Amazon EKS

Definition

Amazon Elastic Kubernetes Service (EKS) is a fully managed Kubernetes service that eliminates the operational overhead of running the Kubernetes control plane. AWS manages the control plane (API server, etcd, scheduler) across multiple Availability Zones, providing 99.95% uptime SLA. You manage the worker nodes (or delegate that to Fargate or managed node groups). EKS is AWS-certified Kubernetes — fully compatible with standard Kubernetes tools and ecosystem.

How EKS Works

Control Plane (AWS Managed)

• Kubernetes API server, scheduler, controller manager, etcd
• Runs across 3 AZs; AWS handles upgrades, patching, HA
• You pay $0.10/hour per cluster for control plane

Worker Nodes (You Choose)

• Self-managed nodes: EC2 instances you provision and manage
• Managed node groups: AWS-managed EC2 fleet with automated updates and lifecycle management
• Fargate profiles: Serverless compute — no nodes to manage; pay per pod

Compute Options

EKS Capabilities (New 2025)

EKS Capabilities streamline Kubernetes development by providing platform capabilities that handle workload orchestration and cloud resource management:

• Managed platform capabilities run on your cluster without you maintaining the underlying infrastructure
• Eliminates maintenance of add-ons like CoreDNS, kube-proxy, VPC CNI, and metrics-server
• Automatically updated and patched by AWS
• Reduces time spent on cluster maintenance

EKS Hybrid Nodes (New 2025)

EKS Hybrid Nodes extend your EKS cluster to on-premises or edge infrastructure:

• Register on-premises servers or VMs as EKS worker nodes
• Use the same Kubernetes API, kubectl, and tooling as cloud nodes
• Consistent GitOps and CI/CD pipelines across cloud and on-premises
• Designed for hybrid compliance requirements or latency-sensitive workloads near on-premises systems

Karpenter for Autoscaling

Karpenter (open-source, AWS-developed) is the recommended node autoscaler for EKS:

• Provisions the right node size for pending pod requirements automatically
• Bins pods efficiently to minimize node count
• Mixes Spot and On-Demand instances to optimize cost
• Consolidates workloads and terminates underutilized nodes
• Significantly faster scaling than the Kubernetes Cluster Autoscaler (seconds vs minutes)

EKS vs ECS vs Fargate

Common Mistakes

Mistake 1: Running EKS without Karpenter for cost optimization. Default Cluster Autoscaler provisions homogeneous node groups; Karpenter optimizes for cost by mixing instance types and spot/on-demand automatically.

Mistake 2: Ignoring EKS version upgrade deadlines. AWS supports each Kubernetes version for ~14 months. Clusters on unsupported versions enter extended support (additional cost). Plan upgrades every 3–6 months.

Mistake 3: Not using IRSA (IAM Roles for Service Accounts). Avoid giving EC2 node instance profiles broad permissions. Use IRSA to assign least-privilege IAM roles directly to Kubernetes service accounts.

Related AWS Services

• Amazon ECR: Container image registry for storing and versioning container images
• AWS Fargate: Serverless compute for EKS pods (no nodes to manage)
• AWS Load Balancer Controller: Manages ALB/NLB resources for Kubernetes Ingress/Service
• Amazon CloudWatch Container Insights: EKS monitoring and observability

Related FactualMinds Content

• AWS Serverless Services
• DevOps Pipeline Setup
• AWS Well-Architected Review Checklist