.github/workflows/trigger-mbridge-tests.yml

# Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0

name: Trigger MBridge Tests
# Remote testing of MBridge from MCore
# Triggers MBridge CI tests with current MCore commit to verify backward compatibility

on:
  # Manual trigger only
  workflow_dispatch:
    inputs:
      mbridge_ref:
        description: 'MBridge branch/ref to trigger'
        required: false
        type: string
        default: 'main'
      run_cicd_main:
        description: 'Run cicd-main.yml (full CI/CD)'
        required: false
        type: boolean
        default: true
      run_install_test:
        description: 'Run install-test.yml (quick install check)'
        required: false
        type: boolean
        default: true
      test_suite:
        description: 'Test suite to run (for cicd-main)'
        required: false
        type: choice
        options:
          - 'all'
          - 'unit-only'
          - 'functional-only'
        default: 'all'

jobs:
  # First job: Get MCore commit info (shared by all matrix jobs)
  get-mcore-info:
    runs-on: ubuntu-latest
    outputs:
      sha: ${{ steps.mcore_info.outputs.sha }}
      short_sha: ${{ steps.mcore_info.outputs.short_sha }}
      branch: ${{ steps.mcore_info.outputs.branch }}
      repo_url: ${{ steps.mcore_info.outputs.repo_url }}
    steps:
      - name: Checkout MCore
        uses: actions/checkout@v4
        with:
          fetch-depth: 0
      
      - name: Get MCore commit info
        id: mcore_info
        run: |
          echo "sha=$(git rev-parse HEAD)" >> $GITHUB_OUTPUT
          echo "short_sha=$(git rev-parse --short HEAD)" >> $GITHUB_OUTPUT
          echo "branch=${GITHUB_REF#refs/heads/}" >> $GITHUB_OUTPUT
          
          # Get repo URL from origin remote, fallback to constructing from github context
          REPO_URL=$(git remote get-url origin 2>/dev/null || echo "${{ github.server_url }}/${{ github.repository }}.git")
          echo "repo_url=${REPO_URL}" >> $GITHUB_OUTPUT

          echo "📦 MCore commit: $(git rev-parse --short HEAD)"
          echo "🌿 Branch: ${GITHUB_REF#refs/heads/}"
          echo "📍 Repo: ${REPO_URL}"

  # Matrix job: Trigger and monitor MBridge workflows in parallel
  trigger-and-monitor:
    needs: [get-mcore-info]
    runs-on: ubuntu-latest
    continue-on-error: true  # Don't fail workflow if monitoring times out
    strategy:
      fail-fast: false  # Continue other matrix jobs even if one fails
      matrix:
        include:
          - workflow: install-test.yml
            name: Install Test
          - workflow: cicd-main.yml
            name: CI/CD Main
    
    name: ${{ matrix.name }}
    
    steps:
      - name: Check if workflow should run
        id: should_run
        run: |
          if [[ "${{ matrix.workflow }}" == "install-test.yml" && "${{ inputs.run_install_test }}" == "true" ]]; then
            echo "run=true" >> $GITHUB_OUTPUT
          elif [[ "${{ matrix.workflow }}" == "cicd-main.yml" && "${{ inputs.run_cicd_main }}" == "true" ]]; then
            echo "run=true" >> $GITHUB_OUTPUT
          else
            echo "run=false" >> $GITHUB_OUTPUT
            echo "⏭️ Skipping ${{ matrix.workflow }} (not enabled)"
          fi
      
      - name: Trigger ${{ matrix.workflow }}
        if: steps.should_run.outputs.run == 'true'
        id: trigger
        env:
          GH_TOKEN: ${{ secrets.PAT }}
        run: |
          echo "🚀 Triggering ${{ matrix.workflow }} | MCore: ${{ needs.get-mcore-info.outputs.short_sha }} | MBridge: ${{ inputs.mbridge_ref }}"
          
          gh workflow run ${{ matrix.workflow }} \
            --repo NVIDIA-NeMo/Megatron-Bridge --ref ${{ inputs.mbridge_ref }} \
            --field mcore_commit=${{ needs.get-mcore-info.outputs.sha }} \
            --field mcore_branch=${{ needs.get-mcore-info.outputs.branch }} \
            --field mcore_repo=${{ needs.get-mcore-info.outputs.repo_url }} \
            --field test_suite=${{ inputs.test_suite }} \
            --field triggered_by=mcore-ci
      
      - name: Get run ID
        if: steps.should_run.outputs.run == 'true'
        id: get_run_id
        env:
          GH_TOKEN: ${{ secrets.PAT }}
        run: |
          sleep 10  # Wait for run to appear
          RUN_ID=$(gh run list \
            --repo NVIDIA-NeMo/Megatron-Bridge \
            --workflow=${{ matrix.workflow }} \
            --limit 5 \
            --json databaseId,createdAt \
            --jq "sort_by(.createdAt) | reverse | .[0] | .databaseId")
          
          echo "run_id=${RUN_ID}" >> $GITHUB_OUTPUT
          echo "📋 Run ID: ${RUN_ID}"
          
          cat >> $GITHUB_STEP_SUMMARY << EOF
          ## 🔄 ${{ matrix.name }} Triggered
          
          **MCore:** \`${{ needs.get-mcore-info.outputs.short_sha }}\` | **MBridge:** \`${{ inputs.mbridge_ref }}\` | **Suite:** \`${{ inputs.test_suite }}\`
          
          - 🔄 [${{ matrix.workflow }}](https://github.com/NVIDIA-NeMo/Megatron-Bridge/actions/runs/${RUN_ID}) - Running...
          - ⏳ Monitoring every 5 minutes until completion
          
          > **Note:** Tests run without approval when triggered from MCore
          EOF
      
      - name: Monitor workflow
        if: steps.should_run.outputs.run == 'true'
        id: monitor
        continue-on-error: true
        env:
          GH_TOKEN: ${{ secrets.PAT }}
        run: |
          RUN_ID="${{ steps.get_run_id.outputs.run_id }}"
          echo "📊 Monitoring ${{ matrix.workflow }} (Run ID: ${RUN_ID})"
          
          gh run watch ${RUN_ID} --repo NVIDIA-NeMo/Megatron-Bridge --exit-status
          
          CONCLUSION=$(gh run view ${RUN_ID} --repo NVIDIA-NeMo/Megatron-Bridge --json conclusion --jq -r .conclusion)
          echo "workflow_status=${CONCLUSION}" >> $GITHUB_ENV
          echo "✅ Completed: ${CONCLUSION}"
      
      - name: Report results
        if: always() && steps.should_run.outputs.run == 'true'
        run: |
          CONCLUSION="${{ env.workflow_status || 'unknown' }}"
          RUN_ID="${{ steps.get_run_id.outputs.run_id }}"
          
          case "$CONCLUSION" in
            "success") ICON="✅"; MSG="passed" ;;
            "failure") ICON="❌"; MSG="failed"; EXIT_CODE=1 ;;
            "cancelled") ICON="🚫"; MSG="cancelled"; EXIT_CODE=0 ;;
            *) ICON="⏳"; MSG="still running or timed out"; EXIT_CODE=0 ;;
          esac
          
          cat >> $GITHUB_STEP_SUMMARY << EOF
          ## 📊 ${{ matrix.name }} Results
          
          ### ${ICON} ${{ matrix.workflow }}
          **Status:** \`${CONCLUSION}\`
          
          [View full results →](https://github.com/NVIDIA-NeMo/Megatron-Bridge/actions/runs/${RUN_ID})
          
          ---
          *Triggered from MCore \`${{ needs.get-mcore-info.outputs.short_sha }}\`*
          EOF
          
          echo "${ICON} ${{ matrix.name }} ${MSG}"
          exit ${EXIT_CODE:-0}