build_pull_request.yml

name: Build on Pull Request

on:
  pull_request:
    types: [opened, synchronize, reopened, labeled]
    branches:
      - "**"
  workflow_dispatch:
    inputs:
      database:
        description: 'Database engine for tests (default: h2)'
        required: false
        default: 'h2'
        type: choice
        options:
          - h2
          - sqlserver
          - sqlserver2017
          - postgres

env:
  DOCKER_HUB_ORGANIZATION: ${{ vars.DOCKER_HUB_ORGANIZATION }}

# ---------------------------------------------------------------------------
# compile  — compiles everything once, packages the JAR, uploads classes
# test     — 3-way matrix downloads compiled output and runs a shard of tests
#
# Wall-clock target:
#   compile ~10 min  (parallel with setup of test shards)
#   tests   ~8 min   (8 shards in parallel after compile finishes)
#   total   ~18 min  (vs ~40+ min single-job)
# ---------------------------------------------------------------------------

jobs:

  # --------------------------------------------------------------------------
  # Job 1: compile
  # --------------------------------------------------------------------------
  compile:
    runs-on: ubuntu-latest
    # The test job has carried a timeout since it was written; compile and report never did,
    # so a hung Maven resolve blocks the build until GitHub's own 6-hour ceiling.
    timeout-minutes: 25
    steps:
      - uses: actions/checkout@v4

      - name: Set up JDK 25
        uses: actions/setup-java@v4
        with:
          java-version: "25"
          distribution: "temurin"
          cache: maven          # caches ~/.m2/repository keyed on pom.xml hash

      - name: Setup production props
        run: |
          cp obp-api/src/main/resources/props/sample.props.template \
             obp-api/src/main/resources/props/production.default.props

      - name: Lint — test-isolation (no setPropsValues at class/feature body)
        run: python3 .github/scripts/check_test_isolation.py

      - name: Lint — resource-doc parity (Lift baseline vs live http4s ResourceDocs)
        run: |
          # Compares every ResourceDoc field (url, verb, summary, description, example
          # bodies, errors, tags) against the frozen Lift-era JSON baseline in
          # scripts/resource_doc_baseline/. Reviewed differences are recorded in
          # parity_allowlist.json, digest-bound to the text they were reviewed against, so a
          # later edit to an allowlisted field resurfaces instead of staying exempt.
          #
          # The script exits non-zero for an unreviewed difference. A STALE entry -- one that
          # matched nothing this run -- it only prints, so that is failed here as well: an
          # exemption that exempts nothing hides the very difference it was written for, and
          # it is the failure mode a hand-maintained allowlist develops first.
          set +e
          report=$(python3 scripts/check_lift_http4s_resource_doc_parity.py --report-stale-allowlist-entries)
          rc=$?
          set -e
          echo "$report"
          if [ "$rc" -ne 0 ]; then
            echo "::error::Resource-doc parity audit found unreviewed differences between the Lift baseline and the live http4s ResourceDocs. Either fix the doc to match the baseline, or record the difference with scripts/resource_doc_baseline/allowlist_helper.py (do not hand-compute digests, and do not edit the JSON baseline). See scripts/resource_doc_baseline/README.md."
            exit 1
          fi
          if ! printf '%s\n' "$report" | awk '/^=== STALE ALLOWLIST ENTRIES/{found=1} found' | grep -qx '  (none)'; then
            echo "::error::parity_allowlist.json contains entries that matched nothing this run. Remove them -- an exemption that exempts nothing gives no cover while looking as though it does."
            exit 1
          fi

      - name: Compile and install (skip test execution)
        run: |
          # -DskipTests  — compile test sources but do NOT run them
          # Test classes must be in target/test-classes for the test shards.
          # `clean` for a guaranteed-correct build (see build_container.yml: a no-clean
          # Zinc incremental cache saved only ~17s — compile is dominated by Maven +
          # fat-jar packaging + install, not Scala compilation).
          MAVEN_OPTS="-Xmx3G -Xss2m -XX:MaxMetaspaceSize=1G" \
            mvn clean install -T 4 -Pprod -DskipTests

      - name: Upload compiled output
        uses: actions/upload-artifact@v4
        with:
          name: compiled-output
          retention-days: 1
          # Upload full target dirs — test shards download and run surefire:test
          # without recompiling (surefire:test goal bypasses compile lifecycle)
          path: |
            obp-api/target/
            obp-commons/target/

      - name: Save .jar artifact
        run: |
          mkdir -p ./pull
          cp obp-api/target/obp-api.jar ./pull/
          cp -r obp-api/target/lib ./pull/lib

      - uses: actions/upload-artifact@v4
        with:
          name: ${{ github.sha }}
          path: pull/

  # --------------------------------------------------------------------------
  # Job 2: test (9-way matrix)
  #
  # Shard assignment (based on actual clean-run timings on ubuntu-latest 2-core):
  #   Shard 1  ~300s  v4_0_0 non-Dynamic (58 classes; Dynamic* split to shard 9)
  #   Shard 2  ~281s  v1_2_1 (largest single suite — isolated)
  #   Shard 3  ~250s  v6_0_0 (split from v2_x; v2_x moved to shard 7)
  #   Shard 4  ~232s  v5_1_0 + v5_0_0 + v3_0_0
  #   Shard 5  ~252s  ResourceDocs + v3_1_0 + v1_4_0 + v1_3_0
  #   Shard 6  ~200s  v7_0_0 + http4sserver + UKOpenBanking
  #   Shard 7  ~280s  model + views + customer + util + berlin + small data + v2_x
  #   Shard 8  ~240s  connector + auth + login + mgmt + metrics + catch-all
  #   Shard 9  ~300s  v4_0_0 Dynamic* (6 classes: 9 400+ lines each)
  # --------------------------------------------------------------------------
  test:
    needs: compile
    runs-on: ubuntu-latest
    timeout-minutes: ${{ (inputs.database == 'sqlserver' || inputs.database == 'sqlserver2017' || inputs.database == 'postgres' || contains(github.event.pull_request.labels.*.name, 'test-sqlserver') || contains(github.event.pull_request.labels.*.name, 'test-postgres') || contains(github.event.head_commit.message, '[test-sqlserver]') || contains(github.event.head_commit.message, '[test-postgres]')) && 90 || 35 }}
    strategy:
      fail-fast: false
      matrix:
        include:
          - shard: 1
            name: "v4 non-Dynamic"
            # v4_0_0 split: non-Dynamic classes only. Dynamic* on shard 9.
            # test_filter is a marker, not a literal wildcardSuites value: the "Run tests"
            # step below discovers code.api.v4_0_0.* classes at runtime and excludes
            # Dynamic* — a static class list here would silently drop any class added
            # after the list was written (the shard-8 catch-all treats the whole
            # code.api.v4_0_0 package as covered by shard 1, so an omitted class would
            # never run on ANY shard). Runtime discovery keeps shard 1 self-updating.
            test_filter: >-
              code.api.v4_0_0
          - shard: 2
            name: "v1_2_1 only (largest unsplittable suite, isolated)"
            # API1_2_1Test is a single 6604-line suite (~333 scenarios, ~281s). Isolated
            # on its own shard: mixing in berlin/management/metrics made this the slowest
            # shard (314s); those moved to shards 7/8 to rebalance.
            test_filter: >-
              code.api.v1_2_1
          - shard: 3
            name: "v6 only"
            # v6_0_0 isolated: 37 files / ~366 scenarios. Previously bundled with v2_x
            # causing 700s+ runs; v2_x moved to shard 7 which had headroom (~98s).
            test_filter: >-
              code.api.v6_0_0
          - shard: 4
            name: "v5_1 + v5_0 + v3_0"
            test_filter: >-
              code.api.v5_1_0
              code.api.v5_0_0
              code.api.v3_0_0
          - shard: 5
            name: "ResourceDocs + v3_1 + v1_4 + v1_3"
            test_filter: >-
              code.api.ResourceDocs1_4_0
              code.api.v3_1_0
              code.api.v1_4_0
              code.api.v1_3_0
          - shard: 6
            name: "v7 + http4sserver + UKOpenBanking"
            test_filter: >-
              code.api.v7_0_0
              code.api.http4sserver
              code.api.UKOpenBanking
          - shard: 7
            name: "model + views + customer + util + small data + berlin + v2_x"
            # v2_0_0/v2_1_0/v2_2_0 moved here from shard 3 to rebalance after v6_0_0 was isolated.
            test_filter: >-
              code.model
              code.views
              code.customer
              code.usercustomerlinks
              code.api.util
              code.errormessages
              code.atms
              code.branches
              code.products
              code.crm
              code.accountHolder
              code.api.berlin
              code.api.v2_1_0
              code.api.v2_2_0
              code.api.v2_0_0
          - shard: 8
            name: "connector + auth + login + mgmt + metrics + remaining (catch-all)"
            # catch-all shard: appends any test package not assigned to shards 1-7
            # Root-level code.api tests use class-name prefix matching (lowercase classes).
            # NOTE: classes that sit DIRECTLY in package code.api must be listed here by
            # FQN-prefix — the catch-all marks the parent package code.api as "covered" once
            # any child (code.api.v4_0_0, …) is assigned, so it never appends code.api itself.
            test_filter: >-
              code.connector
              code.util
              code.api.Authentication
              code.api.dauthTest
              code.api.DirectLoginTest
              code.api.gateWayloginTest
              code.api.OBPRestHelperTest
              code.api.AliveCheckRoutesTest
              code.api.Http4sOpenIdConnect
              code.entitlement
              code.bankaccountcreation
              code.bankconnectors
              code.container
              code.management
              code.metrics
              code.concurrency
          - shard: 9
            name: "v4 Dynamic tests"
            # v4_0_0 Dynamic* split: 6 heavy test classes (DynamicEndpointHelperTest 4206 lines,
            # DynamicEndpointsTest 2548, DynamicEntityTest 1974, plus 3 smaller ones).
            # Prefix code.api.v4_0_0.Dynamic matches all 6 classes; shard 1's runtime
            # discovery excludes this same prefix so no test runs in both shards.
            test_filter: >-
              code.api.v4_0_0.Dynamic
          - shard: 10
            name: "PropGatedPublicEndpoint (read_*_requires_role=true)"
            # JsonSchemaValidationPublicPropTrueTest / AuthenticationTypeValidationPublicPropTrueTest
            # need read_json_schema_validation_requires_role /
            # read_authentication_type_validation_requires_role forced true, which is baked into
            # Http4s400's ResourceDoc error list at object-init time -- a value only one JVM can
            # observe. Every other shard boots with these props unset (false, the default) and
            # the pom.xml default tagsToExclude skips this tag there. This shard is the only place
            # the true branch is exercised; see the "Run tests" step's shard-10 special case below.
            test_filter: >-
              code.api.v4_0_0.JsonSchemaValidationPublicPropTrueTest
              code.api.v4_0_0.AuthenticationTypeValidationPublicPropTrueTest

    services:
      redis:
        image: redis:7-alpine   # pinned + small (~30MB) to cut redis docker-pull flakiness
        ports:
          - 6379:6379
        options: >-
          --health-cmd "redis-cli ping"
          --health-interval 10s
          --health-timeout 5s
          --health-retries 5

    steps:
      - uses: actions/checkout@v4

      - name: Set up the test database
        uses: ./.github/actions/setup-test-database
        with:
          database: ${{ inputs.database }}
          pr-labels: ${{ toJson(github.event.pull_request.labels.*.name) }}
          commit-message: ${{ github.event.head_commit.message }}

      - name: Set up JDK 25
        uses: actions/setup-java@v4
        with:
          java-version: "25"
          distribution: "temurin"
          cache: maven

      - name: Download compiled output
        uses: actions/download-artifact@v4
        with:
          name: compiled-output

      - name: Touch artifact files (prevent Zinc recompilation)
        run: |
          # actions/download-artifact preserves original compile-job timestamps.
          # actions/checkout gives source files the current (later) time.
          # Zinc sees sources newer than classes → full recompile (~215 s wasted).
          # Touching everything in target/ makes all artifact files appear
          # just-downloaded (current time) → newer than sources → Zinc skips.
          find obp-api/target obp-commons/target -type f -exec touch {} + 2>/dev/null || true
          echo "Touched $(find obp-api/target obp-commons/target -type f 2>/dev/null | wc -l) files"

      - name: Install local artifacts into Maven repo
        run: |
          # The compile runner's ~/.m2 is discarded after that job completes.
          # Install the two local multi-module artifacts so scalatest:test can
          # resolve com.tesobe:* without hitting remote repos.
          #
          # 1. Parent POM — obp-commons' pom.xml declares obp-parent as its
          #    <parent>; Maven fetches it when reading transitive deps.
          mvn install:install-file \
            -Dfile=pom.xml \
            -DgroupId=com.tesobe \
            -DartifactId=obp-parent \
            -Dversion=1.10.1 \
            -Dpackaging=pom \
            -DgeneratePom=false
          # 2. obp-commons JAR with its full POM (lists compile deps inherited
          #    by obp-api at test classpath resolution time).
          mvn install:install-file \
            -Dfile=obp-commons/target/obp-commons-1.10.1.jar \
            -DpomFile=obp-commons/pom.xml

      - name: Setup props
        run: |
          cp obp-api/src/main/resources/props/sample.props.template \
             obp-api/src/main/resources/props/production.default.props
          echo connector=star > obp-api/src/main/resources/props/test.default.props

          if [ -n "$TEST_DB_URL" ]; then
            echo "Configuring $TEST_DB connection properties..."
            echo "db.driver=$TEST_DB_DRIVER" >> obp-api/src/main/resources/props/test.default.props
            echo "db.url=$TEST_DB_URL" >> obp-api/src/main/resources/props/test.default.props
          else
            echo "Using default H2 database."
          fi

          echo starConnector_supported_types=mapped,internal >> obp-api/src/main/resources/props/test.default.props
          echo hostname=http://localhost:8016 >> obp-api/src/main/resources/props/test.default.props
          echo tests.port=8016 >> obp-api/src/main/resources/props/test.default.props
          echo End of minimum settings >> obp-api/src/main/resources/props/test.default.props
          echo payments_enabled=false >> obp-api/src/main/resources/props/test.default.props
          echo messageQueue.updateBankAccountsTransaction=false >> obp-api/src/main/resources/props/test.default.props
          echo messageQueue.createBankAccounts=false >> obp-api/src/main/resources/props/test.default.props
          echo allow_sandbox_account_creation=true >> obp-api/src/main/resources/props/test.default.props
          echo allow_sandbox_data_import=true >> obp-api/src/main/resources/props/test.default.props
          echo sandbox_data_import_secret=change_me >> obp-api/src/main/resources/props/test.default.props
          echo allow_account_deletion=true >> obp-api/src/main/resources/props/test.default.props
          echo allowed_internal_redirect_urls = /,/oauth/authorize >> obp-api/src/main/resources/props/test.default.props
          echo transactionRequests_enabled=true >> obp-api/src/main/resources/props/test.default.props
          echo transactionRequests_supported_types=SEPA,SANDBOX_TAN,FREE_FORM,COUNTERPARTY,ACCOUNT,SIMPLE >> obp-api/src/main/resources/props/test.default.props
          echo SIMPLE_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo openredirects.hostname.whitlelist=http://127.0.0.1,http://localhost >> obp-api/src/main/resources/props/test.default.props
          echo remotedata.secret = foobarbaz >> obp-api/src/main/resources/props/test.default.props
          echo allow_public_views=true >> obp-api/src/main/resources/props/test.default.props
          echo SIMPLE_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo ACCOUNT_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo SEPA_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo FREE_FORM_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo COUNTERPARTY_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo SEPA_CREDIT_TRANSFERS_OTP_INSTRUCTION_TRANSPORT=dummy >> obp-api/src/main/resources/props/test.default.props
          echo allow_oauth2_login=true >> obp-api/src/main/resources/props/test.default.props
          echo oauth2.jwk_set.url=https://www.googleapis.com/oauth2/v3/certs >> obp-api/src/main/resources/props/test.default.props
          echo ResetPasswordUrlEnabled=true >> obp-api/src/main/resources/props/test.default.props
          echo consents.allowed=true >> obp-api/src/main/resources/props/test.default.props
          echo hikari.maximumPoolSize=20 >> obp-api/src/main/resources/props/test.default.props
          # Enables the Berlin Group v1.3 alias so ResourceDocRegistryParityTest and
          # ApiCollectionEndpointTest can exercise a real alias operation id end to end.
          echo berlin_group_v1_3_alias_path=0.6/v1 >> obp-api/src/main/resources/props/test.default.props
          echo write_metrics=false >> obp-api/src/main/resources/props/test.default.props
          # Log emails instead of opening a real SMTP socket: without this,
          # LocalMappedConnector.sendCustomerNotification's EMAIL branch calls
          # CommonsEmailWrapper.sendTextEmail which throws ConnectException because
          # there's no mail server in CI. That surfaces as 500 in any test that
          # hits an endpoint triggering the notification (v5 consent flows, etc.).
          echo mail.test.mode=true >> obp-api/src/main/resources/props/test.default.props
          # Permissions granted to runtime-compiled dynamic-endpoint code inside the security sandbox
          # (mirrors default.props / production.default.props). Required so dynamic resource-doc bodies
          # can do JSON extraction (reflection) and read OBP props (getenv); without it the sandbox
          # denies these and DynamicResourceDocTest's native-execution scenarios fail.
          echo 'dynamic_code_sandbox_permissions=[new java.net.NetPermission("specifyStreamHandler"), new java.lang.reflect.ReflectPermission("suppressAccessChecks"), new java.lang.RuntimePermission("getenv.*"), new java.util.PropertyPermission("cglib.useCache", "read"), new java.util.PropertyPermission("net.sf.cglib.test.stressHashCodes", "read"), new java.util.PropertyPermission("cglib.debugLocation", "read"), new java.lang.RuntimePermission("accessDeclaredMembers"), new java.lang.RuntimePermission("getClassLoader")]' >> obp-api/src/main/resources/props/test.default.props
          # allow_user_generated_scala_code defaults to false everywhere (including test/dev)
          # with no run-mode fallback; CI must opt in explicitly so DynamicUtilTest,
          # ConnectorMethodTest, AbacRuleTests, DynamicResourceDocTest, DynamicMessageDocTest
          # and DynamicCodeKillSwitchTest's ON scenarios can still compile/execute dynamic code.
          echo allow_user_generated_scala_code=true >> obp-api/src/main/resources/props/test.default.props

      - name: Run tests — shard ${{ matrix.shard }} (${{ matrix.name }})
        env:
          # This job has declared a redis service since it was written, but nothing ever failed
          # when the service was absent: ConcurrentRateLimiterRaceTest and
          # MethodRoutingCacheInvalidationTest each `assume` a reachable Redis and cancel
          # otherwise, and a cancelled test reports as a pass. Dropping the services: block, or
          # a container that never became healthy, would have taken the rate-limiter and
          # cache-invalidation races out of the run without changing a single report.
          #
          # RedisTestTarget turns that cancellation into a failure wherever this is set.
          # Developers leave it unset and keep the skip.
          OBP_TEST_REDIS_REQUIRED: "true"
        run: |
          # wildcardSuites requires comma-separated package prefixes (-w per entry).
          # The YAML >- scalar collapses newlines to spaces, so we convert here.
          FILTER=$(echo "${{ matrix.test_filter }}" | tr ' ' ',')

          # Shard 1 (v4 non-Dynamic): discover code.api.v4_0_0 classes at runtime and
          # exclude Dynamic* (shard 9's domain), instead of a hand-maintained class list.
          # This is what makes the shard-8 catch-all's "code.api.v4_0_0 is covered by
          # shard 1" assumption actually true — a class list here could drift from the
          # real directory contents and leave a new test running on no shard at all.
          if [ "${{ matrix.shard }}" = "1" ]; then
            FILTER=$(grep -l '^class.*extends' obp-api/src/test/scala/code/api/v4_0_0/*.scala \
                       | xargs -I{} grep -hoP '^class \K[A-Z][A-Za-z0-9_]+' {} \
                       | grep -v '^Dynamic' \
                       | sed 's/^/code.api.v4_0_0./' \
                       | paste -sd, -)
          fi

          # Shard 8 is the catch-all: append any test package not explicitly
          # assigned to shards 1–7 and 9, so new packages are never silently skipped.
          if [ "${{ matrix.shard }}" = "8" ]; then
            # Shard 1 discovers v4 non-Dynamic classes at runtime (see above); shard 9
            # covers Dynamic*. Use code.api.v4_0_0 as the assigned prefix so the
            # catch-all treats the whole v4_0_0 package as covered — true as long as
            # shard 1's discovery and shard 9's Dynamic prefix together span it.
            SHARD1="code.api.v4_0_0"
            SHARD2="code.api.v1_2_1"
            SHARD3="code.api.v6_0_0"
            SHARD4="code.api.v5_1_0 code.api.v5_0_0 code.api.v3_0_0"
            SHARD5="code.api.ResourceDocs1_4_0 code.api.v3_1_0 code.api.v1_4_0 code.api.v1_3_0"
            SHARD6="code.api.v7_0_0 code.api.http4sserver code.api.UKOpenBanking"
            SHARD7="code.model code.views code.customer code.usercustomerlinks \
                    code.api.util code.errormessages code.atms code.branches \
                    code.products code.crm code.accountHolder code.api.berlin \
                    code.api.v2_1_0 code.api.v2_2_0 code.api.v2_0_0"
            ASSIGNED="$SHARD1 $SHARD2 $SHARD3 $SHARD4 $SHARD5 $SHARD6 $SHARD7 ${{ matrix.test_filter }}"

            # Discover all packages that contain at least one .scala test file
            ALL_PKGS=$(find obp-api/src/test/scala obp-commons/src/test/scala \
                         -name "*.scala" 2>/dev/null \
                       | sed 's|.*/test/scala/||; s|/[^/]*\.scala$||; s|/|.|g' \
                       | sort -u)

            EXTRAS=""
            for pkg in $ALL_PKGS; do
              covered=false
              for prefix in $ASSIGNED; do
                if [[ "$pkg" == "$prefix" || "$pkg" == "$prefix."* || "$prefix" == "$pkg."* ]]; then
                  covered=true; break
                fi
              done
              [ "$covered" = "false" ] && EXTRAS="$EXTRAS,$pkg"
            done

            [ -n "$EXTRAS" ] && echo "Catch-all extras added to shard ${{ matrix.shard }}:$EXTRAS"
            FILTER="${FILTER}${EXTRAS}"
          fi

          # Shard 10: PropGatedPublicEndpoint tests need read_json_schema_validation_requires_role /
          # read_authentication_type_validation_requires_role forced true and the
          # PropGatedPublicEndpoint tag included (pom.xml's default tagsToExclude skips it
          # everywhere else). These two OBP_* env vars and the tag override apply only to this
          # shard's mvn invocation below.
          EXTRA_MVN_ARGS=""
          if [ "${{ matrix.shard }}" = "10" ]; then
            export OBP_READ_JSON_SCHEMA_VALIDATION_REQUIRES_ROLE="true"
            export OBP_READ_AUTHENTICATION_TYPE_VALIDATION_REQUIRES_ROLE="true"
            EXTRA_MVN_ARGS="-DtagsToInclude=PropGatedPublicEndpoint -Dtest.tagsToExclude="
          fi

          # `mvn process-resources scalatest:test` — run process-resources (copies the
          # dynamically-generated props from src/main/resources onto the classpath at
          # target/classes/props) then the scalatest:test goal. This SKIPS the
          # compile/testCompile lifecycle phases (~208s/shard: classes already come from
          # the compiled-output artifact + touch trick) while still placing
          # test.default.props on the classpath. (Plain goal-only `scalatest:test` skips
          # process-resources → Props init fails → 0 tests but BUILD SUCCESS = false green.)
          # -pl obp-commons,obp-api: obp-commons' own 5 util suites run on whichever
          # shard's filter matches com.openbankproject.* (the catch-all shard); on every
          # other shard the filter matches nothing in obp-commons → 0 tests there.
          # timeout $TEST_TIMEOUT_SECONDS (25 min on H2, 75 on SQL Server and PostgreSQL): hard-kill to prevent Pekko non-daemon threads
          # (ConsentScheduler etc.) from keeping the JVM alive after tests complete.
          # Exit code 124 (timeout) is treated as success — tests are done, JVM just hung.
          # set +e: GitHub Actions uses -eo pipefail by default; without it, a 124 exit from
          # timeout would abort the step before the rc check below can run.
          set +e
          MAVEN_OPTS="-Xmx3G -Xss2m -XX:MaxMetaspaceSize=1G" \
            timeout "$TEST_TIMEOUT_SECONDS" mvn process-resources scalatest:test -pl obp-commons,obp-api -DfailIfNoTests=false \
              -DwildcardSuites="$FILTER" $EXTRA_MVN_ARGS \
            > maven-build-shard${{ matrix.shard }}.log 2>&1
          rc=$?
          set -e
          # timeout returns 124 when the JVM was killed. That is only benign when the tests had
          # successfully finished but Pekko non-daemon threads kept the JVM alive. We must
          # require proof from the log instead of blindly converting 124 to success.
          if [ $rc -eq 124 ]; then
            if grep -q "BUILD SUCCESS" maven-build-shard${{ matrix.shard }}.log; then
              rc=0
            fi
          fi
          exit $rc

      - name: Report failing tests — shard ${{ matrix.shard }}
        if: always()
        run: |
          echo "Checking shard ${{ matrix.shard }} log for failing tests..."
          if [ ! -f maven-build-shard${{ matrix.shard }}.log ]; then
            echo "No build log found."; exit 0
          fi
          echo "=== RECOMPILATION CHECK ==="
          if grep -c "Compiling " maven-build-shard${{ matrix.shard }}.log > /dev/null 2>&1; then
            echo "WARNING: Scala recompilation occurred on this shard:"
            grep "Compiling " maven-build-shard${{ matrix.shard }}.log | head -10
          else
            echo "OK: no recompilation (Zinc used pre-compiled classes)"
          fi
          echo ""
          echo "=== BRIDGE / UNCAUGHT EXCEPTIONS ==="
          grep -n "\[BRIDGE\] Exception\|Uncaught exception in dispatch\|requestScopeProxy=" \
            maven-build-shard${{ matrix.shard }}.log | head -200 || true
          echo ""
          echo "=== FAILING TEST SCENARIOS (with 30 lines context) ==="
          # maven.test.failure.ignore=true (root pom) makes mvn exit 0 even when a suite
          # aborts entirely — "*** FAILED ***" alone misses that, since scalatest prints
          # "*** RUN ABORTED ***" / "*** SUITE ABORTED ***" instead for e.g. an
          # ExceptionInInitializerError, and neither pattern was being checked. That let
          # a genuinely aborted suite report CI green silently.
          if grep -C 30 -n -E "\*\*\* FAILED \*\*\*|\*\*\* RUN ABORTED \*\*\*|\*\*\* SUITE ABORTED \*\*\*" \
              maven-build-shard${{ matrix.shard }}.log; then
            echo "Failing/aborted tests detected in shard ${{ matrix.shard }}."
            exit 1
          else
            echo "No failing tests detected in shard ${{ matrix.shard }}."
          fi

          # "No failures" is not "it ran". A shard whose filter matches nothing, or
          # whose tag override stops taking effect, executes zero tests, matches no
          # failure pattern and reports green -- the same false green the
          # process-resources note above guards against, from a different direction.
          # Shard 10 is the exposed one: its suites only run while
          # -DtagsToInclude=PropGatedPublicEndpoint -Dtest.tagsToExclude= holds.
          RAN=$(grep -oE "Tests: succeeded [0-9]+" maven-build-shard${{ matrix.shard }}.log \
                | awk '{s+=$3} END {print s+0}')
          echo "Shard ${{ matrix.shard }} executed ${RAN} tests."
          if [ "${RAN}" -eq 0 ]; then
            echo "Shard ${{ matrix.shard }} ran no tests at all — treating as failure."
            exit 1
          fi

      - name: Upload Maven build log — shard ${{ matrix.shard }}
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: maven-build-log-shard${{ matrix.shard }}
          if-no-files-found: ignore
          path: maven-build-shard${{ matrix.shard }}.log

      - name: Upload test reports — shard ${{ matrix.shard }}
        if: always()
        uses: actions/upload-artifact@v4
        with:
          name: test-reports-shard${{ matrix.shard }}
          if-no-files-found: ignore
          path: |
            obp-api/target/surefire-reports/**
            obp-commons/target/surefire-reports/**
            **/target/scalatest-reports/**
            **/target/site/surefire-report.html
            **/target/site/surefire-report/*

  # --------------------------------------------------------------------------
  # Job 3: report — per-test speed (unit/pure vs integration, all http4s)
  # --------------------------------------------------------------------------
  report:
    needs: test
    runs-on: ubuntu-latest
    if: always()
    timeout-minutes: 10
    steps:
      - uses: actions/checkout@v4

      - name: Download test reports — all shards
        uses: actions/download-artifact@v4
        with:
          pattern: test-reports-shard*
          path: all-reports
          merge-multiple: true

      - name: Per-test speed — unit/pure vs integration (all http4s)
        run: python3 .github/scripts/test_speed_report.py all-reports