tests_link_nvidia_host_libraries.yml

# documentation: https://help.github.com/en/articles/workflow-syntax-for-github-actions
name: Test NVIDIA Host Libraries Linking
on:
  push:
  pull_request:
permissions:
  contents: read # to fetch code (actions/checkout)
jobs:
  build:
    runs-on: ubuntu-24.04
    steps:
    - name: checkout
      uses: actions/checkout@d632683dd7b4114ad314bca15554477dd762a938 # v4.2.0

    # we deliberately do not use the eessi/github-action-eessi action,
    # because we want to control when the EESSI environment is initialized
    - name: Mount EESSI CernVM-FS repository
      uses: cvmfs-contrib/github-action-cvmfs@55899ca74cf78ab874bdf47f5a804e47c198743c # v4.0
      with:
          cvmfs_config_package: https://github.com/EESSI/filesystem-layer/releases/download/latest/cvmfs-config-eessi_latest_all.deb
          cvmfs_http_proxy: DIRECT
          cvmfs_repositories: software.eessi.io

    - name: Create EESSI test directories
      run: |
        # Create EESSI directories needed for the script
        mkdir -p /tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64
        echo "EESSI_CVMFS_REPO=/tmp/eessi/cvmfs/software.eessi.io" >> $GITHUB_ENV
        echo "EESSI_CPU_FAMILY=x86_64" >> $GITHUB_ENV
        echo "EESSI_EPREFIX=/tmp/eessi/cvmfs/software.eessi.io/versions/2023.06" >> $GITHUB_ENV

    - name: Setup mock NVIDIA libraries
      run: |
        # Run the script to create mock libraries
        chmod +x ./tests/nvidia-libs/mock-nvidia-libs.sh
        ./tests/nvidia-libs/mock-nvidia-libs.sh
        
        # Add commands to PATH
        echo "PATH=/tmp/nvidia_commands:$PATH" >> $GITHUB_ENV

    - name: Setup mock nvidia-smi
      run: |
        # Create directory for mock nvidia-smi
        mkdir -p /tmp/nvidia-bin
        
        # Copy the mock script
        chmod +x ./tests/nvidia-libs/mock-nvidia-smi.sh
        cp ./tests/nvidia-libs/mock-nvidia-smi.sh /tmp/nvidia-bin/nvidia-smi
        
        # Add to PATH
        echo "PATH=/tmp/nvidia-bin:$PATH" >> $GITHUB_ENV

    - name: Test LD_PRELOAD mode
      run: |
        # Run the script with LD_PRELOAD option (shouldn't create symlinks)
        output=$(./link_nvidia_host_libraries.sh --show-ld-preload || echo "Script returned non-zero: $?")
        
        # Check for expected outputs
        echo "$output" | grep "export EESSI_GPU_COMPAT_LD_PRELOAD=" || { echo "EESSI_GPU_COMPAT_LD_PRELOAD not found in output"; exit 1; }
        echo "$output" | grep "export EESSI_GPU_LD_PRELOAD=" || { echo "EESSI_GPU_LD_PRELOAD not found in output"; exit 1; }
        echo "$output" | grep "export EESSI_OVERRIDE_GPU_CHECK=" || { echo "EESSI_OVERRIDE_GPU_CHECK not found in output"; exit 1; }
        
        # Verify that no symlinks were created
        if [ -e "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/host/driver_version.txt" ]; then
          echo "Error: symlinks were created in LD_PRELOAD mode"
          exit 1
        fi
        
        echo "LD_PRELOAD mode test passed."

    - name: Test normal run (first time)
      run: |
        # Run with verbose mode
        output=$(./link_nvidia_host_libraries.sh --verbose || echo "Script returned non-zero: $?")
        
        # Check if NVIDIA GPU was detected
        echo "$output" | grep "Found NVIDIA GPU driver version 535.129.03" || { echo "Failed to detect NVIDIA driver version"; exit 1; }
        echo "$output" | grep "Found host CUDA version 8.0" || { echo "Failed to detect CUDA version"; exit 1; }
        
        # Check if libraries were found
        echo "$output" | grep "Matched.*CUDA Libraries" || { echo "Failed to match CUDA libraries"; exit 1; }
        
        # Verify symlinks were created
        if [ ! -d "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/host" ]; then
          echo "Error: host directory wasn't created"
          exit 1
        fi
        
        # Check if version files were created
        if [ ! -f "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/host/driver_version.txt" ]; then
          echo "Error: driver_version.txt wasn't created"
          exit 1
        fi
        
        # Check driver version content
        grep "535.129.03" "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/host/driver_version.txt" || { echo "Incorrect driver version"; exit 1; }
        
        # Check if latest symlink was created
        if [ ! -L "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/latest" ]; then
          echo "Error: 'latest' symlink wasn't created"
          exit 1
        fi
        
        # Check if latest points to host
        readlink "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/latest" | grep "host" || { echo "latest doesn't point to host"; exit 1; }
        
        echo "First normal run test passed"

    - name: Test normal run (second time)
      run: |
        # Store file timestamps before second run
        stat_before=$(stat "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/host/driver_version.txt")
        
        # Run script again
        output=$(./link_nvidia_host_libraries.sh || echo "Script returned non-zero: $?")
        
        # Store file timestamps after second run
        stat_after=$(stat "/tmp/eessi/cvmfs/software.eessi.io/host_injections/nvidia/x86_64/host/driver_version.txt")
        
        # Compare timestamps - should be the same (files shouldn't be modified)
        if [[ "$stat_before" != "$stat_after" ]]; then
          echo "Error: files were modified on second run when they shouldn't have been"
          echo "Before: $stat_before"
          echo "After: $stat_after"
          exit 1
        fi
        
        # Check for message indicating that libraries are already linked
        echo "$output" | grep "have already been linked" || { echo "Missing 'already linked' message"; exit 1; }
        
        echo "Second normal run test passed"