test.yml

name: Integration Tests

on:
  pull_request:
    branches: [ main ]
  push:
    branches: [ main ]

jobs:
  integration-test:
    runs-on: ubuntu-latest

    permissions:
      # Allow the workflow to commit the auto-generated
      # test/lyrics_expected_gte_512.json back to main on first run.
      contents: write

    services:
      postgres:
        image: postgres:15-alpine
        env:
          POSTGRES_USER: postgres
          POSTGRES_PASSWORD: postgres
          POSTGRES_DB: audiomuse_migration_test
        ports:
          - 5432:5432
        options: >-
          --health-cmd pg_isready
          --health-interval 10s
          --health-timeout 5s
          --health-retries 5

    steps:
    - uses: actions/checkout@v4
      with:
        # Need a real (non-detached) checkout so we can push back the
        # auto-recorded test/lyrics_expected_gte_512.json. On pull_request events
        # we explicitly check out the PR head branch (not the merge ref)
        # so `git push` lands on the PR branch and the change becomes
        # visible in the PR before merge.
        token: ${{ secrets.GITHUB_TOKEN }}
        fetch-depth: 0
        ref: ${{ github.event.pull_request.head.ref || github.ref }}
        repository: ${{ github.event.pull_request.head.repo.full_name || github.repository }}

    - name: Set up Python
      uses: actions/setup-python@v5
      with:
        python-version: '3.11'

    - name: Install system dependencies
      run: |
        sudo apt-get update
        sudo apt-get install -y ffmpeg

    - name: Install Python dependencies
      run: |
        python -m pip install --upgrade pip
        pip install -r test/requirements.txt

    - name: Download models
      run: |
        dclap_url="https://github.com/NeptuneHub/AudioMuse-AI-DCLAP/releases/download/v1"
        base_url="https://github.com/NeptuneHub/AudioMuse-AI/releases/download/v4.0.0-model"
        mkdir -p test/models
        echo "Downloading DCLAP audio model..."
        curl -L "${dclap_url}/model_epoch_36.onnx" -o "test/models/model_epoch_36.onnx"
        echo "Downloading DCLAP audio model data..."
        curl -L "${dclap_url}/model_epoch_36.onnx.data" -o "test/models/model_epoch_36.onnx.data"
        echo "Downloading CLAP text model..."
        curl -L "${base_url}/clap_text_model.onnx" -o "test/models/clap_text_model.onnx"
        echo "Downloading MusiCNN embedding model..."
        curl -L "${base_url}/musicnn_embedding.onnx" -o "test/models/musicnn_embedding.onnx"
        echo "Downloading MusiCNN prediction model..."
        curl -L "${base_url}/musicnn_prediction.onnx" -o "test/models/musicnn_prediction.onnx"
        echo "Models downloaded successfully"
        ls -lh test/models/

    - name: Download HuggingFace tokenizer cache (release v4.0.0-model)
      run: |
        # The CLAP analyzer loads roberta-base via AutoTokenizer.from_pretrained(..., local_files_only=True).
        # Reuse the same huggingface_models.tar.gz bundled in the GitHub release that the
        # Dockerfile uses, and extract it into an HF cache directory under test/.
        base_url="https://github.com/NeptuneHub/AudioMuse-AI/releases/download/v4.0.0-model"
        hf_cache_dir="${GITHUB_WORKSPACE}/test/.hf_cache"
        mkdir -p "$hf_cache_dir"
        echo "Downloading huggingface_models.tar.gz..."
        curl -L "${base_url}/huggingface_models.tar.gz" -o /tmp/huggingface_models.tar.gz
        echo "Extracting into ${hf_cache_dir}..."
        tar -xzf /tmp/huggingface_models.tar.gz -C "$hf_cache_dir"
        rm -f /tmp/huggingface_models.tar.gz
        if [ ! -d "$hf_cache_dir/hub" ]; then
          echo "ERROR: extraction did not produce a hub/ directory" >&2
          ls -lah "$hf_cache_dir"
          exit 1
        fi
        echo "HF cache ready:"
        du -sh "$hf_cache_dir"
        # Expose HF_HOME to subsequent steps so transformers resolves the cache offline.
        echo "HF_HOME=$hf_cache_dir" >> "$GITHUB_ENV"

    - name: Download gte model bundle (release v4.0.0-model)
      run: |
        # The Dockerfile downloads lyrics_model_gte_vnni.tar.gz from the
        # v4.0.0-model release and extracts it flat into /app/model/. After
        # extraction the layout is:
        #   <models>/gte-multilingual-base-int8.onnx      (INT8 ONNX weights)
        #   <models>/gte-multilingual-base/tokenizer.json (+ tokenizer files)
        # We mirror the same layout under test/models/ for the integration
        # test, which loads via lyrics/gte_onnx.py.
        base_url="https://github.com/NeptuneHub/AudioMuse-AI/releases/download/v4.0.0-model"
        mkdir -p test/models
        if [ ! -f test/models/gte-multilingual-base-int8.onnx ] || [ ! -f test/models/gte-multilingual-base/tokenizer.json ]; then
          echo "Downloading lyrics_model_gte_vnni.tar.gz..."
          curl -L "${base_url}/lyrics_model_gte_vnni.tar.gz" -o /tmp/lyrics_model_gte_vnni.tar.gz
          tar -xzf /tmp/lyrics_model_gte_vnni.tar.gz -C test/models
          rm -f /tmp/lyrics_model_gte_vnni.tar.gz
        fi
        echo "test/models/ after extraction:"
        ls -la test/models/
        echo "gte-multilingual-base/ contents:"
        ls -la test/models/gte-multilingual-base/ || true
        echo "gte model checksums (sha256):"
        sha256sum test/models/gte-multilingual-base-int8.onnx test/models/gte-multilingual-base/tokenizer.json || true
        echo "LYRICS_MODEL_DIR=${GITHUB_WORKSPACE}/test/models" >> "$GITHUB_ENV"
        echo "LYRICS_GTE_ONNX_PATH=${GITHUB_WORKSPACE}/test/models/gte-multilingual-base-int8.onnx" >> "$GITHUB_ENV"
        echo "LYRICS_GTE_TOKENIZER_DIR=${GITHUB_WORKSPACE}/test/models/gte-multilingual-base" >> "$GITHUB_ENV"

    - name: Verify test files exist
      run: |
        echo "Checking for config.py..."
        ls -la config.py
        echo ""
        echo "Checking for ONNX models in test/models/..."
        ls -la test/models/
        echo ""
        echo "Checking for audio files in test/songs/..."
        ls -la test/songs/
        echo ""
        echo "Runner CPU INT8 path (vnni => int32 accumulation, none => int16/VPMADDUBSW):"
        grep -o 'avx512_vnni\|avx_vnni' /proc/cpuinfo | sort -u | head -1 || echo "none (non-VNNI)"

    - name: Run integration tests
      run: |
        # Run the MusiCNN integration test
        pytest test/integration/test_analysis_integration.py -s -v --tb=short

        # Run the CLAP integration test
        pytest test/integration/test_clap_analysis_integration.py -s -v --tb=short

    - name: Run provider migration integration test
      env:
        AUDIOMUSE_TEST_DATABASE_URL: postgresql://postgres:postgres@localhost:5432/audiomuse_migration_test
      run: |
        pytest test/integration/test_provider_migration_integration.py -s -v --tb=short

    - name: Run app endpoint and auth integration tests
      env:
        AUDIOMUSE_TEST_DATABASE_URL: postgresql://postgres:postgres@localhost:5432/audiomuse_migration_test
      run: |
        pytest test/integration/test_app_endpoints_integration.py test/integration/test_auth_users_integration.py -m integration -s -v --tb=short

    - name: Run lyrics integration test
      run: |
        # On first run test/lyrics_expected_gte_512.json does not exist yet: the
        # test enters RECORD mode automatically, writes the file and exits
        # successfully. The next step then commits it back to main so
        # subsequent runs pin against it (cosine similarity >= 0.99).
        pytest test/integration/test_lyrics_analysis_integration.py -s -v --tb=short

    - name: Commit recorded lyrics expected vectors
      # Push back the auto-recorded vectors on:
      #   * push to main
      #   * pull_request originating from the SAME repository (forks can't
      #     get write permission via GITHUB_TOKEN, so we skip them).
      if: |
        (github.event_name == 'push' && github.ref == 'refs/heads/main') ||
        (github.event_name == 'pull_request' &&
         github.event.pull_request.head.repo.full_name == github.repository)
      run: |
        if [ ! -f test/lyrics_expected_gte_512.json ]; then
          echo "No test/lyrics_expected_gte_512.json produced - nothing to commit."
          exit 0
        fi
        # Only commit if the file is new or actually changed.
        if git diff --quiet -- test/lyrics_expected_gte_512.json && \
           ! git status --porcelain test/lyrics_expected_gte_512.json | grep -q '^??'; then
          echo "test/lyrics_expected_gte_512.json is unchanged - skipping commit."
          exit 0
        fi
        git config user.name  "github-actions[bot]"
        git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
        git add test/lyrics_expected_gte_512.json
        git commit -m "test(lyrics): record expected vectors for integration test"
        if [ "${{ github.event_name }}" = "pull_request" ]; then
          target_branch="${{ github.event.pull_request.head.ref }}"
        else
          target_branch="main"
        fi
        echo "Pushing to branch: $target_branch"
        git push origin "HEAD:${target_branch}"

    - name: Upload test results
      if: always()
      uses: actions/upload-artifact@v4
      with:
        name: integration-test-results
        path: |
          test/*.log
          test/temp/
        retention-days: 7