fix(responses): surface upstream error status on get instead of 500 (#32287)

litellm/litellm_core_utils/exception_mapping_utils.py

@@ -1944,7 +1944,7 @@ def _map_azure_exception(
             response=getattr(original_exception, "response", None),
             body=getattr(original_exception, "body", None),
         )
-    elif "invalid_request_error" in error_str:
+    elif "invalid_request_error" in error_str and getattr(original_exception, "status_code", None) in (None, 400):
         raise BadRequestError(
             message=f"AzureException BadRequestError - {message}",
             llm_provider="azure",
@@ -1986,6 +1986,14 @@ def _map_azure_exception(
                 litellm_debug_info=extra_information,
                 response=getattr(original_exception, "response", None),
             )
+        elif original_exception.status_code == 404:
+            raise NotFoundError(
+                message=f"AzureException NotFoundError - {message}",
+                llm_provider="azure",
+                model=model,
+                litellm_debug_info=extra_information,
+                response=getattr(original_exception, "response", None),
+            )
         elif original_exception.status_code == 408:
             raise Timeout(
                 message=f"AzureException Timeout - {message}",

litellm/llms/custom_httpx/llm_http_handler.py

@@ -2912,6 +2912,7 @@ class BaseLLMHTTPHandler:
 
         try:
             response = sync_httpx_client.get(url=url, headers=headers, params=data)
+            response.raise_for_status()
         except Exception as e:
             raise self._handle_error(
                 e=e,
@@ -2983,9 +2984,9 @@ class BaseLLMHTTPHandler:
 
         try:
             response = await async_httpx_client.get(url=url, headers=headers, params=data)
-
+            response.raise_for_status()
         except Exception as e:
-            verbose_logger.exception(f"Error retrieving response: {e}")
+            verbose_logger.debug(f"Error retrieving response: {e}")
             raise self._handle_error(
                 e=e,
                 provider_config=responses_api_provider_config,
@@ -3076,6 +3077,7 @@ class BaseLLMHTTPHandler:
 
         try:
             response = sync_httpx_client.get(url=url, headers=headers, params=params)
+            response.raise_for_status()
         except Exception as e:
             raise self._handle_error(e=e, provider_config=responses_api_provider_config)
 
@@ -3149,6 +3151,7 @@ class BaseLLMHTTPHandler:
 
         try:
             response = await async_httpx_client.get(url=url, headers=headers, params=params)
+            response.raise_for_status()
         except Exception as e:
             raise self._handle_error(e=e, provider_config=responses_api_provider_config)
 

tests/test_litellm/litellm_core_utils/test_exception_mapping_utils.py

@@ -649,3 +649,22 @@ def test_upstream_4xx_without_model_maps_to_bad_request():
 
     assert excinfo.value.status_code == 400
     assert "Cannot cancel a synchronous response." in excinfo.value.message
+
+
+def test_azure_404_with_invalid_request_error_type_maps_to_not_found():
+    from litellm.llms.base_llm.chat.transformation import BaseLLMException
+
+    original_exception = BaseLLMException(
+        status_code=404,
+        message='{"error": {"message": "Response with id \'resp_abc\' not found.", "type": "invalid_request_error"}}',
+    )
+
+    with pytest.raises(litellm.NotFoundError) as excinfo:
+        exception_type(
+            model=None,
+            original_exception=original_exception,
+            custom_llm_provider="azure",
+        )
+
+    assert excinfo.value.status_code == 404
+    assert "Response with id 'resp_abc' not found." in excinfo.value.message

tests/test_litellm/llms/custom_httpx/test_llm_http_handler.py

@@ -1745,3 +1745,95 @@ def test_sync_retrieve_file_content_raises_on_http_error():
         )
 
     assert exc_info.value.status_code == 404
+
+
+_UPSTREAM_NOT_FOUND_BODY = {
+    "error": {
+        "message": "Response with id 'resp_abc' not found.",
+        "type": "invalid_request_error",
+        "param": None,
+        "code": None,
+    }
+}
+
+
+def _async_handler_returning(status_code: int, body: dict) -> AsyncHTTPHandler:
+    handler = AsyncHTTPHandler()
+    handler.client = httpx.AsyncClient(
+        transport=httpx.MockTransport(lambda request: httpx.Response(status_code, json=body))
+    )
+    return handler
+
+
+def _sync_handler_returning(status_code: int, body: dict) -> HTTPHandler:
+    handler = HTTPHandler()
+    handler.client = httpx.Client(transport=httpx.MockTransport(lambda request: httpx.Response(status_code, json=body)))
+    return handler
+
+
+@pytest.mark.asyncio
+async def test_aget_responses_surfaces_upstream_error_status_instead_of_500():
+    client = _async_handler_returning(404, _UPSTREAM_NOT_FOUND_BODY)
+
+    with pytest.raises(litellm.NotFoundError) as excinfo:
+        await litellm.aget_responses(
+            response_id="resp_abc",
+            custom_llm_provider="azure",
+            api_base="https://test.openai.azure.com",
+            api_key="test-key",
+            api_version="2025-03-01-preview",
+            client=client,
+        )
+
+    assert excinfo.value.status_code == 404
+    assert "Response with id 'resp_abc' not found." in excinfo.value.message
+
+
+def test_get_responses_surfaces_upstream_error_status_instead_of_500():
+    client = _sync_handler_returning(404, _UPSTREAM_NOT_FOUND_BODY)
+
+    with pytest.raises(litellm.NotFoundError) as excinfo:
+        litellm.get_responses(
+            response_id="resp_abc",
+            custom_llm_provider="azure",
+            api_base="https://test.openai.azure.com",
+            api_key="test-key",
+            api_version="2025-03-01-preview",
+            client=client,
+        )
+
+    assert excinfo.value.status_code == 404
+    assert "Response with id 'resp_abc' not found." in excinfo.value.message
+
+
+def test_list_input_items_surfaces_upstream_error_status():
+    client = _sync_handler_returning(404, _UPSTREAM_NOT_FOUND_BODY)
+
+    with pytest.raises(litellm.NotFoundError) as excinfo:
+        litellm.list_input_items(
+            response_id="resp_abc",
+            custom_llm_provider="azure",
+            api_base="https://test.openai.azure.com",
+            api_key="test-key",
+            api_version="2025-03-01-preview",
+            client=client,
+        )
+
+    assert excinfo.value.status_code == 404
+
+
+@pytest.mark.asyncio
+async def test_alist_input_items_surfaces_upstream_error_status():
+    client = _async_handler_returning(404, _UPSTREAM_NOT_FOUND_BODY)
+
+    with pytest.raises(litellm.NotFoundError) as excinfo:
+        await litellm.alist_input_items(
+            response_id="resp_abc",
+            custom_llm_provider="azure",
+            api_base="https://test.openai.azure.com",
+            api_key="test-key",
+            api_version="2025-03-01-preview",
+            client=client,
+        )
+
+    assert excinfo.value.status_code == 404