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:
    branches:
      - '*-software.eessi.io'   # Matches any branch ending with '-software.eessi.io'
  pull_request:
    paths:
      - 'scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh'  # PR changes only relevant for this specific file
      - '.github/workflows/tests_link_nvidia_host_libraries.yml'  # Also test when changing the tests themselves
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

    - name: Initialize EESSI
      uses: eessi/github-action-eessi@v3
      
    - name: Setup mock NVIDIA libraries
      run: |
        # Run the script to create mock libraries
        chmod +x ./tests/nvidia-libs/mock-nvidia-libs.sh
        echo ">>> Running ./tests/nvidia-libs/mock-nvidia-libs.sh"
        ./tests/nvidia-libs/mock-nvidia-libs.sh

        # Create symlink to override host's ldconfig, since the script tries to use /sbin/ldconfig first.
        echo "Symlinking ldconfig to /sbin/ldconfig"
        sudo ln -sf /tmp/ldconfig/ldconfig /sbin/ldconfig
        
        # Verify the symlink was created correctly
        ls -la /sbin/ldconfig

    - 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
        echo ">>> Copying ./tests/nvidia-libs/mock-nvidia-smi.sh"
        cp ./tests/nvidia-libs/mock-nvidia-smi.sh /tmp/nvidia-bin/nvidia-smi
        
        # Add to PATH
        echo "Updating PATH"
        echo "PATH=/tmp/nvidia-bin:$PATH" >> $GITHUB_ENV

    - name: Test LD_PRELOAD mode
      run: |
        echo ">>> Testing LD_PRELOAD mode"

        # Run the script with LD_PRELOAD option (shouldn't create symlinks)
        output=$(./scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh --show-ld-preload || { echo "Script returned non-zero: $?"; echo $output; exit 1; })
        
        echo "$output"

        echo ">>> Running checks"

        # 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 "/opt/eessi/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: |
        echo ">>> Testing normal run - first time"

        # Run with verbose mode
        output=$(./scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh --verbose || { echo "Script returned non-zero: $?"; echo $output; exit 1; })

        echo "$output"

        echo ">>> Running checks"
        
        # Check if NVIDIA GPU was detected - Driver version and CUDA version are hardcoded in `tests/nvidia-libs/mock-nvidia-smi.sh`
        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 "/opt/eessi/nvidia/x86_64/host" ]; then
          echo "Error: host directory wasn't created"
          exit 1
        fi
        
        # Check if version files were created
        if [ ! -f "/opt/eessi/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" "/opt/eessi/nvidia/x86_64/host/driver_version.txt" || { echo "Incorrect driver version"; exit 1; }
        
        # Check if latest symlink was created
        if [ ! -L "/opt/eessi/nvidia/x86_64/latest" ]; then
          echo "Error: 'latest' symlink wasn't created"
          exit 1
        fi
        
        # Check if latest points to host
        readlink "/opt/eessi/nvidia/x86_64/latest" | grep "host" || { echo "latest doesn't point to host"; exit 1; }

        # Check if symlinks to libraries were created and point to correct files
        echo ">>> Checking library symlinks"

        # List dir with libraries
        echo "Showing content of /tmp/nvidia_libs"
        echo "$(ls -l /tmp/nvidia_libs)"
        echo "Showing content of /tmp/nvidia_libs_duplicate"
        echo "$(ls -l /tmp/nvidia_libs_duplicate)"
        echo "Showing content of /opt/eessi/nvidia/x86_64/host"
        echo "$(ls -l /opt/eessi/nvidia/x86_64/host)"

        # List expected library names - list of libraries is hardcoded in `tests/nvidia-libs/mock-nvidia-libs.sh`
        libraries=(
          "libcuda.so"
          "libcuda.so.1"
          "libnvidia-ml.so"
          "libnvidia-ml.so.1"
          "libnvidia-ptxjitcompiler.so"
          "libnvidia-ptxjitcompiler.so.1"
          "libcudadebugger.so"
          "libcudadebugger.so.1"
        )

        # Check each expected library symlink
        for lib in "${libraries[@]}"; do
          lib_path="/opt/eessi/nvidia/x86_64/host/$lib"
          
          # Check if the symlink exists
          if [ ! -L "$lib_path" ]; then
            echo "Error: Symlink for $lib was not created"
            exit 1
          fi
          
          # Check if symlink target exists
          target=$(readlink "$lib_path")
          if [ ! -e "$target" ]; then
            echo "Error: Symlink $lib_path points to non-existent file: $target"
            exit 1
          fi
          
          # Verify it points to our mock library in /tmp/nvidia_libs or /tmp/nvidia_libs_duplicate
          if [[ "$target" != "/tmp/nvidia_libs/$lib"* && "$target" != "/tmp/nvidia_libs_duplicate/$lib"* ]]; then
            echo "Error: Symlink $lib_path points to $target, which is not in our mock directories"
            exit 1
          fi
          
          echo ">>> Verified symlink: $lib -> $target"
        done
        
        echo "First normal run test passed"

    - name: Test normal run (second time)
      run: |
        echo ">>> Testing normal run - second time - should be idempotent"
        # Remove all write permissions on /opt/eessi so any attempts to write files fail
        chmod -R a-w /opt/eessi
        
        # Store file timestamps before second run (ignoring access time)
        stat_before=$(stat --format="%n %s %y %U %G %m %i" "/opt/eessi/nvidia/x86_64/host/driver_version.txt")
        
        # Run script again
        output=$(./scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh || { echo "Script returned non-zero: $?"; echo $output; exit 1; })
        
        echo "$output"

        echo ">>> Running checks"

        # Store file timestamps after second run (ignoring access time)
        stat_after=$(stat --format="%n %s %y %U %G %m %i" "/opt/eessi/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"