All Posts
2024-06-30Hünkar Döner

Cost Optimization on AWS EKS

KubernetesEKSCost OptimizationSpot Instances
C

Cost Optimization on AWS EKS

Kubernetes has become the de-facto standard for modern application development. AWS's managed Kubernetes service, Amazon EKS (Elastic Kubernetes Service), simplifies this experience by offering scalability and reliability. However, one of the most frequent complaints in EKS usage is higher-than-expected bills. Between the "Control Plane" fee, hundreds of running pods, load balancers, and network traffic, costs can spiral out of control.

Here are proven strategies you can apply to reduce your costs on AWS EKS:

1. Using Spot Instances: Savings Up to 90%

The biggest item in EKS costs is the EC2 servers running as worker nodes. Using EC2 Spot Instances instead of paying "On-Demand" prices for these servers can reduce costs by up to 90%.

  • How It Works? Spot instances are AWS's idle capacity not currently in use. AWS gives a 2-minute warning when it wants to reclaim these servers.
  • EKS Integration: Kubernetes is inherently fault-tolerant. If your pods are "stateless", Kubernetes automatically restarts them on another node when a node goes down. You can manage this process seamlessly using the AWS Node Termination Handler.

2. Smart Auto-Scaling with Karpenter

Traditional "Cluster Autoscaler" works via node groups (Auto Scaling Groups) and can sometimes be slow or inefficient. AWS's open-source project Karpenter has changed the game.

  • Right-Sizing: Karpenter analyzes the resource needs of pending pods and launches the most suitable and cheapest EC2 type to meet that exact need in seconds. Instead of opening a large node and keeping half of it empty, it provides the resource that fits the need exactly (Bin Packing).

3. Optimize Resource Requests & Limits

The Kubernetes scheduler looks at CPU and Memory requests of pods when placing them. Developers often request more resources than necessary just to "be safe" (Over-provisioning).

  • Problem: If a pod requests 2 CPUs but uses only 0.1 CPU, you pay for those 2 CPUs but cannot use them.
  • Solution: Analyze the actual usage history of pods with tools like Goldilocks, Vertical Pod Autoscaler, or Kubecost and adjust requests closer to actual usage.

4. Manage Development Environments

Production environments must run 24/7, but what about Test, Dev, and Staging environments?

  • Off-Hours Shutdown: Scale down the worker node count in development environments to zero on weekends and nights. This provides savings up to 60%.
  • Use Spot: Run entire development environments on Spot instances.

5. Network and Storage Cleanup

  • EBS Volumes: Regularly identify and delete "unattached" EBS volumes left behind from deleted pods or nodes.
  • NAT Gateway: If pods access the internet via NAT Gateway, prevent unnecessary downloads or use VPC Endpoints.

EKS cost optimization is a technical process requiring continuous monitoring. You might consider getting expert AWS Consultancy to make your container infrastructure more efficient and lower your bill.