Upgrade to Pro — share decks privately, control downloads, hide ads and more …

our_eks_to_ecs_migration_chronicle.pdf

k-ozawa
August 24, 2024

 our_eks_to_ecs_migration_chronicle.pdf

k-ozawa

August 24, 2024
Tweet

More Decks by k-ozawa

Other Decks in Technology

Transcript

  1. Today’s Agenda 1. Background 2. Challenges in Migration Planning 3.

    Solution Approach 4. Conceptual Differences between EKS and ECS 5. Implementation Challenges 6. Current Learnings and Future Expectations
  2. Background Operating a web system on EKS → Considering migration

    to ECS Amazon EKS Something awesome 「Delete “K”!」 Amazon ECS
  3. Reasons for Considering ECS Migration 1. High operational costs for

    Kubernetes updates 2. Team’s lack of Kubernetes experience 3. Re-evaluation of Kubernetes necessity 4. Extensive operational experience with ECS
  4. Challenges in Migration Planning 1. Lack of clear image for

    replacing EKS with ECS 2. Difficulty in grasping the overall picture and mitigating risks 3. Need concretize the migration process
  5. Solution Approach Identifying conceptual differences between EKS and ECS Designing

    a provisional architecture and conducting PoC Amazon EKS Amazon ECS Task Task Task
  6. Kubernetes Managed Services vs AWS Container Orchestration Service Node Group

    Amazon EKS Auto Scaling group EC2 instance Addon Scaling EC2 instance Addon Translate to AWS → Not required with ECS Manage node instances →Alternate with task group Manage node scaling →Same with ECS
  7. How to replace Kubernetes with ECS tasks Service Task ECS

    Service Connect Event Application LoadBalancer Migrate Kubernetes to ECS (Example...)
  8. Kubernetes All-in-One vs ECS + Other AWS Services EKS(Kubernetes) ECS

    Combination of components Combination of ECS and other AWS services Deployment, Ingress, Service, Job, Statefulset, CRD, etc... Functionality realized through ECS and other AWS services
  9. Choosing between Fargate and EC2 Consideration from performance and operational

    perspective Taking into account network mode constraints Amazon EC2 AWS Fargate
  10. Public Access Method in awsvpc Mode No ENI provided with

    public IP address in EC2 awsvpc mode Public subnet Amazon ECS Instance Internet
  11. Scaling Interval EKS ECS Horizonal Pod Autoscaler(HPA) Application AutoScaling using

    CloudWatch metrics Default 15-second interval Default 1-minute interval(for Fargate)
  12. Current Learning and Future Expectations 1. Deepening understanding of containers

    and microservices 2. Potential for reducing operational costs and improving quality 3. Strengthening ability to address technical challenges
  13. Summary Identifying conceptual differences between EKS and ECS • How

    to replace Kubernetes with ECS tasks • After understanding Kubernetes configuration patterns, replace them with the optimal design patterns for AWS Designing a provisional architecture and conducting PoC • Choose a platform based on performance and operational considerations, as well as network mode constraints • Recognize differences in horizontal scaling evaluation intervals and consider the best method