name: macOS Apple Silicon Smoke Test

on:
  schedule:
    # Daily at 2:30 AM UTC
    - cron: '30 2 * * *'
  workflow_dispatch:  # Manual trigger

permissions:
  contents: read

jobs:
  macos-m1-smoke-test:
    # macos-26 (the supported target) is still a preview runner, so gate on GA
    # macos-15 and keep macos-26 non-blocking.
    strategy:
      fail-fast: false
      matrix:
        include:
          - os: macos-15
            required: true
          - os: macos-26
            required: false
    name: macos-m1-smoke-test (${{ matrix.os }})
    runs-on: ${{ matrix.os }}
    continue-on-error: ${{ !matrix.required }}
    timeout-minutes: 30

    steps:
      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0

      - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7.6.0
        with:
          enable-cache: true
          cache-dependency-glob: |
            requirements/**/*.txt
            pyproject.toml
          python-version: '3.12'

      - name: Create virtual environment
        run: |
          uv venv
          echo "$GITHUB_WORKSPACE/.venv/bin" >> "$GITHUB_PATH"

      - name: Install dependencies and build vLLM
        run: |
          uv pip install -r requirements/build/cpu.txt --index-strategy unsafe-best-match
          uv pip install -r requirements/cpu.txt --index-strategy unsafe-best-match
          uv pip install -e . --no-build-isolation
        env:
          CMAKE_BUILD_PARALLEL_LEVEL: 4

      - name: Verify installation
        run: |
          python -c "import vllm; print(f'vLLM version: {vllm.__version__}')"

      - name: Smoke test vllm serve
        run: |
          # Start server in background
          VLLM_CPU_KVCACHE_SPACE=1 \
          vllm serve Qwen/Qwen3-0.6B \
            --max-model-len=2K \
            --load-format=dummy \
            --hf-overrides '{"num_hidden_layers": 2}' \
            --enforce-eager \
            --port 8000 &

          SERVER_PID=$!

          # Wait for server to start
          for i in {1..30}; do
            if curl -s http://localhost:8000/health > /dev/null; then
              echo "Server started successfully"
              break
            fi
            if [ "$i" -eq 30 ]; then
              echo "Server failed to start"
              kill "$SERVER_PID"
              exit 1
            fi
            sleep 2
          done

          # Test health endpoint
          curl -f http://localhost:8000/health

          # Long prompt: hits the split-KV path that short prompts skip (#46769).
          PAYLOAD=$(python -c "import json; print(json.dumps({'model': 'Qwen/Qwen3-0.6B', 'prompt': 'The quick brown fox jumps over the lazy dog. ' * 24, 'max_tokens': 16}))")
          curl -f --max-time 120 http://localhost:8000/v1/completions \
            -H "Content-Type: application/json" \
            -d "$PAYLOAD"

          # Cleanup
          kill "$SERVER_PID"