How to Save Costs on Google Kubernetes Engine (GKE): A Practical Guide

Running workloads on Google Kubernetes Engine (GKE) is powerful, scalable—and, if you’re not careful, potentially expensive. Whether you’re using GKE Standard (where you manage the nodes) or GKE Autopilot (where Google manages the infrastructure for you), there are multiple ways to optimize and reduce your cloud bill without sacrificing reliability.

Below, we’ll break down the most effective strategies to save costs on GKE, with actionable examples.

1. Right-Size Your Resources

In Kubernetes, requested resources (CPU and memory) are what you’re billed for—not necessarily what your workloads actually use.

  • Set realistic requests and limits:
    Don’t over-provision. Use tools like the GKE Workload Recommendations or monitoring data (Cloud Monitoring, Prometheus) to analyze actual usage.
  • Example: 
    resources:
  requests:
    memory: "256Mi"
    cpu: "250m"
  limits:
    memory: "512Mi"
    cpu: "500m"
  • Regularly review: Adjust these values as your workloads grow or shrink. (doesn’t apply on GKE Auto..)

2. Use Spot Pods/Preemptible VMs for Non-Critical Workloads

  • GKE Standard:
    Use Preemptible VMs as node pool types. These cost up to 80% less than regular nodes but can be shut down at any time.
  • GKE Autopilot:
    Use Spot Pods by simply adding a label:

    metadata:
  labels:
    cloud.google.com/gke-spot: "true"

    Spot Pods are up to ~70% cheaper for workloads that can handle interruptions (batch jobs, cron jobs, etc.).

3. Scale Down Unused Resources

  • Autoscaling:
    • Enable Cluster Autoscaler in Standard mode.
    • In Autopilot, unused resources are automatically scaled down, but you should still delete unused deployments, namespaces, and stale workloads.
  • Pause/turn off non-prod environments:
    For development/test environments, automate their creation and deletion based on working hours or demand.

4. Clean Up Orphaned Resources

Cloud projects accumulate forgotten resources:

  • Delete unused PersistentVolumeClaims, LoadBalancers, unused IPs, and old container images.
  • Automate cleanup with scripts or use GKE usage metering to spot forgotten resources.

5. Choose Cheaper Regions and Zones

GKE resource prices vary depending on the location.

  • Use GCP’s pricing calculator to compare regions.
  • Example: us-central1 is usually cheaper than European regions.

6. Optimize Logging and Monitoring

Excessive logging/monitoring generates extra costs:

  • Lower the verbosity level of logs where possible.
  • Use Log Exclusions to filter out unnecessary logs.
  • Review Cloud Monitoring metrics and only collect what’s needed.

7. Take Advantage of Committed Use Discounts (Standard Only)

  • For predictable, steady workloads in GKE Standard, consider Committed Use Contracts.
  • These can save up to 57% on vCPU and memory costs.
    (Autopilot does not currently support committed use discounts.)

8. Monitor Your Spend and Set Budgets

  • Use Google Cloud Budgets & Alerts to get notified before you overspend.
  • Monitor GKE costs via Cost Breakdown reports.

9. Other Power Tips

  • Use minimal base images for containers to reduce network and storage costs.
  • Batch jobs and cron jobs: Always run them as Spot Pods if possible.
  • Check for idle public IPs, which can cost money even when not in use.

Conclusion

Kubernetes is a powerful abstraction, but without attention to detail, your cloud bill can quickly escalate.

By following the above practices—right-sizing, leveraging spot/preemptible resources, cleaning up unused resources, and monitoring spend—you’ll have a lean, efficient, and cost-effective GKE environment.

Happy saving!
If you have more tips or questions, feel free to share them in the comments.