integration.yml

name: Integration Tests

on:
  push:
    branches: [main]
  pull_request:
    branches: [main]
  workflow_dispatch: {}
  # Weekly drift catch — Tuesday 06:00 UTC so failures land in EU/IN
  # working hours, not weekend handover.
  schedule:
    - cron: '0 6 * * 2'

permissions:
  contents: read

# Avoid spawning parallel docker-compose stacks for back-to-back pushes;
# also cancels stale PR runs when a new commit lands. Push to main keys on SHA
# so main runs never queue or cancel each other.
concurrency:
  group: integration-${{ github.event.pull_request.number || github.sha }}
  cancel-in-progress: ${{ github.event_name == 'pull_request' }}

env:
  AXONFLOW_TELEMETRY: 'off'

jobs:
  # WireMock-based integration tests run on every PR + push. No live stack
  # needed — these are contract-style tests over the SDK + agent wire shape.
  # Matrixed across the same JDKs as the unit-test suite in ci.yml.
  contract-integration:
    name: Contract Integration (WireMock, JDK ${{ matrix.java-version }})
    runs-on: ubuntu-latest
    timeout-minutes: 15
    strategy:
      fail-fast: false
      # PR runs only Java 17 (release toolchain). Push, dispatch, and weekly
      # cron run the full [11, 17, 21] matrix to catch version drift.
      matrix:
        java-version: ${{ fromJson(github.event_name == 'pull_request' && '[17]' || '[11, 17, 21]') }}
    steps:
      - name: Checkout SDK
        uses: actions/checkout@v4

      - name: Set up JDK ${{ matrix.java-version }}
        uses: actions/setup-java@v4
        with:
          java-version: ${{ matrix.java-version }}
          distribution: 'temurin'
          cache: 'maven'

      - name: Configure Maven mirror
        uses: ./.github/actions/setup-maven

      # `-DskipUnitTests=true` is now a real toggle (bound to surefire's
      # <skipTests> via pom.xml); previously it was a no-op flag and unit
      # tests were silently re-running here.
      #
      # `-Djacoco.skip=true` because the jacoco:check goal (bound to verify)
      # expects coverage data from the unit tests we just skipped; coverage
      # gating is the unit-test job's responsibility (ci.yml `build (17)`).
      - name: Run integration tests (WireMock)
        run: mvn verify -DskipUnitTests=true -Djacoco.skip=true -B -U

      - name: Upload failsafe reports on failure
        if: failure()
        uses: actions/upload-artifact@v4
        with:
          name: failsafe-reports-jdk${{ matrix.java-version }}
          path: target/failsafe-reports/
          if-no-files-found: ignore

  # Live integration runs against a real community stack via docker compose.
  # Mirrors axonflow-sdk-go/.github/workflows/integration.yml — clones the
  # community repo, brings up docker compose, runs the basic example.
  # Skipped on PR (Go pattern) — PR-level live coverage is added in QF-13.
  live-integration:
    name: Live Integration (Community Stack)
    runs-on: ubuntu-latest
    timeout-minutes: 25
    needs: contract-integration
    if: github.event_name != 'pull_request'
    steps:
      - name: Checkout SDK
        uses: actions/checkout@v4

      - name: Set up JDK 17
        uses: actions/setup-java@v4
        with:
          java-version: '17'
          distribution: 'temurin'
          cache: 'maven'

      - name: Configure Maven mirror
        uses: ./.github/actions/setup-maven

      # 3-attempt retry for transient Maven Central flakes — same pattern
      # ci.yml's `Build with Maven` and `Run unit tests` use.
      - name: Install SDK to local Maven repo
        run: |
          for i in 1 2 3; do
            echo "Attempt $i: mvn install"
            if mvn install -DskipTests -B -U; then break; fi
            if [ $i -eq 3 ]; then exit 1; fi
            sleep 30
          done

      # Pin the basic example's SDK dep to whatever we just installed
      # locally, so the example resolves the freshly-built artifact and
      # not whatever version is published to Central. Without this, when
      # the parent pom bumps from 6.1.0 → 6.2.0 the example silently
      # keeps testing the OLD 6.1.0 from Central.
      - name: Sync example SDK version with parent
        run: |
          PARENT_VERSION=$(mvn -B -q -DforceStdout help:evaluate -Dexpression=project.version)
          echo "Parent SDK version: ${PARENT_VERSION}"
          # Replace the axonflow-sdk dependency version in examples/basic/pom.xml.
          # Anchored on the artifactId on the previous line to avoid touching
          # other deps. Asserts the regex matched (count >= 1) so a layout
          # drift fails CI loud; a no-op rewrite (versions already match)
          # is fine.
          python3 - <<PY
          import re, pathlib
          p = pathlib.Path("examples/basic/pom.xml")
          s = p.read_text()
          new, count = re.subn(
              r"(<artifactId>axonflow-sdk</artifactId>\s*<version>)[^<]+(</version>)",
              rf"\g<1>${PARENT_VERSION}\g<2>",
              s,
          )
          assert count >= 1, "Regex did not match — examples/basic/pom.xml layout drifted"
          if new != s:
              p.write_text(new)
              print(f"Rewrote example pom version to ${PARENT_VERSION}")
          else:
              print(f"Example pom already at parent version ${PARENT_VERSION} (no-op)")
          PY
          grep -A 1 "axonflow-sdk" examples/basic/pom.xml | head -4

      - name: Clone community stack
        run: git clone --depth 1 https://github.com/getaxonflow/axonflow.git ../axonflow

      - name: Start community stack
        run: |
          cd ../axonflow
          docker compose up -d --wait --wait-timeout 120

          # Belt-and-suspenders: also poll /health since not every compose
          # service has a healthcheck wired.
          echo "Waiting for agent to be healthy..."
          timeout 120 bash -c 'until curl -sf http://localhost:8080/health; do sleep 2; done'
          echo "Agent is healthy"

          echo "Waiting for orchestrator to be healthy..."
          timeout 60 bash -c 'until curl -sf http://localhost:8081/health; do sleep 2; done'
          echo "Orchestrator is healthy"

      - name: Run basic example against live stack
        env:
          AXONFLOW_AGENT_URL: http://localhost:8080
          AXONFLOW_CLIENT_ID: demo-client
          AXONFLOW_CLIENT_SECRET: demo-secret
        working-directory: examples/basic
        run: timeout 90 mvn -q compile exec:java

      # Logs MUST be captured before `Stop community stack` runs — `compose
      # down` destroys the containers and `compose logs` then returns
      # nothing.
      - name: Show docker logs on failure
        if: failure()
        run: |
          if [ -d "../axonflow" ]; then
            cd ../axonflow
            docker compose logs --tail=200 || true
          fi

      - name: Upload docker logs on failure
        if: failure()
        uses: actions/upload-artifact@v4
        with:
          name: docker-compose-logs
          path: ../axonflow/docker-compose-logs.txt
          if-no-files-found: ignore

      - name: Stop community stack
        if: always()
        run: |
          if [ -d "../axonflow" ]; then
            cd ../axonflow
            # Persist logs to disk so the upload step can grab them even after teardown.
            docker compose logs --tail=500 > docker-compose-logs.txt 2>/dev/null || true
            docker compose down --volumes --remove-orphans || true
          fi