feat(log):add_request_and_response_log (#3373)

fastdeploy/entrypoints/openai/serving_chat.py

@@ -239,6 +239,7 @@ class OpenAIServingChat:
                                     prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=num_cached_tokens),
                                 )
                             yield f"data: {chunk.model_dump_json(exclude_unset=True)} \n\n"
+                            api_server_logger.info(f"Chat Streaming response send_idx 0: {chunk.model_dump_json()}")
                         first_iteration = False
 
                     output = res["outputs"]
@@ -273,6 +274,7 @@ class OpenAIServingChat:
                         logprobs=logprobs_res,
                         arrival_time=arrival_time,
                     )
+
                     if res["finished"]:
                         num_choices -= 1
                         work_process_metrics.e2e_request_latency.observe(
@@ -304,6 +306,9 @@ class OpenAIServingChat:
                     if len(choices) == max_streaming_response_tokens or res["finished"]:
                         chunk.choices = choices
                         yield f"data: {chunk.model_dump_json(exclude_unset=True)}\n\n"
+                        # 打印尾包
+                        if res["finished"]:
+                            api_server_logger.info(f"Chat Streaming response last send: {chunk.model_dump_json()}")
                         choices = []
 
                 if choices:
@@ -456,13 +461,15 @@ class OpenAIServingChat:
             prompt_tokens_details=PromptTokenUsageInfo(cached_tokens=final_res.get("num_cached_tokens", 0)),
         )
         work_process_metrics.e2e_request_latency.observe(time.time() - final_res["metrics"]["request_start_time"])
-        return ChatCompletionResponse(
+        res = ChatCompletionResponse(
             id=request_id,
             created=created_time,
             model=model_name,
             choices=choices,
             usage=usage,
         )
+        api_server_logger.info(f"Chat response: {res.model_dump_json()}")
+        return res
 
     def _create_chat_logprobs(
         self,