fix: E2E tests

Signed-off-by: Alon Kellner <[email protected]>

Author: AlonKellner-RedHat

src/guidellm/__main__.py

@@ -388,8 +388,10 @@ def benchmark():
     "--detect-saturation",  # alias
     default=None,
     help=(
-        "Enable over-saturation detection. Can be a flag (bool) or a JSON dict with "
-        'configuration (e.g., \'{"enabled": true, "min_seconds": 30}\'). '
+        "Enable over-saturation detection. "
+        "Use --over-saturation=True for boolean flag, "
+        "or a JSON dict with configuration "
+        '(e.g., \'{"enabled": true, "min_seconds": 30}\'). '
         "Defaults to None (disabled)."
     ),
     type=click.UNPROCESSED,

tests/e2e/test_over_saturated_benchmark.py

@@ -72,3 +72,50 @@ def test_over_saturated_benchmark(server: VllmSimServer):
     )
 
     cleanup_report_file(report_path)
+
+
+@pytest.mark.timeout(60)
+def test_over_saturated_benchmark_with_dict_config(server: VllmSimServer):
+    """
+    Test over-saturation detection with dictionary configuration instead of boolean.
+    """
+    report_path = Path("tests/e2e/over_saturated_benchmarks_dict.json")
+    rate = 100
+
+    # Create and configure the guidellm client
+    client = GuidellmClient(target=server.get_url(), output_path=report_path)
+
+    cleanup_report_file(report_path)
+    # Start the benchmark with dictionary configuration for over-saturation
+    client.start_benchmark(
+        rate=rate,
+        max_seconds=20,
+        over_saturation={
+            "enabled": True,
+            "min_seconds": 0,
+            "max_window_seconds": 120.0,
+            "moe_threshold": 2.0,
+            "minimum_window_size": 5,
+        },
+        extra_env={
+            "GUIDELLM__CONSTRAINT_OVER_SATURATION_MIN_SECONDS": "0",
+            "GOMAXPROCS": "1",
+        },
+    )
+
+    # Wait for the benchmark to complete
+    client.wait_for_completion(timeout=55)
+
+    # Assert no Python exceptions occurred
+    assert_no_python_exceptions(client.stderr)
+
+    # Load and validate the report
+    report = load_benchmark_report(report_path)
+    benchmark = report["benchmarks"][0]
+
+    # Check that the over-saturation constraint was triggered
+    assert_constraint_triggered(
+        benchmark, "over_saturation", {"is_over_saturated": True}
+    )
+
+    cleanup_report_file(report_path)

tests/e2e/utils.py

@@ -54,15 +54,18 @@ def start_benchmark(
         """
         Start a guidellm benchmark command.
 
-        :param rate_type: Type of rate control (constant, etc.)
+        :param profile: Type of rate control (constant, etc.)
         :param rate: Request rate
         :param max_seconds: Maximum duration in seconds
         :param max_requests: Maximum number of requests
         :param max_error_rate: Maximum error rate before stopping
         :param over_saturation: Over-saturation detection configuration (bool or dict).
+            When bool is True, passes --over-saturation=True to avoid Click parsing
+            issues.
         :param data: Data configuration string
         :param processor: Processor/tokenizer to use
         :param additional_args: Additional command line arguments
+        :param extra_env: Additional environment variables to set
         """
         guidellm_exe = get_guidellm_executable()
 
@@ -88,7 +91,7 @@ def start_benchmark(
         if over_saturation is not None:
             if isinstance(over_saturation, bool):
                 if over_saturation:
-                    cmd_parts.append("--over-saturation")
+                    cmd_parts.append("--over-saturation=True")
             elif isinstance(over_saturation, dict):
                 import json