ci(build): split allTests into parallel per-target jobs to end OOM flakes

.github/workflows/build.yml

@@ -11,10 +11,31 @@ concurrency:
   cancel-in-progress: true
 
 jobs:
-  build-job:
+  # `allTests` used to run in a single job alongside `assemble`. On a 7 GB
+  # ubuntu-latest runner the combined peak RSS (Gradle JVM + Kotlin daemon +
+  # per-target test forks + Node for JS/Wasm + the native toolchain, all at
+  # once) intermittently exceeded physical memory. The runner was then
+  # OOM-killed and GitHub reported "The operation was canceled" after ~15 min
+  # with no BUILD FAILED — the flaky ~20-minute reds.
+  #
+  # Splitting the test matrix so each leg runs one target family keeps every
+  # job's memory well under the runner limit, and the legs run in parallel so
+  # wall-clock is the slowest single leg rather than the sum. `assemble` runs
+  # as its own (memory-light) job.
+  test:
+    name: test (${{ matrix.name }})
     runs-on: ubuntu-latest
-    timeout-minutes: 60
-
+    timeout-minutes: 40
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - name: jvm
+            tasks: jvmTest
+          - name: js-wasm
+            tasks: jsTest wasmJsTest wasmWasiTest
+          - name: native
+            tasks: linuxX64Test
     steps:
       - name: Checkout
         uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
@@ -33,45 +54,112 @@ jobs:
         with:
           node-version: '20'
 
-      - name: Print git commit variables
-        run: |
-          echo "TAG: $CURRENT_TAG"
+      # Warm the Gradle dependency/wrapper cache so each parallel leg does not
+      # cold-download the full dependency graph. Matches docs.yml.
+      - name: Cache Gradle
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: |
+            ~/.gradle/caches
+            ~/.gradle/wrapper
+          key: gradle-${{ runner.os }}-${{ hashFiles('**/*.gradle*', '**/gradle-wrapper.properties', '**/libs.versions.toml') }}
+          restore-keys: |
+            gradle-${{ runner.os }}-
 
       - name: Disk space (observability)
         run: df -h
 
-      - name: Build and Test (PR)
-        if: github.event_name == 'pull_request'
-        env:
-          GRADLE_OPTS: -Dorg.gradle.jvmargs="-Xmx4g -Dfile.encoding=UTF-8"
-        run: |
-          ./gradlew --no-daemon --stacktrace --info \
-            -Dorg.gradle.caching=true \
-            -Dorg.gradle.configuration-cache=true \
-            clean assemble allTests
-
-      - name: Build and Test (push)
-        if: github.event_name != 'pull_request'
+      - name: Test (${{ matrix.name }})
         env:
           GRADLE_OPTS: -Dorg.gradle.jvmargs="-Xmx4g -Dfile.encoding=UTF-8"
         run: |
-          ./gradlew --no-daemon --stacktrace --info \
+          ./gradlew --no-daemon --stacktrace \
             -Dorg.gradle.caching=true \
             -Dorg.gradle.configuration-cache=true \
-            assemble allTests
+            ${{ matrix.tasks }}
 
-      - name: Memory info (on failure)
-        if: failure()
+      # `cancelled()` matters: an OOM-killed runner reports the job as
+      # cancelled, not failed, so `if: failure()` alone never captured the
+      # memory snapshot on exactly the runs we needed it for.
+      - name: Memory info (on failure or cancellation)
+        if: failure() || cancelled()
         run: |
           free -h || true
-          cat /proc/meminfo | head -n 50 || true
+          head -n 50 /proc/meminfo || true
 
-      - name: Upload all test reports
+      - name: Upload test reports (${{ matrix.name }})
         if: always()
         uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
-          name: test-reports
+          name: test-reports-${{ matrix.name }}
           path: |
             **/build/reports/tests/**
             **/build/test-results/**
           retention-days: 14
+
+  assemble:
+    name: assemble
+    runs-on: ubuntu-latest
+    timeout-minutes: 40
+    steps:
+      - name: Checkout
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Copy CI gradle.properties
+        run: mkdir -p ~/.gradle ; cp .github/ci-gradle.properties ~/.gradle/gradle.properties
+
+      - name: Set up JDK 25
+        uses: actions/setup-java@03ad4de0992f5dab5e18fcb136590ce7c4a0ac95 # v5.6.0
+        with:
+          distribution: 'zulu'
+          java-version: 25
+
+      - name: Set up Node.js
+        uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: '20'
+
+      - name: Cache Gradle
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: |
+            ~/.gradle/caches
+            ~/.gradle/wrapper
+          key: gradle-${{ runner.os }}-${{ hashFiles('**/*.gradle*', '**/gradle-wrapper.properties', '**/libs.versions.toml') }}
+          restore-keys: |
+            gradle-${{ runner.os }}-
+
+      - name: Assemble
+        env:
+          GRADLE_OPTS: -Dorg.gradle.jvmargs="-Xmx4g -Dfile.encoding=UTF-8"
+        run: |
+          ./gradlew --no-daemon --stacktrace \
+            -Dorg.gradle.caching=true \
+            -Dorg.gradle.configuration-cache=true \
+            assemble
+
+      - name: Memory info (on failure or cancellation)
+        if: failure() || cancelled()
+        run: |
+          free -h || true
+          head -n 50 /proc/meminfo || true
+
+  # Aggregate gate. Branch protection references the single "build-job" check;
+  # keeping a job with that exact name (now backed by the parallel test matrix
+  # + assemble) preserves the required-status-check contract without needing to
+  # re-point branch protection. Fails if any dependency did not succeed.
+  build-job:
+    name: build-job
+    needs: [ test, assemble ]
+    if: always()
+    runs-on: ubuntu-latest
+    steps:
+      - name: Verify test matrix and assemble succeeded
+        run: |
+          echo "test result:     ${{ needs.test.result }}"
+          echo "assemble result: ${{ needs.assemble.result }}"
+          if [ "${{ needs.test.result }}" != "success" ] || [ "${{ needs.assemble.result }}" != "success" ]; then
+            echo "::error::One or more build jobs did not succeed (see the test/assemble jobs above)."
+            exit 1
+          fi
+          echo "All build jobs succeeded."