Address CodeQL warnings

.gitignore

@@ -11,4 +11,4 @@ artifacts
 
 # Test exclusions
 qa/L0_openai/openai
-qa/L0_openai/tensorrtllm_models
+tensorrtllm_models

python/openai/openai/src/utils/triton.py

@@ -139,7 +139,7 @@ def get_output(response):
     if "text_output" in response.outputs:
         try:
             return response.outputs["text_output"].to_string_array()[0]
-        except:
+        except Exception:
             return str(response.outputs["text_output"].to_bytes_array()[0])
     return ""
 

python/openai/openai/tests/conftest.py

@@ -36,21 +36,23 @@
 TEST_PROMPT = "What is machine learning?"
 TEST_MESSAGES = [{"role": "user", "content": TEST_PROMPT}]
 TEST_TOKENIZER = "meta-llama/Meta-Llama-3.1-8B-Instruct"
+
+# Infer the test environment for simplicity in local dev/testing.
 try:
     import vllm as _
 
     TEST_BACKEND = "vllm"
     TEST_MODEL = "llama-3.1-8b-instruct"
 except ImportError:
-    pass
+    print("No vllm installation found.")
 
 try:
     import tensorrt_llm as _
 
     TEST_BACKEND = "tensorrtllm"
     TEST_MODEL = "tensorrt_llm_bls"
 except ImportError:
-    pass
+    print("No tensorrt_llm installation found.")
 
 if not TEST_BACKEND or not TEST_MODEL:
     raise Exception("Unknown test environment")

python/openai/openai/tests/test_chat_completions.py

@@ -49,7 +49,7 @@ def test_chat_completions_defaults(self, client, model: str, messages: List[dict
         assert message["content"].strip()
         assert message["role"] == "assistant"
         # "usage" currently not supported
-        assert response.json()["usage"] == None
+        assert not response.json()["usage"]
 
     def test_chat_completions_system_prompt(self, client, model: str):
         # NOTE: Currently just sanity check that there are no issues when a

python/openai/openai/tests/test_completions.py

@@ -46,7 +46,7 @@ def test_completions_defaults(self, client, model: str, prompt: str):
         #       or tested with dummy identity model.
         assert response.json()["choices"][0]["text"].strip()
         # "usage" currently not supported
-        assert response.json()["usage"] == None
+        assert not response.json()["usage"]
 
     @pytest.mark.parametrize(
         "sampling_parameter, value",

python/openai/openai/tests/test_observability.py

@@ -57,15 +57,6 @@ def test_startup_success(self, client):
         response = client.get("/health")
         assert response.status_code == 200
 
-    def test_startup_fail(self):
-        os.environ["TRITON_MODEL_REPOSITORY"] = "/does/not/exist"
-        with pytest.raises(Exception):
-            # Test that FastAPI lifespan startup fails when initializing Triton
-            # with unknown model repository.
-            app = init_app()
-            with TestClient(app):
-                pass
-
     ### Metrics ###
     def test_startup_metrics(self, client):
         response = client.get("/metrics")
@@ -96,3 +87,16 @@ def test_models_get(self, client, model):
         assert model_resp["object"] == "model"
         assert model_resp["created"] > 0
         assert model_resp["owned_by"] == "Triton Inference Server"
+
+
+# For tests that won't use the same pytest fixture for server startup across
+# the whole class test suite.
+class TestObservabilityCustomFixture:
+    def test_startup_fail(self):
+        os.environ["TRITON_MODEL_REPOSITORY"] = "/does/not/exist"
+        with pytest.raises(Exception):
+            # Test that FastAPI lifespan startup fails when initializing Triton
+            # with unknown model repository.
+            app = init_app()
+            with TestClient(app):
+                pass