fix: Cleanup pytest markers, enable gpu_0 tests on trtllm arm, reduce log noise (#6124)

Signed-off-by: Dmitry Tokarev <dtokarev@nvidia.com>

tests/fault_tolerance/migration/test_vllm.py

@@ -26,6 +26,7 @@ from .utils import DynamoFrontendProcess, run_migration_test
 logger = logging.getLogger(__name__)
 
 pytestmark = [
+    pytest.mark.fault_tolerance,
     pytest.mark.vllm,
     pytest.mark.gpu_1,
     pytest.mark.e2e,

tests/fault_tolerance/test_vllm_health_check.py

@@ -16,6 +16,11 @@ from tests.utils.payloads import check_models_api, completions_response_handler
 
 logger = logging.getLogger(__name__)
 
+pytestmark = [
+    pytest.mark.fault_tolerance,
+    pytest.mark.vllm,
+]
+
 
 class DynamoWorkerProcess(ManagedProcess):
     """Process manager for Dynamo worker with vLLM backend"""
@@ -123,7 +128,6 @@ def send_completion_request(
         raise
 
 
-@pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
 @pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)
@@ -181,7 +185,6 @@ def test_vllm_health_check_active(request, runtime_services):
                 )
 
 
-@pytest.mark.vllm
 @pytest.mark.gpu_1
 @pytest.mark.e2e
 @pytest.mark.model(FAULT_TOLERANCE_MODEL_NAME)

tests/kvbm_integration/test_determinism_disagg.py

@@ -37,6 +37,8 @@ from .common import check_module_available
 # Todo: enable the rest when kvbm is built in the ci
 pytestmark = [
     pytest.mark.kvbm,
+    pytest.mark.vllm,
+    pytest.mark.trtllm,
     pytest.mark.e2e,
     pytest.mark.slow,
     pytest.mark.gpu_2,

tests/planner/unit/test_prometheus.py

@@ -29,6 +29,7 @@ pytestmark = [
     pytest.mark.pre_merge,
     pytest.mark.unit,
     pytest.mark.planner,
+    pytest.mark.vllm,
 ]
 
 

tests/planner/unit/test_sla_planner_scaling.py

@@ -22,6 +22,7 @@ pytestmark = [
     pytest.mark.pre_merge,
     pytest.mark.unit,
     pytest.mark.planner,
+    pytest.mark.vllm,
 ]
 
 

tests/planner/unit/test_virtual_connector.py

@@ -18,6 +18,7 @@ pytestmark = [
     pytest.mark.pre_merge,
     pytest.mark.unit,
     pytest.mark.planner,
+    pytest.mark.sglang,
 ]
 logger = logging.getLogger(__name__)
 

tests/profiler/test_profile_sla_aiconfigurator.py

@@ -20,6 +20,10 @@ sys.path.insert(0, str(project_root))
 from benchmarks.profiler.profile_sla import run_profile  # noqa: E402
 from benchmarks.profiler.utils.model_info import ModelInfo  # noqa: E402
 
+pytestmark = [
+    pytest.mark.aiconfigurator,
+]
+
 
 # Override the logger fixture from conftest.py to prevent directory creation
 @pytest.fixture(autouse=True)
@@ -110,10 +114,11 @@ class TestProfileSlaAiconfigurator:
         with pytest.raises(ValueError, match="Database not found"):
             await run_profile(llm_args)
 
+    @pytest.mark.trtllm
     @pytest.mark.pre_merge
     @pytest.mark.parallel
     @pytest.mark.asyncio
-    @pytest.mark.gpu_1
+    @pytest.mark.gpu_0
     @pytest.mark.integration
     async def test_trtllm_aiconfigurator_single_model(self, llm_args):
         # Test that profile_sla works with the model & backend in the llm_args fixture.
@@ -124,17 +129,19 @@ class TestProfileSlaAiconfigurator:
     @pytest.mark.gpu_1
     @pytest.mark.integration
     @pytest.mark.nightly
+    # fmt: off
     @pytest.mark.parametrize(
         "backend, aic_backend_version",
         [
-            ("trtllm", None),
-            ("trtllm", "1.2.0rc5"),
-            ("vllm", None),
-            ("vllm", "0.12.0"),
-            ("sglang", None),
-            ("sglang", "0.5.6.post2"),
+            pytest.param("trtllm", None,          marks=pytest.mark.trtllm),
+            pytest.param("trtllm", "1.2.0rc5",    marks=pytest.mark.trtllm),
+            pytest.param("vllm",   None,          marks=pytest.mark.vllm),
+            pytest.param("vllm",   "0.12.0",      marks=pytest.mark.vllm),
+            pytest.param("sglang", None,          marks=pytest.mark.sglang),
+            pytest.param("sglang", "0.5.6.post2", marks=pytest.mark.sglang),
         ],
     )
+    # fmt: on
     @pytest.mark.parametrize(
         "hf_model_id",
         [

tests/router/test_router_e2e_with_sglang.py

@@ -29,6 +29,7 @@ MODEL_NAME = "silence09/DeepSeek-R1-Small-2layers"
 
 pytestmark = [
     pytest.mark.e2e,
+    pytest.mark.router,
     pytest.mark.sglang,
     pytest.mark.model(MODEL_NAME),
 ]

tests/router/test_router_e2e_with_trtllm.py

@@ -30,6 +30,7 @@ TRTLLM_BLOCK_SIZE = 32  # fixed internally to 32
 
 pytestmark = [
     pytest.mark.e2e,
+    pytest.mark.router,
     pytest.mark.trtllm,
     pytest.mark.model(MODEL_NAME),
 ]

tests/router/test_router_e2e_with_vllm.py

@@ -29,6 +29,7 @@ MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
 
 pytestmark = [
     pytest.mark.e2e,
+    pytest.mark.router,
     pytest.mark.vllm,
     pytest.mark.model(MODEL_NAME),
 ]