Taylor Vick</a> on Unsplash<h2 id="strategy-3-implement-cluster-autoscaler-properly">Strategy 3: Implement Cluster Autoscaler Properly</h2>Default autoscaler settings are conservative. Tune them:<div><pre><code># cluster-autoscaler deployment args
- --scale-down-delay-after-add=5m # Default 10m
- --scale-down-unneeded-time=5m # Default 10m
- --scale-down-utilization-threshold=0.5
- --skip-nodes-with-local-storage=false
</code></pre></div><h3 id="consider-karpenter-aws">Consider Karpenter (AWS)</h3>Karpenter is faster and smarter than Cluster Autoscaler:<ul><li>Provisions nodes in seconds, not minutes</li><li>Automatically selects optimal instance types</li><li>Better bin-packing</li></ul><h2 id="strategy-4-schedule-non-production-workloads">Strategy 4: Schedule Non-Production Workloads</h2>Dev and staging don’t need to run at 3 AM.<h3 id="using-kube-downscaler">Using Kube-downscaler</h3><div><pre><code>apiVersion: apps/v1
kind: Deployment
metadata:
 name: my-app
 annotations:
 downscaler/uptime: "Mon-Fri 08:00-20:00 Asia/Seoul"
</code></pre></div><h3 id="savings-potential">Savings Potential</h3><ul><li>8 hours/day × 5 days = 40 hours running</li><li>vs 168 hours/week</li><li>76% savings on dev/staging</li></ul><h2 id="strategy-5-use-resource-quotas-and-limit-ranges">Strategy 5: Use Resource Quotas and Limit Ranges</h2>Prevent teams from over-requesting:<div><pre><code>apiVersion: v1
kind: LimitRange
metadata:
 name: default-limits
 namespace: team-a
spec:
 limits:
 - default:
 memory: 512Mi
 cpu: 500m
 defaultRequest:
 memory: 256Mi
 cpu: 100m
 type: Container
</code></pre></div><div><pre><code>apiVersion: v1
kind: ResourceQuota
metadata:
 name: team-quota
 namespace: team-a
spec:
 hard:
 requests.cpu: "20"
 requests.memory: 40Gi
 limits.cpu: "40"
 limits.memory: 80Gi
</code></pre></div><h2 id="strategy-6-optimize-persistent-storage">Strategy 6: Optimize Persistent Storage</h2>Storage costs sneak up on you.<h3 id="clean-up-orphaned-pvcs">Clean Up Orphaned PVCs</h3><div><pre><code># Find unbound PVCs
kubectl get pvc --all-namespaces | grep -v Bound

# Find PVs without claims
kubectl get pv | grep Released
</code></pre></div><h3 id="use-appropriate-storage-classes">Use Appropriate Storage Classes</h3><table><thead><tr><th>Workload</th><th>Storage Type</th><th>Cost</th></tr></thead><tbody><tr><td>Databases</td><td>SSD (gp3)</td><td>$$</td></tr><tr><td>Logs</td><td>HDD (st1)</td><td>$</td></tr><tr><td>Temp data</td><td>Local SSD</td><td>$</td></tr><tr><td>Archives</td><td>S3/GCS</td><td>¢</td></tr></tbody></table><h2 id="strategy-7-monitor-and-set-budgets">Strategy 7: Monitor and Set Budgets</h2>You can’t optimize what you don’t measure.<h3 id="kubecost-setup">Kubecost Setup</h3><div><pre><code>helm install kubecost cost-analyzer \
 --repo https://kubecost.github.io/cost-analyzer/ \
 --namespace kubecost \
 --create-namespace
</code></pre></div><h3 id="set-cloud-provider-budgets">Set Cloud Provider Budgets</h3>Create alerts at 50%, 80%, and 100% of expected spend. Don’t wait for the bill.<h2 id="quick-wins-checklist">Quick Wins Checklist</h2><ul><li>Enable VPA in recommendation mode</li><li>Move CI/CD to spot instances</li><li>Schedule dev cluster downtime</li><li>Set default resource limits</li><li>Clean up orphaned PVCs</li><li>Install Kubecost or similar</li><li>Create budget alerts</li></ul><h2 id="expected-savings-by-strategy">Expected Savings by Strategy</h2><table><thead><tr><th>Strategy</th><th>Effort</th><th>Savings</th></tr></thead><tbody><tr><td>Right-sizing</td><td>Medium</td><td>20-30%</td></tr><tr><td>Spot instances</td><td>Medium</td><td>30-40%</td></tr><tr><td>Autoscaler tuning</td><td>Low</td><td>10-15%</td></tr><tr><td>Scheduling</td><td>Low</td><td>15-25%</td></tr><tr><td>Resource quotas</td><td>Low</td><td>10-20%</td></tr><tr><td>Storage optimization</td><td>Medium</td><td>5-15%</td></tr></tbody></table>Combined, these strategies typically achieve 40-60% cost reduction.<h2 id="conclusion">Conclusion</h2>Kubernetes cost optimization isn’t a one-time project—it’s an ongoing practice. Start with the quick wins (scheduling, autoscaler tuning), then tackle the bigger items (right-sizing, spot instances).The goal isn’t to minimize spending at all costs. It’s to maximize value per dollar spent. Sometimes that means spending more on production reliability while aggressively cutting dev/staging costs.Start measuring today. Your finance team will thank you. <div>이 글이 도움이 되셨다면 공감 및 광고 클릭을 부탁드립니다 :) </div></div></div>

Dev Note

Kubernetes Cost Optimization: 7 Proven Strategies to Cut Cloud Spending by 50%
Running Kubernetes in production is powerful—but expensive. The average organization wastes 30-40% of their cloud budget on idle or over-provisioned resources. Here’s how to fix that.
Photo by Growtika on Unsplash
The Hidden Cost of Kubernetes
Before diving into solutions, let’s understand where money disappears:
Over-provisioned pods: Requesting 2GB RAM when using 200MB
Idle resources: Dev clusters running 24/7
Unused persistent volumes: Orphaned PVCs eating storage costs
Inefficient autoscaling: Scaling too aggressively or too slowly
Strategy 1: Right-Size Your Workloads
The biggest win. Most teams set resource requests based on guesses, not data.
Use Vertical Pod Autoscaler (VPA) for 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"  # Just get recommendations first
Check recommendations:
kubectl describe vpa my-app-vpa
Tools for Right-Sizing
Kubecost: Free tier available, shows cost by namespace/deployment
Goldilocks: Open-source VPA dashboard
AWS Compute Optimizer: If running EKS
Strategy 2: Leverage Spot/Preemptible Instances
Spot instances cost 60-90% less than on-demand. Perfect for:
Batch processing
CI/CD runners
Stateless workloads
Dev/staging environments
Implementation with Node Affinity
affinity:
  nodeAffinity:
    preferredDuringSchedulingIgnoredDuringExecution:
    - weight: 100
      preference:
        matchExpressions:
        - key: kubernetes.io/os
          operator: In
          values:
          - linux
        - key: node.kubernetes.io/lifecycle
          operator: In
          values:
          - spot
Handle Spot Interruptions
Use AWS Node Termination Handler or Karpenter to gracefully drain nodes before termination.
Photo by Taylor Vick on Unsplash
Strategy 3: Implement Cluster Autoscaler Properly
Default autoscaler settings are conservative. Tune them:
# cluster-autoscaler deployment args
- --scale-down-delay-after-add=5m      # Default 10m
- --scale-down-unneeded-time=5m        # Default 10m
- --scale-down-utilization-threshold=0.5
- --skip-nodes-with-local-storage=false
Consider Karpenter (AWS)
Karpenter is faster and smarter than Cluster Autoscaler:
Provisions nodes in seconds, not minutes
Automatically selects optimal instance types
Better bin-packing
Strategy 4: Schedule Non-Production Workloads
Dev and staging don’t need to run at 3 AM.
Using Kube-downscaler
apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
  annotations:
    downscaler/uptime: "Mon-Fri 08:00-20:00 Asia/Seoul"
Savings Potential
8 hours/day × 5 days = 40 hours running
vs 168 hours/week
76% savings on dev/staging
Strategy 5: Use Resource Quotas and Limit Ranges
Prevent teams from over-requesting:
apiVersion: v1
kind: LimitRange
metadata:
  name: default-limits
  namespace: team-a
spec:
  limits:
  - default:
      memory: 512Mi
      cpu: 500m
    defaultRequest:
      memory: 256Mi
      cpu: 100m
    type: Container
apiVersion: v1
kind: ResourceQuota
metadata:
  name: team-quota
  namespace: team-a
spec:
  hard:
    requests.cpu: "20"
    requests.memory: 40Gi
    limits.cpu: "40"
    limits.memory: 80Gi
Strategy 6: Optimize Persistent Storage
Storage costs sneak up on you.
Clean Up Orphaned PVCs
# Find unbound PVCs
kubectl get pvc --all-namespaces | grep -v Bound
# Find PVs without claims
kubectl get pv | grep Released
Use Appropriate Storage Classes
Workload
      Storage Type
      Cost
    
Databases
      SSD (gp3)
      $$
    
Logs
      HDD (st1)
      $
    
Temp data
      Local SSD
      $
    
Archives
      S3/GCS
      ¢
    
Strategy 7: Monitor and Set Budgets
You can’t optimize what you don’t measure.
Kubecost Setup
helm install kubecost cost-analyzer \
  --repo https://kubecost.github.io/cost-analyzer/ \
  --namespace kubecost \
  --create-namespace
Set Cloud Provider Budgets
Create alerts at 50%, 80%, and 100% of expected spend. Don’t wait for the bill.
Quick Wins Checklist
Enable VPA in recommendation mode
Move CI/CD to spot instances
Schedule dev cluster downtime
Set default resource limits
Clean up orphaned PVCs
Install Kubecost or similar
Create budget alerts
Expected Savings by Strategy
Strategy
      Effort
      Savings
    
Right-sizing
      Medium
      20-30%
    
Spot instances
      Medium
      30-40%
    
Autoscaler tuning
      Low
      10-15%
    
Scheduling
      Low
      15-25%
    
Resource quotas
      Low
      10-20%
    
Storage optimization
      Medium
      5-15%
    
Combined, these strategies typically achieve 40-60% cost reduction.
Conclusion
Kubernetes cost optimization isn’t a one-time project—it’s an ongoing practice. Start with the quick wins (scheduling, autoscaler tuning), then tackle the bigger items (right-sizing, spot instances).
The goal isn’t to minimize spending at all costs. It’s to maximize value per dollar spent. Sometimes that means spending more on production reliability while aggressively cutting dev/staging costs.
Start measuring today. Your finance team will thank you.

Kubernetes Cost Optimization: 7 Proven Strategies to Cut Cloud Spending by 50%

댓글