Disable parallel for smoke test

tests/conftest.py

@@ -260,6 +260,15 @@ def pytest_configure(config):
 
     pytest.terminate_on_failure = config.getoption('--terminate-on-failure')
 
+    # Disable parallelism for smoke tests only to save memory in Buildkite.
+    # Check if any of the test paths are under smoke_tests/
+    if hasattr(config, 'args') and config.args:
+        is_smoke_test = any('smoke_tests' in str(arg) for arg in config.args)
+        if is_smoke_test and smoke_tests_utils.is_in_buildkite_env():
+            # Override xdist settings to disable parallelism
+            config.option.numprocesses = 0
+            config.option.dist = 'no'
+
 
 def _get_cloud_to_run(config) -> List[str]:
     cloud_to_run = []

tests/smoke_tests/docker/docker_utils.py

@@ -97,6 +97,7 @@ def create_and_setup_new_container(target_container_name: str,
     if is_inside_docker():
         # Run the new container directly
         run_cmd = (f'docker run -d '
+                   f'--cpus=4 '
                    f'--name {target_container_name} '
                    f'-p {api_server_host_port}:{api_server_container_port} '
                    f'-p {metrics_host_port}:{metrics_container_port} '

tests/smoke_tests/test_api_server.py

@@ -405,9 +405,9 @@ def _rss_aggregate_threshold_checker(
                 total_increase: float, total_increase_pct: float) -> List[str]:
             """Aggregate threshold checker for RSS metrics."""
             failures = []
-            if total_increase_pct > 20:
+            if total_increase_pct > 30:
                 failures.append(
-                    f"Average memory increase too high: {total_increase_pct:.1f}% (limit: 20%)"
+                    f"Average memory increase too high: {total_increase_pct:.1f}% (limit: 30%)"
                 )
             return failures