[openapi server] log exception in exception handler(2/N) (#36201)

Signed-off-by: Andy Xie <andy.xning@gmail.com>

vllm/entrypoints/openai/chat_completion/api_router.py

@@ -50,10 +50,7 @@ async def create_chat_completion(request: ChatCompletionRequest, raw_request: Re
     )
     handler = chat(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Chat Completions API"
-        )
+        raise NotImplementedError("The model does not support Chat Completions API")
 
     generator = await handler.create_chat_completion(request, raw_request)
 

vllm/entrypoints/openai/completion/api_router.py

@@ -49,10 +49,7 @@ async def create_completion(request: CompletionRequest, raw_request: Request):
     )
     handler = completion(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Completions API"
-        )
+        raise NotImplementedError("The model does not support Completions API")
 
     generator = await handler.create_completion(request, raw_request)
 

vllm/entrypoints/openai/models/serving.py

@@ -7,7 +7,6 @@ from http import HTTPStatus
 
 from vllm.engine.protocol import EngineClient
 from vllm.entrypoints.openai.engine.protocol import (
-    ErrorInfo,
     ErrorResponse,
     ModelCard,
     ModelList,
@@ -18,7 +17,8 @@ from vllm.entrypoints.serve.lora.protocol import (
     LoadLoRAAdapterRequest,
     UnloadLoRAAdapterRequest,
 )
-from vllm.entrypoints.utils import sanitize_message
+from vllm.entrypoints.utils import create_error_response
+from vllm.exceptions import LoRAAdapterNotFoundError
 from vllm.logger import init_logger
 from vllm.lora.request import LoRARequest
 from vllm.lora.resolver import LoRAResolver, LoRAResolverRegistry
@@ -152,15 +152,15 @@ class OpenAIServingModels:
             try:
                 await self.engine_client.add_lora(lora_request)
             except Exception as e:
-                error_type = "BadRequestError"
-                status_code = HTTPStatus.BAD_REQUEST
-                if "No adapter found" in str(e):
-                    error_type = "NotFoundError"
-                    status_code = HTTPStatus.NOT_FOUND
-
-                return create_error_response(
-                    message=str(e), err_type=error_type, status_code=status_code
-                )
+                if str(
+                    LoRAAdapterNotFoundError(
+                        lora_request.lora_name, lora_request.lora_path
+                    )
+                ) in str(e):
+                    raise LoRAAdapterNotFoundError(
+                        lora_request.lora_name, lora_request.lora_path
+                    ) from e
+                raise
 
             self.lora_requests[lora_name] = lora_request
             logger.info(
@@ -292,17 +292,3 @@ class OpenAIServingModels:
                     err_type="NotFoundError",
                     status_code=HTTPStatus.NOT_FOUND,
                 )
-
-
-def create_error_response(
-    message: str,
-    err_type: str = "BadRequestError",
-    status_code: HTTPStatus = HTTPStatus.BAD_REQUEST,
-) -> ErrorResponse:
-    return ErrorResponse(
-        error=ErrorInfo(
-            message=sanitize_message(message),
-            type=err_type,
-            code=status_code.value,
-        )
-    )

vllm/entrypoints/openai/responses/api_router.py

@@ -59,10 +59,7 @@ async def _convert_stream_to_sse_events(
 async def create_responses(request: ResponsesRequest, raw_request: Request):
     handler = responses(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Responses API"
-        )
+        raise NotImplementedError("The model does not support Responses API")
 
     generator = await handler.create_responses(request, raw_request)
 
@@ -88,10 +85,7 @@ async def retrieve_responses(
 ):
     handler = responses(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Responses API"
-        )
+        raise NotImplementedError("The model does not support Responses API")
 
     response = await handler.retrieve_responses(
         response_id,
@@ -115,10 +109,7 @@ async def retrieve_responses(
 async def cancel_responses(response_id: str, raw_request: Request):
     handler = responses(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Responses API"
-        )
+        raise NotImplementedError("The model does not support Responses API")
 
     response = await handler.cancel_responses(response_id)
 

vllm/entrypoints/openai/speech_to_text/api_router.py

@@ -65,10 +65,7 @@ async def create_transcriptions(
 ):
     handler = transcription(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Transcriptions API"
-        )
+        raise NotImplementedError("The model does not support Transcriptions API")
 
     audio_data = await request.file.read()
 
@@ -101,10 +98,7 @@ async def create_translations(
 ):
     handler = translation(raw_request)
     if handler is None:
-        base_server = raw_request.app.state.openai_serving_tokenization
-        return base_server.create_error_response(
-            message="The model does not support Translations API"
-        )
+        raise NotImplementedError("The model does not support Translations API")
 
     audio_data = await request.file.read()