nightly_orchestrator.yml

name: Nightly Orchestrator

on:
  schedule:
    - cron: '0 6 * * *'  # 6AM UTC daily
  workflow_dispatch:
    inputs:
      mode:
        description: 'Run mode: full (run tests + deploy dashboard), dry-run (run tests, skip dashboard deploy), smoke-test (skip tests, validate CI plumbing only)'
        type: choice
        options:
          - full
          - dry-run
          - smoke-test
        default: full

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

jobs:
  # =============================================================================
  # Resolve the effective mode (schedule always uses "full")
  # =============================================================================
  config:
    name: Configure
    runs-on: ubuntu-latest
    outputs:
      mode: ${{ steps.resolve.outputs.mode }}
      run_tests: ${{ steps.resolve.outputs.run_tests }}
      deploy_dashboard: ${{ steps.resolve.outputs.deploy_dashboard }}
    steps:
      - name: Resolve mode
        id: resolve
        run: |
          MODE="${{ inputs.mode || 'full' }}"
          echo "mode=$MODE" >> "$GITHUB_OUTPUT"

          if [ "$MODE" = "smoke-test" ]; then
            echo "run_tests=false" >> "$GITHUB_OUTPUT"
            echo "deploy_dashboard=true" >> "$GITHUB_OUTPUT"
          elif [ "$MODE" = "dry-run" ]; then
            echo "run_tests=true" >> "$GITHUB_OUTPUT"
            echo "deploy_dashboard=false" >> "$GITHUB_OUTPUT"
          else
            echo "run_tests=true" >> "$GITHUB_OUTPUT"
            echo "deploy_dashboard=true" >> "$GITHUB_OUTPUT"
          fi

          echo "### Nightly Orchestrator Configuration" >> "$GITHUB_STEP_SUMMARY"
          echo "- **Mode**: $MODE" >> "$GITHUB_STEP_SUMMARY"

  # =============================================================================
  # Test workflows (skipped in smoke-test mode)
  # =============================================================================
  test-linux:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-linux.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  test-linux-llm:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-linux-llm.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  test-linux-habitat:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-linux-habitat.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  test-linux-tutorials:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-linux-tutorials.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  test-linux-sota:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-linux-sota.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  test-linux-libs:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-linux-libs.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  test-windows-optdepts:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/test-windows-optdepts.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  lint:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/lint.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  docs:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/docs.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: write

  benchmarks:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/benchmarks.yml
    with:
      skip-upload: true
    secrets: inherit
    permissions:
      id-token: write
      contents: write
      deployments: write
      pull-requests: write

  validate-test-partitioning:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/validate-test-partitioning.yml
    secrets: inherit

  nightly-build:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/nightly_build.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  # The four build-wheels-* jobs run sequentially, not in parallel. The
  # pytorch/test-infra build_wheels_{linux,macos,windows}.yml reusable
  # workflows they call each define a workflow-level concurrency group keyed
  # on the caller's workflow name and sha (identical for all four when called
  # from this orchestrator) with cancel-in-progress: true. When invoked in
  # parallel, each new claim of the group cancels the previous child's jobs,
  # which marked every scheduled orchestrator run as cancelled. Chaining via
  # needs serializes the claims; !cancelled() keeps the chain (and the status
  # dashboard coverage) going when an earlier wheel build fails.
  build-wheels-linux:
    needs: config
    if: needs.config.outputs.run_tests == 'true'
    uses: ./.github/workflows/build-wheels-linux.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  build-wheels-m1:
    needs:
      - config
      - build-wheels-linux
    if: ${{ !cancelled() && needs.config.outputs.run_tests == 'true' }}
    uses: ./.github/workflows/build-wheels-m1.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  build-wheels-windows:
    needs:
      - config
      - build-wheels-m1
    if: ${{ !cancelled() && needs.config.outputs.run_tests == 'true' }}
    uses: ./.github/workflows/build-wheels-windows.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  build-wheels-aarch64-linux:
    needs:
      - config
      - build-wheels-windows
    if: ${{ !cancelled() && needs.config.outputs.run_tests == 'true' }}
    uses: ./.github/workflows/build-wheels-aarch64-linux.yml
    secrets: inherit
    permissions:
      id-token: write
      contents: read

  # =============================================================================
  # Status collector - updates the nightly dashboard after all workflows complete
  # =============================================================================
  update-status-dashboard:
    name: Update Status Dashboard
    runs-on: ubuntu-latest
    permissions:
      contents: write
    needs:
      - config
      - test-linux
      - test-linux-llm
      - test-linux-habitat
      - test-linux-tutorials
      - test-linux-sota
      - test-linux-libs
      - test-windows-optdepts
      - lint
      - docs
      - benchmarks
      - validate-test-partitioning
      - nightly-build
      - build-wheels-linux
      - build-wheels-m1
      - build-wheels-windows
      - build-wheels-aarch64-linux
    if: always()
    steps:
      - name: Trigger status collector
        uses: peter-evans/repository-dispatch@v4
        with:
          event-type: nightly-orchestrator-completed
          client-payload: |
            {
              "run_id": "${{ github.run_id }}",
              "triggered_at": "${{ github.event.repository.updated_at }}",
              "deploy": ${{ needs.config.outputs.deploy_dashboard }}
            }