Skip to content

AWS EKS — Elastic Kubernetes Service

AWS EKS — Elastic Kubernetes Service

Amazon Elastic Kubernetes Service (EKS) is a managed Kubernetes service that removes the complexity of running and maintaining the Kubernetes control plane. You focus on deploying applications; AWS manages etcd, the API server, and control-plane updates.

In Azure terms: AWS EKS = Azure Kubernetes Service (AKS)

EKS Architecture

EKS Cluster
├── Control Plane (managed by AWS)
│   ├── API Server
│   ├── etcd
│   └── Scheduler / Controller Manager
└── Data Plane (your nodes)
    ├── Managed Node Group (EC2)
    ├── Self-managed Node Group (EC2)
    └── Fargate Profile (serverless)

Node Types

Type	Description	Best For
Managed Node Groups	AWS manages EC2 nodes — provisioning, updates, draining	Standard workloads
Self-managed Nodes	You manage EC2 nodes manually	Custom AMIs, fine control
Fargate Profiles	Serverless — no EC2 to manage; pods run on Fargate	Batch jobs, isolation, simplicity
Karpenter	Open-source node provisioner — faster scaling than Cluster Autoscaler	Cost-optimized autoscaling

IAM Integration — IRSA

IRSA (IAM Roles for Service Accounts) lets Kubernetes pods assume IAM roles without storing credentials:

# Service account with IRSA annotation
apiVersion: v1
kind: ServiceAccount
metadata:
  name: s3-reader
  namespace: default
  annotations:
    eks.amazonaws.com/role-arn: arn:aws:iam::123456789:role/eks-s3-read-role

Any pod using this service account can call S3 with the permissions of that IAM role — no credentials in the container. This is equivalent to Azure Workload Identity in AKS.

Key EKS Add-ons

Add-on	Purpose
AWS Load Balancer Controller	Provisions ALB/NLB from Kubernetes Ingress/Service objects
Amazon EBS CSI Driver	Dynamic provisioning of EBS volumes for PersistentVolumeClaims
Amazon EFS CSI Driver	Shared ReadWriteMany volumes via EFS
CoreDNS	DNS resolution inside the cluster
kube-proxy	Network rules on each node
VPC CNI	Pod networking — each pod gets a real VPC IP address
AWS Secrets Manager CSI	Mount secrets from Secrets Manager/SSM as files

EKS vs AKS Comparison

Feature	AWS EKS	Azure AKS
Managed control plane	Yes	Yes
Node types	EC2, Fargate	VMs, Virtual Nodes (ACI)
Node autoscaling	Cluster Autoscaler / Karpenter	Cluster Autoscaler / KEDA
Pod identity	IRSA	Azure Workload Identity
Container registry	ECR	ACR
Ingress	AWS Load Balancer Controller (ALB)	AGIC (App Gateway) / NGINX
Storage CSI	EBS, EFS CSI drivers	Azure Disk, Azure File CSI
Monitoring	CloudWatch Container Insights + X-Ray	Azure Monitor / Container Insights
Networking	VPC CNI (native pod IPs)	Azure CNI / Kubenet
Pricing	Control plane: $0.10/hour + node costs	Control plane: free + node costs

EKS control plane costs $0.10/hour (~$72/month) per cluster. AKS control plane is free.

Getting Started with EKS

# Install eksctl
# https://eksctl.io/installation/

# Create a cluster (takes ~15 minutes)
eksctl create cluster \
  --name my-cluster \
  --region us-east-1 \
  --nodegroup-name standard-nodes \
  --node-type t3.medium \
  --nodes 3 \
  --nodes-min 1 \
  --nodes-max 5 \
  --managed

# Configure kubectl
aws eks update-kubeconfig --region us-east-1 --name my-cluster

# Verify
kubectl get nodes
kubectl get pods -A

# Deploy a sample app
kubectl create deployment nginx --image=nginx
kubectl expose deployment nginx --port=80 --type=LoadBalancer

# Delete the cluster
eksctl delete cluster --name my-cluster --region us-east-1

Fargate Profile

Run pods on Fargate — no node management:

# eksctl Fargate profile
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig
metadata:
  name: my-cluster
  region: us-east-1

fargateProfiles:
  - name: default
    selectors:
      - namespace: default
      - namespace: kube-system

Cost Optimization Tips

Use Spot instances for non-critical workloads in managed node groups
Use Karpenter instead of Cluster Autoscaler for faster, cost-aware scaling
Right-size nodes with AWS Compute Optimizer recommendations
Use Fargate only for small or bursty workloads (Fargate is more expensive per CPU/RAM)
Set resource requests and limits on all pods to prevent over-provisioning

Useful Links