[ML] Use optimal test parallelism based on CPU count

Each test suite spawns ctest --parallel <all_cpus> internally, so
running all suites concurrently causes resource contention. Limit
the number of concurrent suites to ceil(nproc/3) on Unix and 2 on
Windows, matching empirically determined optimal values from
cross-platform benchmarking (see PR elastic#2900).

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: edsavage

build.gradle

@@ -112,6 +112,11 @@ if (isWindows && msystem != 'MINGW64') {
 }
 
 project.ext.numCpus = Runtime.runtime.availableProcessors()
+// Optimal number of test suites to run concurrently via cmake --build -j.
+// Each suite spawns ctest --parallel <all_cpus> internally, so running too
+// many suites simultaneously causes resource contention. These values were
+// determined empirically (see PR #2900).
+project.ext.testParallel = isWindows ? 2 : Math.max(1, (int) Math.ceil(numCpus / 3.0))
 project.ext.makeEnvironment = [ 'CPP_CROSS_COMPILE': cppCrossCompile,
                                 'VERSION_QUALIFIER': versionQualifier,
                                 'SNAPSHOT': (isSnapshot ? 'yes' : 'no'),
@@ -223,7 +228,7 @@ String artifactGroupPath = project.group.replaceAll("\\.", "/")
 task test(type: Exec) {
   environment makeEnvironment
   commandLine shell
-  args shellFlag, "${setEnv} && cmake --build ${cmakeBuildDir} --config ${cmakeBuildType} -v -j ${numCpus} -t test_individually"
+  args shellFlag, "${setEnv} && cmake --build ${cmakeBuildDir} --config ${cmakeBuildType} -v -j ${testParallel} -t test_individually"
   workingDir "${projectDir}"
   dependsOn 'strip'
   description = 'Run C++ tests'

dev-tools/docker/docker_entrypoint.sh

@@ -66,6 +66,9 @@ if [ "x$1" = "x--test" ] ; then
     # failure is the unit tests, and then the detailed test results can be
     # copied from the image
     echo passed > build/test_status.txt
-    cmake --build cmake-build-docker ${CMAKE_VERBOSE} -j $(nproc) -t test_individually || echo failed > build/test_status.txt
+    # Each test suite spawns ctest --parallel <nproc> internally, so limit
+    # the number of suites running concurrently to avoid resource contention.
+    TEST_PARALLEL=$(( ($(nproc) + 2) / 3 ))
+    cmake --build cmake-build-docker ${CMAKE_VERBOSE} -j ${TEST_PARALLEL} -t test_individually || echo failed > build/test_status.txt
 fi