Overview

Introduction

Project MOKU focuses on dynamically provisioning IT infrastructure and deploying software based on user requirements. Before this project, AI researchers with limited knowledge of IT infrastructure faced significant challenges in setting up hardware and software environments in the cloud. These challenges resulted in two main problems:

1. Researchers were often allocated more computing resources than necessary, leading to inefficiency.
2. Setting up the desired virtual environments on allocated resources was time-consuming for each researcher.

To address these issues, I leveraged Terraform and Kubernetes to dynamically provision infrastructure and virtual environments tailored to the needs of AI researchers, based on web-based user input. This approach significantly improved development efficiency while reducing costs.

Task

• Cluster Lifecycle Management
  • Frontend (Backstage based)
  • Backend (Backstage based)
  • Infrastructure-as-Code (IaC) and CI/CD Pipeline
• Cluster Resource Management via Web Terminal
  • Frontend
  • Web Terminal Server
  • Deployment

Approach

• Cluster Lifecycle Management

  

  • Frontend (Backstage based)

    • Cluster resource status, component configuration, and error overview
    • Multi-cluster management UI

  • Backend (Backstage based)

    • APIs and business logic (user data, cluster status, etc.)
    • GitHub Actions triggers to create, pause, resume, and delete clusters

  • Infrastructure-as-Code (IaC) and CI/CD Pipeline

    

    • Cluster Creation/Deletion
      • GitHub Actions with Bash scripts and Terraform files to provision or delete cloud infrastructure based on user input
      • GitHub Actions to generate or delete Helm charts in the GitHub repository for software tools, Kubernetes add-ons, web terminal server, etc.
      • **Argo CD to automatically sync repository Helm charts with target clusters**
    • Cluster Pause/Resume
      • Pause: GitHub Actions with Bash scripts and Terraform to scale down the cluster, delete all workloads, and release computing resources
      • Resume: GitHub Actions with Bash scripts and Terraform to recreate all computing resources and workloads

• Cluster Resource Management via Web Terminal

  • Frontend

    • Web terminal UI for each cluster

  • Web Terminal Server

    • Seamless setup with kube-context, tools, and basic software
    • Execution of commands sent from the web terminal

  • Deployment

    

    • Automatic creation of command-line processing server when a cluster is created