Stirling-PDF/.github/workflows/build-enterprise.yml

name: Enterprise E2E (Playwright)

# Enterprise Playwright suite — exercises premium-key gated features (audit,
# teams, analytics) plus full OAuth + SAML logins via the Keycloak compose
# stacks under testing/compose. Slow and secret-gated, so it runs in four
# situations:
#
#   - PRs that touch proprietary / premium / SSO compose / enterprise tests
#     (driven by build.yml via workflow_call, path-filtered against
#     .github/config/.files.yaml `proprietary`),
#   - every push to main (post-merge safety net),
#   - on a nightly cron schedule (catches Keycloak image drift, license
#     expiry, upstream proprietary changes),
#   - manual workflow_dispatch.

on:
  workflow_call:
    inputs:
      use_shared_cache:
        required: false
        type: boolean
        default: false
  push:
    branches: ["main"]
  schedule:
    - cron: "0 4 * * *"
  workflow_dispatch:

# No `concurrency:` block here on purpose. When this workflow is called via
# workflow_call from build.yml, ${{ github.workflow }}/event_name/pr_number
# resolve to the *caller's* values, producing the same group key as build.yml
# and causing the workflow_call instantiation to self-cancel — the job
# silently fails to spawn while `${{ needs.X.result }}` still reports
# `failure`. Standalone runs (push-to-main, nightly cron, dispatch) don't
# overlap often enough to need explicit concurrency control.

permissions:
  contents: read

jobs:
  pick:
    uses: ./.github/workflows/_runner-pick.yml

  playwright-e2e-enterprise:
    environment:
      name: ci-unsigned
      deployment: false
    needs: pick
    # Skip on fork PRs / untrusted authors: they have no PREMIUM_KEY_ENTERPRISE,
    # so the suite can't boot premium and would fail. See the header comment.
    # GitHub reports the skipped reusable workflow as success.
    if: needs.pick.outputs.is_fork != 'true'
    runs-on: ubuntu-latest
    timeout-minutes: 45
    env:
      PREMIUM_KEY: ${{ secrets.PREMIUM_KEY_ENTERPRISE }}
      PREMIUM_ENABLED: "true"
      SYSTEM_ENABLEANALYTICS: "false"
    steps:
      - name: Harden Runner
        uses: step-security/harden-runner@b09bb98e06d4d774595224525879c09bc6e98c40 # v2.20.1
        with:
          egress-policy: audit
      - name: Checkout repository
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2

      - name: Restore cache Gradle User Home
        if: inputs.use_shared_cache
        uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
        with:
          path: |
            ~/.gradle/caches
            ~/.gradle/wrapper
          key: gradle-v1-${{ runner.os }}-${{ runner.arch }}-jdk-25-${{ hashFiles('gradle/wrapper/gradle-wrapper.properties', 'gradle/libs.versions.toml', 'buildSrc/**', 'settings.gradle', 'build.gradle', 'app/**/build.gradle', 'gradle/**/*.gradle') }}

      - name: Restore cache Gradle
        if: inputs.use_shared_cache == false
        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
        with:
          path: |
            ~/.gradle/caches
            ~/.gradle/wrapper
          key: gradle-playwright-e2e-v1-${{ runner.os }}-${{ runner.arch }}-jdk-25-${{ hashFiles('gradle/wrapper/gradle-wrapper.properties', 'gradle/libs.versions.toml', 'buildSrc/**', 'settings.gradle', 'build.gradle', 'app/**/build.gradle', 'gradle/**/*.gradle') }}

      - name: Set up JDK 25
        uses: actions/setup-java@b6effb05e454b25005698d916606bdc6ffcbf961 # v5.7.0
        with:
          java-version: "25"
          distribution: "temurin"

      - name: Set up Node.js
        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
        with:
          node-version: "22"
          cache: "npm"
          cache-dependency-path: frontend/package-lock.json
      - name: Install Task
        uses: go-task/setup-task@01a4adf9db2d14c1de7a560f09170b6e0df736aa # v2.1.0
      - name: Install Playwright (chromium only)
        run: task e2e:install -- chromium
      - name: Build frontend (needed for playwright's vite preview webServer)
        # Enterprise tests target :8080 (Spring Boot's bundled frontend), but
        # playwright's webServer config still launches `vite preview --port 5173`
        # before any test run, which needs dist/ to exist. VITE_BUILD_FOR_PREVIEW
        # forces absolute asset paths so vite preview can serve deep SPA routes.
        env:
          VITE_BUILD_FOR_PREVIEW: "1"
        run: task frontend:build

      - name: Resolve kubernetes.docker.internal to localhost
        # The compose stacks set KC_HOSTNAME=kubernetes.docker.internal so
        # Keycloak issues redirect URIs against that host. Docker Desktop
        # auto-resolves it; GHA runners don't. Map it to 127.0.0.1 so the
        # browser-driven OAuth flow lands back on Stirling-PDF correctly.
        run: |
          echo "127.0.0.1 kubernetes.docker.internal" | sudo tee -a /etc/hosts

      # Helper function used by all phases — boots `:stirling-pdf:bootRun`
      # with the React frontend baked in (-PbuildWithFrontend=true) so the
      # SPA serves on :8080 and OAuth/SAML callbacks land on the same host
      # that the browser is interacting with.
      - name: Define helpers
        run: |
          {
            echo 'wait_for_backend() {'
            echo '  start=$SECONDS'
            echo '  for i in $(seq 1 300); do'
            echo '    if curl -fsS http://localhost:8080/api/v1/info/status >/dev/null 2>&1; then'
            echo '      echo "Backend up after $((SECONDS - start))s"; return 0'
            echo '    fi; sleep 2'
            echo '  done'
            echo '  tail -200 /tmp/backend.log || true; return 1'
            echo '}'
            echo 'stop_backend() {'
            echo '  if [ -f /tmp/backend.pid ]; then'
            echo '    kill "$(cat /tmp/backend.pid)" 2>/dev/null || true'
            echo '    rm -f /tmp/backend.pid'
            echo '  fi'
            echo '  pkill -f "gradlew :stirling-pdf:bootRun" 2>/dev/null || true'
            echo '  for i in $(seq 1 30); do'
            echo '    curl -fsS http://localhost:8080/api/v1/info/status >/dev/null 2>&1 || return 0'
            echo '    sleep 1'
            echo '  done'
            echo '}'
          } > /tmp/helpers.sh
          chmod +x /tmp/helpers.sh

      # ───────── OAuth round-trip ─────────
      - name: Bring up Keycloak (OAuth realm)
        working-directory: testing/compose
        run: docker compose -f docker-compose-keycloak-oauth.yml up -d --no-deps keycloak-oauth-db keycloak-oauth
      - name: Wait for Keycloak (OAuth) ready
        working-directory: testing/compose
        run: |
          for i in $(seq 1 60); do
            bash validate-oauth-test.sh 2>/dev/null && exit 0 || true
            # validate script also pings stirling on :8080 — accept just the
            # keycloak realm as our gate here, stirling boots in the next step
            curl -fsS http://localhost:9080/realms/stirling-oauth >/dev/null 2>&1 && exit 0
            sleep 5
          done
          docker compose -f docker-compose-keycloak-oauth.yml logs --tail=200 keycloak-oauth
          exit 1
      - name: Boot Stirling-PDF (frontend baked in, OAuth env)
        env:
          SECURITY_ENABLELOGIN: "true"
          SECURITY_LOGINMETHOD: "all"
          SECURITY_OAUTH2_ENABLED: "true"
          SECURITY_OAUTH2_AUTOCREATEUSER: "true"
          # Keycloak issues redirect URIs against KC_HOSTNAME, which the
          # compose default sets to kubernetes.docker.internal. Match here
          # (resolves to localhost via /etc/hosts mapping above).
          SECURITY_OAUTH2_CLIENT_KEYCLOAK_ISSUER: "http://kubernetes.docker.internal:9080/realms/stirling-oauth"
          SECURITY_OAUTH2_CLIENT_KEYCLOAK_CLIENTID: "stirling-pdf-client"
          SECURITY_OAUTH2_CLIENT_KEYCLOAK_CLIENTSECRET: "test-client-secret-change-in-production"
          SECURITY_OAUTH2_CLIENT_KEYCLOAK_USEASUSERNAME: "email"
          SECURITY_OAUTH2_CLIENT_KEYCLOAK_SCOPES: "openid,profile,email"
        run: |
          source /tmp/helpers.sh
          nohup ./gradlew :stirling-pdf:bootRun -PbuildWithFrontend=true > /tmp/backend.log 2>&1 &
          echo $! > /tmp/backend.pid
          wait_for_backend
      - name: Run enterprise OAuth Playwright tests
        id: oauth-tests
        env:
          PLAYWRIGHT_JSON_OUTPUT_FILE: ${{ github.workspace }}/frontend/playwright-report/results-oauth.json
        run: task e2e:enterprise -- --grep "OAuth"
      - name: Stop backend + tear down OAuth Keycloak
        if: always()
        run: |
          source /tmp/helpers.sh
          stop_backend
          (cd testing/compose && docker compose -f docker-compose-keycloak-oauth.yml down -v)

      # ───────── SAML round-trip ─────────
      - name: Bring up Keycloak (SAML realm)
        working-directory: testing/compose
        run: docker compose -f docker-compose-keycloak-saml.yml up -d --no-deps keycloak-saml-db keycloak-saml
      - name: Wait for Keycloak (SAML) ready
        working-directory: testing/compose
        run: |
          for i in $(seq 1 60); do
            curl -fsS http://localhost:9080/realms/stirling-saml >/dev/null 2>&1 && exit 0
            sleep 5
          done
          docker compose -f docker-compose-keycloak-saml.yml logs --tail=200 keycloak-saml
          exit 1
      - name: Generate SAML SP certs + fetch Keycloak IdP cert
        # The .pem/.crt/.key files are gitignored (test-only certs); the
        # docker-based start-saml-test.sh generates them at runtime, so do
        # the same in CI before bootRun reads them.
        working-directory: testing/compose
        run: |
          openssl req -x509 -newkey rsa:2048 \
            -keyout saml-private-key.key \
            -out saml-public-cert.crt \
            -days 3650 -nodes \
            -subj "/CN=stirling-pdf-saml-sp" >/dev/null 2>&1
          # Fetch Keycloak's SAML signing cert from the realm descriptor
          CERT_BODY=$(curl -sf http://localhost:9080/realms/stirling-saml/protocol/saml/descriptor \
            | awk 'BEGIN{RS="<[^>]*X509Certificate>|</[^>]*X509Certificate>"} NR==2{gsub(/[[:space:]]+/,""); print; exit}')
          {
            echo "-----BEGIN CERTIFICATE-----"
            echo "$CERT_BODY"
            echo "-----END CERTIFICATE-----"
          } > keycloak-saml-cert.pem
          test -s saml-private-key.key
          test -s saml-public-cert.crt
          test -s keycloak-saml-cert.pem
          echo "✓ SAML certs prepared"
      - name: Boot Stirling-PDF (frontend baked in, SAML env)
        env:
          SECURITY_ENABLELOGIN: "true"
          SECURITY_LOGINMETHOD: "all"
          SECURITY_SAML2_ENABLED: "true"
          SECURITY_SAML2_AUTOCREATEUSER: "true"
          SECURITY_SAML2_PROVIDER: "keycloak"
          SECURITY_SAML2_REGISTRATIONID: "keycloak"
          SECURITY_SAML2_IDP_ISSUER: "http://localhost:9080/realms/stirling-saml"
          SECURITY_SAML2_IDP_ENTITYID: "http://localhost:9080/realms/stirling-saml"
          SECURITY_SAML2_IDP_METADATAURI: "http://localhost:9080/realms/stirling-saml/protocol/saml/descriptor"
          SECURITY_SAML2_IDPSINGLELOGINURL: "http://localhost:9080/realms/stirling-saml/protocol/saml"
          SECURITY_SAML2_IDPSINGLELOGOUTURL: "http://localhost:9080/realms/stirling-saml/protocol/saml"
          SECURITY_SAML2_IDP_CERT: "${{ github.workspace }}/testing/compose/keycloak-saml-cert.pem"
          SECURITY_SAML2_PRIVATEKEY: "${{ github.workspace }}/testing/compose/saml-private-key.key"
          SECURITY_SAML2_SP_CERT: "${{ github.workspace }}/testing/compose/saml-public-cert.crt"
          # Realm registers the SP entity as the metadata URL — see
          # keycloak-realm-saml.json `clientId`. Match it here so Keycloak
          # accepts the AuthnRequest issuer.
          SECURITY_SAML2_SP_ENTITYID: "http://localhost:8080/saml2/service-provider-metadata/keycloak"
          SECURITY_SAML2_SP_ACS: "http://localhost:8080/login/saml2/sso/keycloak"
          SECURITY_SAML2_SP_SLS: "http://localhost:8080/logout/saml2/slo"
        run: |
          source /tmp/helpers.sh
          nohup ./gradlew :stirling-pdf:bootRun -PbuildWithFrontend=true > /tmp/backend.log 2>&1 &
          echo $! > /tmp/backend.pid
          wait_for_backend
      - name: Run enterprise SAML Playwright tests
        id: saml-tests
        env:
          PLAYWRIGHT_JSON_OUTPUT_FILE: ${{ github.workspace }}/frontend/playwright-report/results-saml.json
        run: task e2e:enterprise -- --grep "SAML"
      - name: Stop backend + tear down SAML Keycloak
        if: always()
        run: |
          source /tmp/helpers.sh
          stop_backend
          (cd testing/compose && docker compose -f docker-compose-keycloak-saml.yml down -v)

      # ───────── License-gated feature tests (no IdP needed) ─────────
      - name: Wipe DB so InitialSecuritySetup re-runs with admin/adminadmin
        # Earlier phases (OAuth, SAML) create the default admin/stirling user.
        # InitialSecuritySetup only honours SECURITY_INITIALLOGIN_* when the
        # admin user doesn't already exist, so the persisted DB has to be
        # cleared between phases for the feature env vars to take effect.
        run: |
          rm -f app/core/configs/stirling-pdf-DB*.mv.db
          rm -rf app/core/configs/backup
      - name: Boot Stirling-PDF (frontend baked in, premium only)
        env:
          SECURITY_INITIALLOGIN_USERNAME: admin
          SECURITY_INITIALLOGIN_PASSWORD: adminadmin
          SECURITY_ENABLELOGIN: "true"
          SECURITY_LOGINMETHOD: "all"
        run: |
          source /tmp/helpers.sh
          nohup ./gradlew :stirling-pdf:bootRun -PbuildWithFrontend=true > /tmp/backend.log 2>&1 &
          echo $! > /tmp/backend.pid
          wait_for_backend
      - name: Run enterprise feature Playwright tests
        id: feature-tests
        env:
          PLAYWRIGHT_JSON_OUTPUT_FILE: ${{ github.workspace }}/frontend/playwright-report/results-feature.json
        run: task e2e:enterprise -- --grep "Enterprise license"
      - name: Print backend log on failure
        if: failure()
        run: |
          echo "::group::Enterprise backend log"
          tail -500 /tmp/backend.log || true
          echo "::endgroup::"
      - name: Stop backend (final)
        if: always()
        run: |
          source /tmp/helpers.sh
          stop_backend
      - name: Flag flaky tests
        # Runs regardless of the test outcomes: a flaky test (passed on retry)
        # leaves its step green, so this is the only place it surfaces. Merges
        # all three phase reports (some may be absent if an earlier phase hard-
        # failed and skipped the rest). Emits ::warning:: annotations + a job
        # summary; never fails the job.
        if: always()
        working-directory: frontend
        run: >
          npx tsx editor/scripts/report-flaky-tests.mts
          "${{ github.workspace }}/frontend/playwright-report/results-oauth.json"
          "${{ github.workspace }}/frontend/playwright-report/results-saml.json"
          "${{ github.workspace }}/frontend/playwright-report/results-feature.json"
      - name: Upload Playwright report
        if: always()
        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: playwright-report-enterprise-${{ github.run_id }}
          path: frontend/playwright-report/
          retention-days: 7

      - name: Cleanup temporary files
        if: always()
        run: |
          rm -f /tmp/helpers.sh /tmp/backend.log /tmp/backend.pid
        continue-on-error: true

  # Multi-node regression: builds + seeds the clustered stack (testing/compose/docker-compose-multinode.yml)
  # and runs behave features/multinode. Licence-gated, so it runs after the Playwright job (not in parallel).
  multinode-e2e:
    environment:
      name: ci-unsigned
      deployment: false
    needs: [pick, playwright-e2e-enterprise]
    # Nightly cron + manual dispatch only (heavy build), fork-gated for the licence secret.
    if: >-
      always() && needs.pick.outputs.is_fork != 'true'
      && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch')
    runs-on: ${{ needs.pick.outputs.is_fork == 'true' && 'ubuntu-latest' || format('depot-ubuntu-24.04-{0}', inputs.depot_cores || '8') }}
    timeout-minutes: 60
    env:
      PREMIUM_KEY: ${{ secrets.PREMIUM_KEY_ENTERPRISE }}
      PREMIUM_ENABLED: "true"
      SYSTEM_ENABLEANALYTICS: "false"
      DEPOT_TOKEN: ${{ secrets.DEPOT_TOKEN }}
      MN_COMPOSE: docker-compose-multinode.yml
    steps:
      - name: Harden Runner
        uses: step-security/harden-runner@b09bb98e06d4d774595224525879c09bc6e98c40 # v2.20.1
        with:
          egress-policy: audit
      - name: Checkout repository
        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
      - name: Install uv
        uses: astral-sh/setup-uv@c771a70e6277c0a99b617c7a806ffedaca235ff9 # v9.0.0
        with:
          enable-cache: true
          cache-dependency-glob: |
            engine/pyproject.toml
            engine/uv.lock
      - name: Install behave test deps
        run: |
          uv sync --project engine --locked --group cucumber
      - name: Build the multi-node image
        working-directory: testing/compose
        run: docker compose -f "$MN_COMPOSE" build
      - name: Bring up the cluster and wait for both nodes healthy
        working-directory: testing/compose
        run: |
          docker compose -f "$MN_COMPOSE" up -d
          for i in $(seq 1 90); do
            h1=$(docker inspect -f '{{.State.Health.Status}}' multinode-stirling-1 2>/dev/null || echo starting)
            h2=$(docker inspect -f '{{.State.Health.Status}}' multinode-stirling-2 2>/dev/null || echo starting)
            if [ "$h1" = healthy ] && [ "$h2" = healthy ]; then echo "both nodes healthy"; exit 0; fi
            sleep 5
          done
          echo "::error::nodes did not become healthy"
          docker compose -f "$MN_COMPOSE" logs --tail=200 stirling-1 stirling-2
          exit 1
      - name: Seed the cluster (teams, users, S3 connection, policy)
        working-directory: testing/compose
        run: docker compose -f "$MN_COMPOSE" --profile seed run --rm seed
      - name: Run multi-node regression (implemented guarantees)
        working-directory: testing/cucumber
        # -e overrides behave.ini's exclusion of features/multinode; ~@known_gap skips any tracked-gap scenarios.
        run: uv run --project ../../engine --locked --group cucumber python -m behave features/multinode -e "features/enterprise" --tags="~@known_gap ~@destructive" --no-capture -f plain
      - name: Run multi-node failover (destructive)
        working-directory: testing/cucumber
        run: uv run --project ../../engine --locked --group cucumber python -m behave features/multinode -e "features/enterprise" --tags="@destructive ~@known_gap" --no-capture -f plain
      - name: Dump node logs on failure
        if: failure()
        working-directory: testing/compose
        run: docker compose -f "$MN_COMPOSE" logs --tail=400 stirling-1 stirling-2
      - name: Tear down
        if: always()
        working-directory: testing/compose
        run: docker compose -f "$MN_COMPOSE" --profile seed down -v --remove-orphans