preset-image-build.yml

name: Build and Push Preset Models

concurrency:
  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
  cancel-in-progress: true

on:
  pull_request:
    branches:
      - main
    paths:
      - 'presets/inference/**'
      - 'presets/models/supported_models.yaml'
  push:
    branches:
      - main
    paths:
      - 'presets/inference/**'
      - 'presets/models/supported_models.yaml'
  workflow_dispatch:
    inputs:
      force-run-all:
        type: boolean
        default: false
        description: "Run all models for build"

env:
    GO_VERSION: "1.20"
    BRANCH_NAME: ${{ github.head_ref || github.ref_name }}

permissions:
  id-token: write
  contents: read

jobs:
  determine-models: 
    runs-on: ubuntu-latest
    environment: preset-env
    outputs:
      matrix: ${{ steps.affected_models.outputs.matrix }}
      is_matrix_empty: ${{ steps.check_matrix_empty.outputs.is_empty }}
    steps: 
      - name: Checkout
        uses: actions/checkout@v4
        with:
          submodules: true
          fetch-depth: 0
      
      - name: Set FORCE_RUN_ALL Flag
        run: echo "FORCE_RUN_ALL=${{ github.event_name == 'workflow_dispatch' && github.event.inputs.force-run-all == 'true' }}" >> $GITHUB_ENV
      
      # This script should output a JSON array of model names
      - name: Determine Affected Models
        id: affected_models
        run: |
          PR_BRANCH=${{ env.BRANCH_NAME }} \
          FORCE_RUN_ALL=${{ env.FORCE_RUN_ALL }} \
          PR_REPO_URL=${{ github.event.pull_request.head.repo.clone_url }} \
          python3 .github/workflows/kind-cluster/determine_models.py
            
      - name: Print Determined Models
        run: |
          echo "Output from affected_models: ${{ steps.affected_models.outputs.matrix }}"
      
      - name: Check if Matrix is Empty
        id: check_matrix_empty
        run: |
          if [ "${{ steps.affected_models.outputs.matrix }}" == "[]" ] || [ -z "${{ steps.affected_models.outputs.matrix }}" ]; then
            echo "is_empty=true" >> $GITHUB_OUTPUT
          else
            echo "is_empty=false" >> $GITHUB_OUTPUT
          fi
  
  build-models:
    needs: determine-models
    if: needs.determine-models.outputs.is_matrix_empty == 'false'
    runs-on: [self-hosted, 'hostname:model-server']
    environment: preset-env
    strategy:
      fail-fast: false
      matrix:
        model: ${{fromJson(needs.determine-models.outputs.matrix)}}
      max-parallel: 3
    steps:
      - name: Checkout
        uses: actions/checkout@v4
        with:
          submodules: true
          fetch-depth: 0
      
      - name: Check Available Disk Space
        run: df -h

      - name: Install Azure CLI latest
        run: |
              if ! which az > /dev/null; then
                  echo "Azure CLI not found. Installing..."
                  curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash
              else
                  echo "Azure CLI already installed."
              fi
    
      - name: 'Az CLI login'
        uses: azure/login@v1.6.1
        with:
            client-id: ${{ secrets.AZURE_CLIENT_ID }}
            tenant-id: ${{ secrets.AZURE_TENANT_ID }}
            allow-no-subscriptions: true
        
      - name: 'Set subscription'
        run: az account set --subscription ${{secrets.AZURE_SUBSCRIPTION_ID}}

      - name: 'Get ACR Info'
        id: acr_info
        run: |
          ACR_NAME="${{ secrets.ACR_AMRT_USERNAME }}"
          ACR_USERNAME=${{ secrets.ACR_AMRT_USERNAME }}
          ACR_PASSWORD=${{ secrets.ACR_AMRT_PASSWORD }} 
          echo "ACR_NAME=$ACR_NAME" >> $GITHUB_OUTPUT
          echo "ACR_USERNAME=$ACR_USERNAME" >> $GITHUB_OUTPUT
          echo "ACR_PASSWORD=$ACR_PASSWORD" >> $GITHUB_OUTPUT
      
      - name: 'Check if Image exists in Test ACR'
        id: check_test_image
        run: |
            ACR_NAME=${{ steps.acr_info.outputs.ACR_USERNAME }}
            IMAGE_NAME=${{ matrix.model.name }}
            TAG=${{ matrix.model.tag }}

            # Use '|| true' to prevent script from exiting with an error if the repository is not found
            TAGS=$(az acr repository show-tags -n $ACR_NAME --repository $IMAGE_NAME --output tsv || true)

            if [[ -z "$TAGS" ]]; then
                echo "Image $IMAGE_NAME:$TAG or repository not found in $ACR_NAME."
                echo "IMAGE_EXISTS=false" >> $GITHUB_OUTPUT
            else
                if echo "$TAGS" | grep -q "^$TAG$"; then
                    echo "IMAGE_EXISTS=true" >> $GITHUB_OUTPUT
                else
                    echo "IMAGE_EXISTS=false" >> $GITHUB_OUTPUT
                    echo "Image $IMAGE_NAME:$TAG not found in $ACR_NAME."
                fi
            fi
      
      - name: Launch Python Script to Kickoff Build Jobs
        if: steps.check_test_image.outputs.IMAGE_EXISTS == 'false'
        id: launch_script
        run: |
          PR_BRANCH=${{ env.BRANCH_NAME }} \
          ACR_NAME=${{ steps.acr_info.outputs.ACR_NAME }} \
          ACR_USERNAME=${{ steps.acr_info.outputs.ACR_USERNAME }} \
          ACR_PASSWORD=${{ steps.acr_info.outputs.ACR_PASSWORD }} \
          MODEL_NAME=${{ matrix.model.name }} \
          MODEL_TYPE=${{matrix.model.type}} \
          MODEL_VERSION=${{ matrix.model.version }} \
          MODEL_RUNTIME=${{ matrix.model.runtime }} \
          MODEL_TAG=${{ matrix.model.tag }} \
          python3 .github/workflows/kind-cluster/main.py
      
      # Check the exit status of the Python script
      - name: Check Python Script Status
        if: ${{ always() }}
        run: |
          if [[ "${{ steps.check_test_image.outputs.IMAGE_EXISTS }}" == "true" ]]; then
            echo "Image already exists; skipping the status step."
          elif [[ "${{ steps.launch_script.outcome }}" != "success" ]]; then
            echo "Python script failed to execute successfully."
            exit 1  # Fail the job due to script failure
          else
            echo "Python script executed successfully."
          fi
      # Cleanup Resources
      - name: Cleanup
        if: ${{ always() }}
        run: |
          if [[ "${{ steps.check_test_image.outputs.IMAGE_EXISTS }}" == "false" ]]; then
            kubectl get job --no-headers -o custom-columns=":metadata.name" | grep "^docker-build-job-${{ matrix.model.name }}-[0-9]" | xargs -r kubectl delete job
          fi