deploy-aks-callable.yml

name: Deploy AKS (Reusable)

on:
  workflow_call:
    inputs:
      location:
        description: "Azure region"
        required: false
        type: string
        default: "westus3"
      command:
        description: "deploy-aks.sh command to run"
        required: false
        type: string
        default: "deploy-aks"
      node_pool_vm_size:
        description: "GPU node pool VM size"
        required: false
        type: string
        default: "Standard_D2ads_v5"
      node_pool_node_count:
        description: "GPU node pool node count"
        required: false
        type: string
        default: "2"
      system_pool_vm_size:
        description: "Optional system nodepool VM size"
        required: false
        type: string
        default: "Standard_D2ads_v5"
      install_network_operator:
        description: "Install NVIDIA Network Operator for InfiniBand/RDMA support"
        required: false
        type: boolean
        default: false
      install_gpu_operator:
        description: "Install NVIDIA GPU Operator for GPU workload management"
        required: false
        type: boolean
        default: false
      cleanup:
        description: "Delete the resource group after the run"
        required: false
        type: boolean
        default: true

jobs:
  deploy:
    runs-on: ubuntu-latest
    environment: testing

    steps:
      - name: Checkout repository
        uses: actions/checkout@v4

      - name: Validate deploy script is syntactically valid
        run: |
          set -euo pipefail
          bash -n infrastructure_references/aks/scripts/deploy-aks.sh

      - name: Azure Login
        uses: azure/login@v2
        with:
          client-id: ${{ secrets.AZURE_CLIENT_ID }}
          tenant-id: ${{ secrets.AZURE_TENANT_ID }}
          subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }}

      - name: Install jq and kustomize
        run: |
          set -euo pipefail
          sudo apt-get update
          sudo apt-get install -y jq
          curl -s "https://raw.githubusercontent.com/kubernetes-sigs/kustomize/master/hack/install_kustomize.sh"  | bash

      - name: Setup kubectl
        uses: azure/setup-kubectl@v3

      - name: Setup helm
        uses: azure/setup-helm@v3

      - name: Verify CLI tools
        run: |
          set -euo pipefail
          kubectl version --client=true
          helm version
          kustomize version
          az version

      - name: Set script permissions
        run: |
          set -euo pipefail
          chmod +x infrastructure_references/aks/scripts/deploy-aks.sh

      - name: Generate resource group and cluster names
        id: names
        run: |
          set -euo pipefail
          timestamp=$(date -u +"%y%d%m%H%M")
          rg_name="ai-infra-aks-${timestamp}"
          cluster_name="ai-infra-${timestamp}"
          echo "resource_group=${rg_name}" >> "$GITHUB_OUTPUT"
          echo "cluster_name=${cluster_name}" >> "$GITHUB_OUTPUT"

      - name: Run deploy-aks.sh
        env:
          AZURE_REGION: ${{ inputs.location }}
          AZURE_RESOURCE_GROUP: ${{ steps.names.outputs.resource_group }}
          CLUSTER_NAME: ${{ steps.names.outputs.cluster_name }}
          SYSTEM_POOL_VM_SIZE: ${{ inputs.system_pool_vm_size }}
          NODE_POOL_VM_SIZE: ${{ inputs.node_pool_vm_size }}
          NODE_POOL_NODE_COUNT: ${{ inputs.node_pool_node_count }}
          INSTALL_NETWORK_OPERATOR: ${{ inputs.install_network_operator }}
          INSTALL_GPU_OPERATOR: ${{ inputs.install_gpu_operator }}
        run: |
          set -euo pipefail
          ./infrastructure_references/aks/scripts/deploy-aks.sh "${{ inputs.command }}"

      - name: Cleanup resource group after run
        if: always() && inputs.cleanup == true
        run: |
          set -euo pipefail
          echo "Cleaning up resource group: ${{ steps.names.outputs.resource_group }}"
          if az group exists --name "${{ steps.names.outputs.resource_group }}" --output tsv 2>/dev/null | grep -q "true"; then
            az group delete --name "${{ steps.names.outputs.resource_group }}" --yes --no-wait
            echo "✅ Resource group deletion initiated" >> "$GITHUB_STEP_SUMMARY"
          else
            echo "Resource group does not exist, nothing to clean up"
          fi