Ref: https://learn.cantrill.io/courses/1820301/lectures/42220586

• Summary
• Contents:

Amazon EKS - Key Concepts

• 🔧 AWS’s Kubernetes-as-a-Service (K8s-aaS)
  • AWS-managed implementation of K8s
  • Extends K8s to work seamlessly within AWS
  • 💡 K8s is Open-Source Software (OSS) & cloud-agnostic
    • AWS implemented K8s on their platform, they offer K8s as a service with EKS
    • Containers implemented in K8s and want to use AWS? → probably use EKS
• Different options to run EKS:
  • In AWS cloud (most common)
  • AWS Outposts, i.e. private, on-premises AWS
  • EKS Anywhere = create EKS clusters on-premises or anywhere else
  • EKS Distro = AWS released their EKS product as OSS
• Seamless integration with other AWS products
  • ECR, ELB, IAM, VPC…
  • Persistent storage: EBS, EFS, FSx for Lustre, FSx for NetApp ONTAP
• EKS Cluster = EKS Cluster Control Plane + EKS Nodes
  • Cluster Control Plane managed by AWS
    • scales based on load
    • runs across multiple AZs
    • etcd (key-value store) distributed across multiple AZs
  • Options for Node management:
    1. Self-managed
       • EC2 instances managed by customer
       • Billed based on normal EC2 pricing
    2. Managed node groups
       • EC2 instances managed by EKS
       • EKS handles provisioning & lifecycle management
    3. Fargate pods
       • Uses AWS Fargate for pods (containers) → similar to ECS-Fargate
       • Serverless
       • No need to worry about provisioning, configuration, scaling…
  • ‼️ Decide node management based on your business requirements!

    • Docs: https://docs.aws.amazon.com/eks/latest/userguide/eks-compute.html

    • E.g. must support Windows, GPU, Inferentia, Bottlerocket, Outposts, Local Zones…?

      → check node type in your region!! Make sure it's capable of needed features!

    • ❗ If not careful, project can be impacted very negatively

Amazon EKS - Architecture

Diagram: https://github.com/acantril/aws-sa-associate-saac03/blob/main/1000-CONTAINERS_and_ECS/00_LEARNINGAIDS/EKS101-1.png

• EKS Cluster Control Plane runs on an AWS-managed VPC across multiple AZs
• Nodes/Pods deployed to customer-managed VPC
  • Worker nodes (EC2 instances) deployed here
  • Fargate Pod ENIs injected here
  • Resource consumption done via ingress endpoint
• Node communication with Cluster Control Plane (kube-api-traffic) has two options:
  1. ENIs injected (by Cluster Control Plane) inside Customer VPC
  2. Public control plane endpoint
• ❗ Administration of the EKS Cluster Control Plane ONLY via public control plane endpoint!