rag/.github/workflows/training-e2e.yaml

name: traning E2E

on:
  schedule: # schedule the job to run every day at midnight
    - cron: '0 12 * * *'

  pull_request:
    branches:
      - main
    paths:
      - .github/workflows/training-e2e.yaml
      - ./training/**

  workflow_dispatch:

env:
  TF_VAR_aws_region: "eu-west-2"
  TF_VAR_aws_ami_owners: '["309956199498"]'
  TF_VAR_aws_ami_name: '["*RHEL-9.4*"]'
  TF_VAR_aws_volume_size: 500
  TF_VAR_aws_access_key: ${{ secrets.AWS_ACCESS_KEY_ID }}
  TF_VAR_aws_secret_key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
  HF_TOKEN: ${{ secrets.HF_TOKEN }}

concurrency:
  group: ${{ github.workflow }}
  cancel-in-progress: false

jobs:
  e2e:
    if: github.repository == 'containers/ai-lab-recipes' && !contains(github.event.pull_request.labels.*.name, 'hold-tests')
    runs-on: ubuntu-24.04
    strategy:
      fail-fast: false
      max-parallel: 1
      matrix:
        include:
          - arch: amd64
            aws_image_type: g5.8xlarge
            image_name: nvidia-bootc
            aws_ami_architecture: x86_64
    steps:
      - name: Checkout
        uses: actions/checkout@v4.1.7
        with:
          path: main

      - name: Checkout terraform module
        id: checkout-module
        uses: actions/checkout@v4.1.7
        with:
          repository: containers/terraform-test-environment-module
          path: terraform-test-environment-module
          ref: 'main'

      - name: Setup Terraform
        uses: hashicorp/setup-terraform@v3.1.2
        with:
          terraform_version: "1.7.5"
          terraform_wrapper: false

      - name: Init
        run: terraform init
        working-directory: terraform-test-environment-module

      - name: Bootstrap
        id: up
        run: terraform apply -auto-approve -lock=false
        working-directory: terraform-test-environment-module
        env:
          TF_VAR_aws_instance_type: ${{ matrix.aws_image_type }}
          TF_VAR_aws_ami_architecture: ${{ matrix.aws_ami_architecture }}

      - name: Terraform Output
        id: terraform-output
        run: |
          echo "id=$(terraform output id | xargs)" >> $GITHUB_OUTPUT
          echo "url=$(terraform output host | xargs)" >> $GITHUB_OUTPUT
          echo "ssh_public_key=$(terraform output ssh_public_key | xargs)" >> $GITHUB_OUTPUT
          echo "pem_filename=$(terraform output pem_filename | xargs)" >> $GITHUB_OUTPUT
        working-directory: terraform-test-environment-module

      - name: Ansible Collections
        run: ansible-galaxy install -r ./tests/provision/requirements.yml
        working-directory: ./main/training

      - name: Provision
        run: |
          ansible-playbook ./main/training/tests/provision/playbook.yml \
            -i terraform-test-environment-module/hosts.ini \
            --private-key=terraform-test-environment-module/${{ steps.terraform-output.outputs.pem_filename }} \
            --extra-vars "image_name=${{ matrix.image_name }}" \
            --extra-vars "ssh_public_key='${{ steps.terraform-output.outputs.ssh_public_key }}'" \
            --extra-vars "registry_user=${{ secrets.REGISTRY_USER }}" \
            --extra-vars "registry_password=${{ secrets.REGISTRY_PASSWORD }}"
        env:
          ANSIBLE_CONFIG: ./main/training/tests/ansible.cfg

      # - name: Setup tmate session
      #   uses: mxschmitt/action-tmate@v3.18
      #   timeout-minutes: 20
      #   with:
      #     detached: true
      #     limit-access-to-actor: false

      - name: Setup tmate session
        uses: mxschmitt/action-tmate@v3.18
        timeout-minutes: 60
        with:
          detached: true
          limit-access-to-actor: false

      - name: Run tests
        run: |
          ansible-playbook ./main/training/tests/e2e-tests/playbook.yml \
            -i terraform-test-environment-module/hosts.ini \
            --private-key=terraform-test-environment-module/${{ steps.terraform-output.outputs.pem_filename }} \
            --extra-vars "HF_TOKEN=${{ secrets.HF_TOKEN }}" \
            --extra-vars "image_name=${{ matrix.image_name }}" \
            --extra-vars "ssh_public_key='${{ steps.terraform-output.outputs.ssh_public_key }}'" \
            --extra-vars "registry_user=${{ secrets.REGISTRY_USER }}" \
            --extra-vars "registry_password=${{ secrets.REGISTRY_PASSWORD }}"
        env:
          ANSIBLE_CONFIG: ./main/training/tests/ansible.cfg

      # This should exist in the final workflow
      # - name: Setup tmate session
      #   if: ${{ failure() }}
      #   uses: mxschmitt/action-tmate@v3.18
      #   timeout-minutes: 15
      #   with:
      #     detached: true
      #     limit-access-to-actor: false

      - name: Destroy Test Environment
        id: down
        if: always()
        run: terraform destroy -auto-approve -lock=false
        working-directory: terraform-test-environment-module
        env:
          TF_VAR_aws_instance_type: ${{ matrix.aws_image_type }}
          TF_VAR_aws_ami_architecture: ${{ matrix.aws_ami_architecture }}

      - name: Publish Job Results to Slack
        id: slack
        if: always()
        uses: slackapi/slack-github-action@v1.26.0
        with:
          payload: |
            {
              "text": "${{ github.workflow }} workflow status: ${{ job.status }}\n${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}"
            }
        env:
          SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}