nightly_status_collector.yml

name: Nightly Status Collector

on:
  # Run after the orchestrator completes (triggered by repository dispatch)
  repository_dispatch:
    types: [nightly-orchestrator-completed]
  # Also run on schedule to catch any missed updates
  schedule:
    - cron: '0 18 * * *' # 6PM UTC (nightly starts at 6AM UTC, usually done by then)
  # Manual trigger for backfills
  workflow_dispatch:
    inputs:
      days:
        description: 'Number of days to collect (for backfill)'
        type: number
        default: 7
      force:
        description: 'Force recollection even if data already exists'
        type: boolean
        default: false
      skip-deploy:
        description: 'Collect status but skip deploying to gh-pages (dry-run)'
        type: boolean
        default: false

permissions:
  contents: write
  actions: read

jobs:
  collect-status:
    runs-on: ubuntu-latest
    outputs:
      should-deploy: ${{ steps.resolve-deploy.outputs.deploy }}
    steps:
      - name: Resolve deploy flag
        id: resolve-deploy
        run: |
          # repository_dispatch: read from client payload
          # workflow_dispatch: read from inputs
          # schedule: always deploy
          if [ "${{ github.event_name }}" = "repository_dispatch" ]; then
            DEPLOY="${{ github.event.client_payload.deploy }}"
          elif [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
            if [ "${{ inputs.skip-deploy }}" = "true" ]; then
              DEPLOY="false"
            else
              DEPLOY="true"
            fi
          else
            DEPLOY="true"
          fi
          echo "deploy=${DEPLOY:-true}" >> "$GITHUB_OUTPUT"
          echo "Deploy dashboard: ${DEPLOY:-true}"

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

      - name: Checkout gh-pages branch for existing data
        uses: actions/checkout@v4
        with:
          ref: gh-pages
          path: gh-pages-data
        continue-on-error: true

      - name: Install uv
        uses: astral-sh/setup-uv@v4
        with:
          enable-cache: false

      - name: Collect nightly status
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
        run: |
          DAYS="${{ inputs.days || 7 }}"

          # Copy existing history if available
          mkdir -p nightly-status
          if [ -f "gh-pages-data/nightly-status/history.json" ]; then
            cp gh-pages-data/nightly-status/history.json nightly-status/history.json
            echo "Loaded existing history"
          fi

          # Extract repo name from github.repository (format: owner/repo)
          REPO_NAME="${{ github.repository }}"
          REPO_NAME="${REPO_NAME#*/}"

          # Build force flag if requested
          FORCE_FLAG=""
          if [ "${{ inputs.force }}" = "true" ]; then
            FORCE_FLAG="--force"
          fi

          # Run the collector script with inline dependencies
          uv run --no-project --with requests --script .github/scripts/collect_nightly_status.py \
            --owner "${{ github.repository_owner }}" \
            --repo "$REPO_NAME" \
            --days "$DAYS" \
            --output-dir nightly-status \
            --history-file nightly-status/history.json \
            $FORCE_FLAG

      - name: Prepare dashboard files
        run: |
          # Copy dashboard HTML
          cp .github/nightly-dashboard/index.html nightly-status/

      - name: Deploy nightly-status to GitHub Pages
        if: steps.resolve-deploy.outputs.deploy == 'true'
        uses: peaceiris/actions-gh-pages@v4
        with:
          github_token: ${{ secrets.GITHUB_TOKEN }}
          publish_dir: ./nightly-status
          destination_dir: nightly-status
          keep_files: true
          commit_message: 'Update nightly status dashboard'

      - name: Dry-run summary
        if: steps.resolve-deploy.outputs.deploy != 'true'
        run: |
          echo "### Dry-run: skipped dashboard deployment" >> "$GITHUB_STEP_SUMMARY"
          echo "Collected status data is available in the workflow artifacts below." >> "$GITHUB_STEP_SUMMARY"
          cat nightly-status/history.json | head -50 >> "$GITHUB_STEP_SUMMARY"

      - name: Upload collected data (artifact)
        uses: actions/upload-artifact@v4
        with:
          name: nightly-status-data
          path: nightly-status/
          retention-days: 7

  notify-if-degraded:
    runs-on: ubuntu-latest
    needs: collect-status
    steps:
      - name: Checkout for data
        uses: actions/checkout@v4
        with:
          ref: gh-pages
          path: gh-pages-data
        continue-on-error: true

      - name: Check for degraded status
        id: check
        run: |
          if [ ! -f "gh-pages-data/nightly-status/history.json" ]; then
            echo "No history file found, skipping notification check"
            exit 0
          fi

          # Check if success rate dropped below 70% in last 7 days
          SUCCESS_RATE=$(jq -r '.stats.last_7_days.success_rate // 100' gh-pages-data/nightly-status/history.json)

          if [ "$SUCCESS_RATE" != "null" ] && [ "$(echo "$SUCCESS_RATE < 70" | bc -l)" -eq 1 ]; then
            echo "degraded=true" >> "$GITHUB_OUTPUT"
            echo "success_rate=$SUCCESS_RATE" >> "$GITHUB_OUTPUT"
          else
            echo "degraded=false" >> "$GITHUB_OUTPUT"
          fi

      - name: Post warning (if degraded)
        if: steps.check.outputs.degraded == 'true'
        run: |
          echo "::warning::Nightly workflow success rate has dropped to ${{ steps.check.outputs.success_rate }}% (below 70% threshold)"