ci(profiling): helper container resource limits (#3673)

The shared runner pool defaults leave helper container limits unset,
causing pods to be Burstable QoS even when the build container has
matching requests and limits. This adds explicit helper container
resource limits (requests == limits) so the pod achieves Guaranteed
QoS class, ensuring proper CPU cgroup enforcement.

This fixes the issue where nproc reports the node's full CPU count
(e.g. 66) instead of the requested 3 CPUs.

Co-authored-by: Daniel Esponda <daniel.esponda@datadoghq.com>

Author: morrisonlevi

.gitlab/generate-profiler.php

@@ -17,11 +17,18 @@
   stage: test
   tags: [ "arch:${ARCH}" ]
   image: registry.ddbuild.io/images/mirror/datadog/dd-trace-ci:${IMAGE_PREFIX}${PHP_MAJOR_MINOR}${IMAGE_SUFFIX}
+  # Setting the *_REQUEST and *_LIMIT variables to be the same, and setting
+  # them for both the build and helper allows using Guaranteed QoS instead of
+  # Burstable. This means nproc and similar tools will work as expected.
   variables:
     KUBERNETES_CPU_REQUEST: 3
     KUBERNETES_CPU_LIMIT: 3
     KUBERNETES_MEMORY_REQUEST: 6Gi
     KUBERNETES_MEMORY_LIMIT: 6Gi
+    KUBERNETES_HELPER_CPU_REQUEST: 1
+    KUBERNETES_HELPER_CPU_LIMIT: 1
+    KUBERNETES_HELPER_MEMORY_REQUEST: 2Gi
+    KUBERNETES_HELPER_MEMORY_LIMIT: 2Gi
     CARGO_TARGET_DIR: /mnt/ramdisk/cargo # ramdisk??
     libdir: /tmp/datadog-profiling
   parallel:
@@ -41,11 +48,6 @@
     - cd profiling
     - 'echo "nproc: $(nproc)"'
     - 'echo "KUBERNETES_CPU_REQUEST: ${KUBERNETES_CPU_REQUEST:-<unset>}"'
-    - |
-      if [ -n "${KUBERNETES_CPU_REQUEST:-}" ]; then
-        export CARGO_BUILD_JOBS="${KUBERNETES_CPU_REQUEST}"
-      fi
-      echo "CARGO_BUILD_JOBS: ${CARGO_BUILD_JOBS}"
     - export TEST_PHP_EXECUTABLE=$(which php)
     - run_tests_php=$(find $(php-config --prefix) -name run-tests.php) # don't anticipate there being more than one
     - cp -v "${run_tests_php}" tests