test-ci.yml

name: CI Tests

on:
  pull_request:
  workflow_dispatch:
    inputs:
      use_local_images:
        description: 'Build images locally instead of pulling from DockerHub'
        required: false
        type: boolean
        default: true

env:
  NODE_VERSION: "22"
  PYTHON_VERSION: "3.13"
  PLAYWRIGHT_BROWSERS_PATH: "ms-playwright"
  PLAYWRIGHT_VERSION: "1.57.0"
  OPENSEARCH_PASSWORD: ${{ vars.OPENSEARCH_PASSWORD || secrets.OPENSEARCH_PASSWORD || 'OpenRag#2025!' }}
  OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
  LANGFLOW_AUTO_LOGIN: "True"
  LANGFLOW_NEW_USER_IS_ACTIVE: "True"
  LANGFLOW_ENABLE_SUPERUSER_CLI: "True"
  LANGFLOW_CHAT_FLOW_ID: ${{ vars.LANGFLOW_CHAT_FLOW_ID || '1098eea1-6649-4e1d-aed1-b77249fb8dd0' }}
  LANGFLOW_INGEST_FLOW_ID: ${{ vars.LANGFLOW_INGEST_FLOW_ID || '5488df7c-b93f-4f87-a446-b67028bc0813' }}
  NUDGES_FLOW_ID: ${{ vars.NUDGES_FLOW_ID || 'ebc01d31-1976-46ce-a385-b0240327226c' }}
  CI_IMAGE_TAG: ci-${{ github.sha }}
  ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
  WATSONX_API_KEY: ${{ secrets.WATSONX_API_KEY }}
  WATSONX_ENDPOINT: ${{ secrets.WATSONX_ENDPOINT }}
  WATSONX_PROJECT_ID: ${{ secrets.WATSONX_PROJECT_ID }}
  OLLAMA_ENDPOINT: ${{ secrets.OLLAMA_ENDPOINT }}
  GOOGLE_OAUTH_CLIENT_ID: ${{ secrets.GOOGLE_OAUTH_CLIENT_ID }}
  GOOGLE_OAUTH_CLIENT_SECRET: ${{ secrets.GOOGLE_OAUTH_CLIENT_SECRET }}
  AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
  AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
  LANGFLOW_URL_INGEST_FLOW_ID: ${{ vars.LANGFLOW_URL_INGEST_FLOW_ID || '72c3d17c-2dac-4a73-b48a-6518473d7830' }}
  # Single-node test cluster never reaches the multi-node readiness counts
  OPENSEARCH_NODE_COUNT_CHECK_ENABLED: "false"

jobs:
  changes:
    runs-on: ubuntu-latest
    outputs:
      e2e: ${{ steps.filter.outputs.e2e }}
      integration: ${{ steps.filter.outputs.integration }}
    steps:
      - name: Checkout
        uses: actions/checkout@v4

      - name: Detect changes
        uses: dorny/paths-filter@v3
        id: filter
        with:
          filters: |
            e2e:
              - '**'
              - '!docs/**'
              - '!**/*.md'
            integration:
              - '**'
              - '!docs/**'
              - '!**/*.md'

  build-images:
    needs: changes
    if: ${{ needs.changes.outputs.e2e == 'true' || needs.changes.outputs.integration == 'true' || github.event_name == 'workflow_dispatch' }}
    runs-on:
      labels: ["self-hosted", "linux", "ARM64", "langflow-ai-arm64-40gb-ephemeral-sudo"]
    steps:
      - run: df -h

      - name: Cleanup Docker cache
        run: |
          docker compose -f docker-compose.yml down -v --remove-orphans || true

      - name: Cleanup root-owned files (OpenSearch data, config, Langflow data, keys, data, flows, documents)
        run: |
          for i in 1 2 3; do
            docker run --rm -v $(pwd):/work alpine sh -c "rm -rf /work/opensearch-data /work/config /work/langflow-data /work/keys /work/data /work/flows /work/openrag-documents" && break
            echo "Attempt $i failed, retrying in 5s..."
            sleep 5
          done || true

      - run: df -h

      - name: Checkout
        uses: actions/checkout@v4

      - name: Verify workspace
        run: |
          echo "Current directory: $(pwd)"
          echo "Workspace: ${GITHUB_WORKSPACE}"
          ls -la

      - name: Build OpenRAG images
        run: |
          USE_LOCAL="${{ inputs.use_local_images }}"
          for i in 1 2 3; do
            echo "Build attempt $i..."
            if [ "${{ github.event_name }}" == "pull_request" ] || [ "$USE_LOCAL" != "false" ]; then
              echo "Building PR images locally..."
              if OPENRAG_VERSION="${CI_IMAGE_TAG}" make ci-build-images; then
                break
              fi
            else
              echo "Preparing DockerHub images..."
              if docker pull langflowai/openrag-backend:latest && \
                 docker pull langflowai/openrag-frontend:latest && \
                 docker pull langflowai/openrag-langflow:latest && \
                 docker tag langflowai/openrag-backend:latest "langflowai/openrag-backend:${CI_IMAGE_TAG}" && \
                 docker tag langflowai/openrag-frontend:latest "langflowai/openrag-frontend:${CI_IMAGE_TAG}" && \
                 docker tag langflowai/openrag-langflow:latest "langflowai/openrag-langflow:${CI_IMAGE_TAG}" && \
                 docker build --no-cache -t "langflowai/openrag-opensearch:${CI_IMAGE_TAG}" -f Dockerfile .; then
                break
              fi
            fi
            echo "Build attempt $i failed. Retrying in 10 seconds..."
            sleep 10
            if [ $i -eq 3 ]; then
              echo "Build failed after 3 attempts."
              exit 1
            fi
          done

      - name: Save OpenRAG images
        run: OPENRAG_VERSION="${CI_IMAGE_TAG}" make ci-save-images

      - name: Upload OpenRAG image artifact
        uses: actions/upload-artifact@v4
        with:
          name: openrag-ci-images-${{ github.sha }}
          path: .ci-artifacts/openrag-ci-images.tar
          retention-days: 1
          compression-level: 0

  test-suite:
    needs: [changes, build-images]
    if: ${{ needs.changes.outputs.integration == 'true' || github.event_name == 'workflow_dispatch' }}
    runs-on:
      labels: ["self-hosted", "linux", "ARM64", "langflow-ai-arm64-40gb-ephemeral-sudo"]
    strategy:
      fail-fast: false
      matrix:
        suite: [core, sdk-python, sdk-typescript]
    name: tests (${{ matrix.suite }})
    steps:
      - run: df -h

      - name: Cleanup Docker cache
        run: |
          docker compose -f docker-compose.yml down -v --remove-orphans || true

      - name: Cleanup root-owned files (OpenSearch data, config, Langflow data, keys, data, flows, documents)
        run: |
          for i in 1 2 3; do
            docker run --rm -v $(pwd):/work alpine sh -c "rm -rf /work/opensearch-data /work/config /work/langflow-data /work/keys /work/data /work/flows /work/openrag-documents" && break
            echo "Attempt $i failed, retrying in 5s..."
            sleep 5
          done || true

      - run: df -h

      - name: Checkout
        uses: actions/checkout@v4

      - name: Verify workspace
        run: |
          echo "Current directory: $(pwd)"
          echo "Workspace: ${GITHUB_WORKSPACE}"
          ls -la

      - name: Set up UV
        uses: astral-sh/setup-uv@v3
        with:
          version: latest

      - name: Set up Node.js
        if: matrix.suite == 'sdk-typescript'
        uses: actions/setup-node@v4
        with:
          node-version: '20'

      - name: Python version
        run: uv python install 3.13

      - name: Download OpenRAG image artifact
        uses: actions/download-artifact@v4
        with:
          name: openrag-ci-images-${{ github.sha }}
          path: .ci-artifacts

      - name: Load OpenRAG images
        run: docker load -i .ci-artifacts/openrag-ci-images.tar

      - name: Run ${{ matrix.suite }} integration suite
        env:
          OPENSEARCH_HOST: localhost
          OPENSEARCH_PORT: 9200
          OPENSEARCH_USERNAME: admin
          OPENSEARCH_PASSWORD: ${{ env.OPENSEARCH_PASSWORD }}
          LOG_LEVEL: DEBUG
          # Force no-auth mode so tests bypass OAuth
          GOOGLE_OAUTH_CLIENT_ID: ""
          GOOGLE_OAUTH_CLIENT_SECRET: ""
          # Disable startup ingest noise unless a test enables it
          DISABLE_STARTUP_INGEST: "true"
        run: |
          chmod +x scripts/ci/run_integration_suite.sh
          OPENRAG_VERSION="${CI_IMAGE_TAG}" TEST_SUITE="${{ matrix.suite }}" make test-ci-suite
          echo "Keys directory after tests:"
          ls -la keys/ || echo "No keys directory"

      - name: Upload service logs
        uses: actions/upload-artifact@v4
        if: failure()
        with:
          name: service-logs-${{ matrix.suite }}
          path: service-logs/
          retention-days: 7

  e2e-run:
    needs: [changes, build-images]
    if: ${{ needs.changes.outputs.e2e == 'true' || github.event_name == 'workflow_dispatch' }}
    runs-on:
      labels: ["self-hosted", "linux", "ARM64", "langflow-ai-arm64-40gb-ephemeral-sudo"]
    steps:
      - name: Cleanup Docker cache
        run: |
          docker compose -f docker-compose.yml down -v --remove-orphans || true

      - name: Cleanup root-owned files (OpenSearch data, config, Langflow data, keys, data, flows, documents)
        run: |
          for i in 1 2 3; do
            docker run --rm -v $(pwd):/work alpine sh -c "rm -rf /work/opensearch-data /work/config /work/langflow-data /work/keys /work/data /work/flows /work/openrag-documents" && break
            echo "Attempt $i failed, retrying in 5s..."
            sleep 5
          done || true

      - name: Checkout
        uses: actions/checkout@v4

      - name: Set up UV
        uses: astral-sh/setup-uv@v3
        with:
          version: latest

      - name: Set up Node.js
        uses: actions/setup-node@v4
        with:
          node-version: ${{ env.NODE_VERSION }}

      - name: Install Python
        run: uv python install 3.13

      - name: Install frontend dependencies
        working-directory: frontend
        run: npm ci

      - name: Install Playwright Browsers
        working-directory: frontend
        run: |
          for i in 1 2 3; do
            echo "Installing Playwright browsers (attempt $i)..."
            if npx playwright install --with-deps chromium; then
              break
            fi
            echo "Attempt $i failed. Retrying in 10 seconds..."
            sleep 10
            if [ $i -eq 3 ]; then
              echo "Failed to install Playwright browsers after 3 attempts."
              exit 1
            fi
          done

      - name: Download OpenRAG image artifact
        uses: actions/download-artifact@v4
        with:
          name: openrag-ci-images-${{ github.sha }}
          path: .ci-artifacts

      - name: Load OpenRAG images
        run: docker load -i .ci-artifacts/openrag-ci-images.tar

      - name: Setup E2E infrastructure
        run: |
          chmod +x scripts/setup-e2e.sh
          OPENRAG_VERSION="${CI_IMAGE_TAG}" ./scripts/setup-e2e.sh

      - name: Run Playwright tests
        working-directory: frontend
        env:
          CI: "true"
          OPENSEARCH_HOST: localhost
          OPENSEARCH_PORT: "9200"
          OPENSEARCH_USERNAME: admin
          OPENSEARCH_PASSWORD: ${{ env.OPENSEARCH_PASSWORD }}
          GOOGLE_OAUTH_CLIENT_ID: ""
          GOOGLE_OAUTH_CLIENT_SECRET: ""
        run: npx playwright test

      - name: Collect service logs on failure
        if: failure()
        run: |
          mkdir -p service-logs
          docker logs os > service-logs/opensearch.log 2>&1 || true
          docker logs openrag-backend > service-logs/backend.log 2>&1 || true
          docker logs langflow > service-logs/langflow.log 2>&1 || true
          docker logs openrag-backend-proxy > service-logs/backend-proxy.log 2>&1 || true

      - name: Upload service logs
        uses: actions/upload-artifact@v4
        if: failure()
        with:
          name: service-logs
          path: service-logs/
          retention-days: 7

      - name: Upload Playwright report
        uses: actions/upload-artifact@v4
        if: ${{ !cancelled() }}
        with:
          name: playwright-report
          path: frontend/playwright-report/
          retention-days: 14

      - name: Upload test results
        uses: actions/upload-artifact@v4
        if: ${{ !cancelled() }}
        with:
          name: playwright-test-results
          path: frontend/test-results/
          retention-days: 7

      - name: Teardown infrastructure
        if: always()
        run: |
          make docling-stop || true
          docker rm -f openrag-backend-proxy 2>/dev/null || true
          make clean || true
          docker system prune -f || true

  tests:
    needs: [changes, build-images, test-suite]
    if: always()
    runs-on:
      labels: ["self-hosted", "linux", "ARM64", "langflow-ai-arm64-40gb-ephemeral-sudo"]
    steps:
      - name: Check split integration result
        run: |
          if [ "${{ needs.changes.outputs.integration }}" = "true" ] || [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
            if [ "${{ needs.build-images.result }}" != "success" ]; then
              echo "build-images failed"
              exit 1
            fi
            if [ "${{ needs.test-suite.result }}" != "success" ]; then
              echo "one or more integration suites failed"
              exit 1
            fi
          fi
          echo "all integration suites passed or skipped"

  e2e:
    needs: [changes, build-images, e2e-run]
    if: always()
    runs-on:
      labels: ["self-hosted", "linux", "ARM64", "langflow-ai-arm64-40gb-ephemeral-sudo"]
    steps:
      - name: Check E2E result
        run: |
          if [ "${{ needs.changes.outputs.e2e }}" = "true" ] || [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
            if [ "${{ needs.build-images.result }}" != "success" ]; then
              echo "build-images failed"
              exit 1
            fi
            if [ "${{ needs.e2e-run.result }}" != "success" ]; then
              echo "E2E tests failed"
              exit 1
            fi
          fi
          echo "E2E tests passed or skipped"