functional_tests_common.yml

name: Common Functional Tests

on:
  workflow_call:
    inputs:
      platform:
        required: true
        type: string
        description: Platform name (e.g., cuda, default)
      device:
        required: true
        type: string
        description: Device type (e.g., a100, a800, h100, generic)
      task:
        required: true
        type: string
        description: Task name (e.g., train, hetero_train)
      model:
        required: true
        type: string
        description: Model name (e.g., aquila, mixtral, deepseek)
      case:
        required: false
        type: string
        description: Specific test case (e.g., tp2_pp2, tp4_pp2)
      image:
        required: true
        type: string
      runs_on:
        required: true
        type: string
      container_volumes:
        required: true
        type: string
      container_options:
        required: true
        type: string
      source_artifact:
        required: true
        type: string
        description: Name of the artifact containing source code

jobs:
  functional_test:
    runs-on: ${{ fromJson(inputs.runs_on) }}
    container:
      image: ${{ inputs.image }}
      ports:
        - 80
      volumes: ${{ fromJson(inputs.container_volumes) }}
      options: ${{ inputs.container_options }}

    steps:
      - name: Download source code artifact
        uses: actions/download-artifact@v4
        with:
          name: ${{ inputs.source_artifact }}
          path: /tmp

      - name: Extract source code
        shell: bash
        run: |
          mkdir -p /__w/FlagScale/FlagScale
          tar -xzf /tmp/flagscale-source.tar.gz -C /__w/FlagScale/FlagScale

      - name: Set safe directory
        shell: bash
        run: |
          git config --global --add safe.directory /__w/FlagScale/FlagScale

      - name: Run functional tests
        id: functional_test
        shell: bash
        run: |
          set -euo pipefail
          cd /__w/FlagScale/FlagScale

          PLATFORM='${{ inputs.platform }}'
          DEVICE='${{ inputs.device }}'
          TASK='${{ inputs.task }}'
          MODEL='${{ inputs.model }}'
          CASE='${{ inputs.case }}'
          PROJECT_ROOT="/__w/FlagScale/FlagScale"

          echo "Running functional tests"
          echo "Platform: $PLATFORM"
          echo "Device: $DEVICE"
          echo "Task: $TASK"
          echo "Model: $MODEL"
          echo "Case: ${CASE:-all}"
          echo "Project root: $PROJECT_ROOT"

          # Load retry functions
          source "$PROJECT_ROOT/.github/workflows/scripts/retry_functions.sh"

          # Setup backend
          source /root/miniconda3/bin/activate flagscale-train

          COMMANDS=(
            "rm -rf ${GITHUB_WORKSPACE}/Megatron-LM-FL"
            "git clone https://github.com/flagos-ai/Megatron-LM-FL.git"
            "pip install httpx multi-storage-client boto3 google-cloud-storage fastapi uvicorn griffe"
            "cd ${GITHUB_WORKSPACE}/Megatron-LM-FL && pip install --no-build-isolation . -vvv"
          )
          retry_commands 3 "${COMMANDS[@]}"

          # Ensure we're back in the project root
          cd "$PROJECT_ROOT"

          # Build run_tests.sh command
          TEST_CMD="bash $PROJECT_ROOT/tests/test_utils/runners/run_tests.sh \
            --platform $PLATFORM \
            --device $DEVICE \
            --type functional \
            --task $TASK \
            --model $MODEL"

          # Add case filter if specified
          if [ -n "$CASE" ]; then
            TEST_CMD="$TEST_CMD --list $CASE"
          fi

          # Run functional tests
          eval "$TEST_CMD"
          exit_code=$?

          if [ $exit_code -eq 0 ]; then
            echo "✅ Functional tests passed for $PLATFORM/$DEVICE/$TASK/$MODEL${CASE:+/$CASE}"
          else
            echo "❌ Functional tests failed for $PLATFORM/$DEVICE/$TASK/$MODEL${CASE:+/$CASE} (exit code: $exit_code)"
          fi

          echo "exit_code=$exit_code" >> $GITHUB_OUTPUT
          exit $exit_code
        timeout-minutes: 180