Improve error handling and logging for streaming

src/litserve/server.py

@@ -19,6 +19,7 @@
 import logging
 import multiprocessing as mp
 import os
+import pickle
 import secrets
 import sys
 import threading
@@ -325,17 +326,23 @@ async def handle_request(self, request, request_type) -> Response:
             return response
 
         except HTTPException as e:
-            raise e
+            raise e from None
 
         except Exception as e:
-            logger.exception(f"Error handling request: {e}")
-            raise HTTPException(status_code=500, detail="Internal server error")
+            logger.error(f"Unhandled exception: {e}", exc_info=True)
+            raise HTTPException(status_code=500, detail="Internal server error") from e
 
     async def _handle_error_response(self, response):
-        logger.error("Error in request: %s", response)
+        """Raise HTTPException as is and rest as 500 after logging the error."""
+        if isinstance(response, bytes):
+            response = pickle.loads(response)
+
         if isinstance(response, HTTPException):
             raise response
 
+        if isinstance(response, Exception):
+            logger.error(f"Error while handling request: {response}")
+
         raise HTTPException(status_code=500, detail="Internal server error")
 
 
@@ -800,7 +807,7 @@ def launch_inference_worker(self, lit_api: LitAPI):
                     self.workers_setup_status,
                     self._callback_runner,
                 ),
-                name=f"lit-inference-{endpoint}_{worker_id}",
+                name="inference-worker",
             )
             process.start()
             process_list.append(process)
@@ -1363,9 +1370,9 @@ def _start_server(self, port, num_uvicorn_servers, log_level, sockets, uvicorn_w
             server = uvicorn.Server(config=uvicorn_config)
             if uvicorn_worker_type == "process":
                 ctx = mp.get_context("fork")
-                w = ctx.Process(target=server.run, args=(sockets,), name=f"lit-uvicorn-{response_queue_id}")
+                w = ctx.Process(target=server.run, args=(sockets,), name=f"LitServer-{response_queue_id}")
             elif uvicorn_worker_type == "thread":
-                w = threading.Thread(target=server.run, args=(sockets,), name=f"lit-uvicorn-{response_queue_id}")
+                w = threading.Thread(target=server.run, args=(sockets,), name=f"LitServer-{response_queue_id}")
             else:
                 raise ValueError("Invalid value for api_server_worker_type. Must be 'process' or 'thread'")
             w.start()

src/litserve/specs/openai.py

@@ -327,6 +327,26 @@ async def encode_response(self, output):
 """  # noqa: E501
 
 
+def _openai_format_error(error: Exception):
+    if isinstance(error, HTTPException):
+        return "data: " + json.dumps({
+            "error": {
+                "message": error.detail,
+                "type": "internal",
+                "param": None,
+                "code": "internal_error",
+            }
+        })
+    return "data: " + json.dumps({
+        "error": {
+            "message": "Internal server error",
+            "type": "internal",
+            "param": None,
+            "code": "internal_error",
+        }
+    })
+
+
 class OpenAISpec(LitSpec):
     def __init__(
         self,
@@ -486,82 +506,91 @@ async def chat_completion(self, request: ChatCompletionRequest, background_tasks
         return await response_task
 
     async def streaming_completion(self, request: ChatCompletionRequest, pipe_responses: List):
-        model = request.model
-        usage_info = None
-        async for streaming_response in azip(*pipe_responses):
-            choices = []
-            usage_infos = []
-            # iterate over n choices
-            for i, (response, status) in enumerate(streaming_response):
-                if status == LitAPIStatus.ERROR and isinstance(response, HTTPException):
-                    raise response
-                elif status == LitAPIStatus.ERROR:
-                    logger.error("Error in streaming response: %s", response)
-                    raise HTTPException(status_code=500)
-                encoded_response = json.loads(response)
-                logger.debug(encoded_response)
-                chat_msg = ChoiceDelta(**encoded_response)
-                usage_infos.append(UsageInfo(**encoded_response))
-                choice = ChatCompletionStreamingChoice(
-                    index=i, delta=chat_msg, system_fingerprint="", finish_reason=None
-                )
+        try:
+            model = request.model
+            usage_info = None
+            async for streaming_response in azip(*pipe_responses):
+                choices = []
+                usage_infos = []
+                # iterate over n choices
+                for i, (response, status) in enumerate(streaming_response):
+                    if status == LitAPIStatus.ERROR and isinstance(response, HTTPException):
+                        raise response
+                    elif status == LitAPIStatus.ERROR:
+                        logger.error("Error in streaming response: %s", response)
+                        raise HTTPException(status_code=500)
+                    encoded_response = json.loads(response)
+                    logger.debug(encoded_response)
+                    chat_msg = ChoiceDelta(**encoded_response)
+                    usage_infos.append(UsageInfo(**encoded_response))
+                    choice = ChatCompletionStreamingChoice(
+                        index=i, delta=chat_msg, system_fingerprint="", finish_reason=None
+                    )
 
-                choices.append(choice)
+                    choices.append(choice)
+
+                # Only use the last item from encode_response
+                usage_info = sum(usage_infos)
+                chunk = ChatCompletionChunk(model=model, choices=choices, usage=None)
+                logger.debug(chunk)
+                yield f"data: {chunk.model_dump_json(by_alias=True)}\n\n"
 
-            # Only use the last item from encode_response
-            usage_info = sum(usage_infos)
-            chunk = ChatCompletionChunk(model=model, choices=choices, usage=None)
-            logger.debug(chunk)
-            yield f"data: {chunk.model_dump_json(by_alias=True)}\n\n"
-
-        choices = [
-            ChatCompletionStreamingChoice(
-                index=i,
-                delta=ChoiceDelta(),
-                finish_reason="stop",
+            choices = [
+                ChatCompletionStreamingChoice(
+                    index=i,
+                    delta=ChoiceDelta(),
+                    finish_reason="stop",
+                )
+                for i in range(request.n)
+            ]
+            last_chunk = ChatCompletionChunk(
+                model=model,
+                choices=choices,
+                usage=usage_info,
             )
-            for i in range(request.n)
-        ]
-        last_chunk = ChatCompletionChunk(
-            model=model,
-            choices=choices,
-            usage=usage_info,
-        )
-        yield f"data: {last_chunk.model_dump_json(by_alias=True)}\n\n"
-        yield "data: [DONE]\n\n"
+            yield f"data: {last_chunk.model_dump_json(by_alias=True)}\n\n"
+            yield "data: [DONE]\n\n"
+        except Exception as e:
+            logger.error("Error in streaming response: %s", e, exc_info=True)
+            yield _openai_format_error(e)
+            return
 
     async def non_streaming_completion(self, request: ChatCompletionRequest, generator_list: List[AsyncGenerator]):
-        model = request.model
-        usage_infos = []
-        choices = []
-        # iterate over n choices
-        for i, streaming_response in enumerate(generator_list):
-            msgs = []
-            tool_calls = None
-            usage = None
-            async for response, status in streaming_response:
-                if status == LitAPIStatus.ERROR and isinstance(response, HTTPException):
-                    raise response
-                if status == LitAPIStatus.ERROR:
-                    logger.error("Error in OpenAI non-streaming response: %s", response)
-                    raise HTTPException(status_code=500)
-
-                # data from LitAPI.encode_response
-                encoded_response = json.loads(response)
-                logger.debug(encoded_response)
-                chat_msg = ChatMessage(**encoded_response)
-                usage = UsageInfo(**encoded_response)
-                usage_infos.append(usage)  # Aggregate usage info across all choices
-                msgs.append(chat_msg.content)
-                if chat_msg.tool_calls:
-                    tool_calls = chat_msg.tool_calls
-
-            content = "".join(msg for msg in msgs if msg is not None)
-            msg = {"role": "assistant", "content": content, "tool_calls": tool_calls}
-            choice = ChatCompletionResponseChoice(index=i, message=msg, finish_reason="stop")
-            choices.append(choice)
-
-        return ChatCompletionResponse(model=model, choices=choices, usage=sum(usage_infos))
+        try:
+            model = request.model
+            usage_infos = []
+            choices = []
+            # iterate over n choices
+            for i, streaming_response in enumerate(generator_list):
+                msgs = []
+                tool_calls = None
+                usage = None
+                async for response, status in streaming_response:
+                    if status == LitAPIStatus.ERROR and isinstance(response, HTTPException):
+                        raise response
+                    if status == LitAPIStatus.ERROR:
+                        logger.error("Error in OpenAI non-streaming response: %s", response)
+                        raise HTTPException(status_code=500)
+
+                    # data from LitAPI.encode_response
+                    encoded_response = json.loads(response)
+                    logger.debug(encoded_response)
+                    chat_msg = ChatMessage(**encoded_response)
+                    usage = UsageInfo(**encoded_response)
+                    usage_infos.append(usage)  # Aggregate usage info across all choices
+                    msgs.append(chat_msg.content)
+                    if chat_msg.tool_calls:
+                        tool_calls = chat_msg.tool_calls
+
+                content = "".join(msg for msg in msgs if msg is not None)
+                msg = {"role": "assistant", "content": content, "tool_calls": tool_calls}
+                choice = ChatCompletionResponseChoice(index=i, message=msg, finish_reason="stop")
+                choices.append(choice)
+
+            return ChatCompletionResponse(model=model, choices=choices, usage=sum(usage_infos))
+        except Exception as e:
+            logger.error("Error in non-streaming response: %s", e, exc_info=True)
+            raise HTTPException(status_code=500)
 
 
 class _AsyncOpenAISpecWrapper(_AsyncSpecWrapper):

src/litserve/utils.py

@@ -31,6 +31,10 @@
 
 logger = logging.getLogger(__name__)
 
+_DEFAULT_LOG_FORMAT = (
+    "%(asctime)s - %(processName)s[%(process)d] - %(name)s - %(levelname)s - %(filename)s:%(lineno)d - %(message)s"
+)
+
 
 class LitAPIStatus:
     OK = "OK"
@@ -128,7 +132,7 @@ def _get_default_handler(stream, format):
 
 def configure_logging(
     level: Union[str, int] = logging.INFO,
-    format: str = "%(processName)s[%(process)d] - %(name)s - %(levelname)s - %(message)s",
+    format: str = _DEFAULT_LOG_FORMAT,
     stream: TextIO = sys.stdout,
     use_rich: bool = False,
 ):