Update app.py

app.py

@@ -62,7 +62,6 @@ class KeyInfo:
 keys_info: Dict[str, KeyInfo] = {}
 chat_keys = []
 first_key = None
-
 # 初始化keys
 def init_keys():
     global keys_info, chat_keys, first_key
@@ -132,6 +131,64 @@ async def access_control(request: Request, call_next):
             raise HTTPException(status_code=403, detail="IP not allowed")
             
     return await call_next(request)
+
+# 流式响应生成器
+async def stream_generator(response):
+    async for chunk in response.aiter_bytes():
+        try:
+            # 解码字节为文本
+            chunk_str = chunk.decode('utf-8')
+            # 处理多行数据
+            for line in chunk_str.split('\n'):
+                if line.strip():  # 忽略空行
+                    if line.startswith('data: '):
+                        yield f"{line}\n\n"
+                    else:
+                        yield f"data: {line}\n\n"
+        except Exception as e:
+            print(f"Stream error: {str(e)}")
+            continue
+
+# 处理API请求
+async def handle_api_request(url: str, headers: dict, method: str = "GET", body: dict = None, for_chat: bool = False):
+    max_retries = 3
+    current_try = 0
+    
+    while current_try < max_retries:
+        try:
+            # 获取key
+            if for_chat:
+                key = get_chat_key()
+            else:
+                key = first_key
+                
+            if not key:
+                raise HTTPException(status_code=500, detail="No API keys available")
+                
+            headers["Authorization"] = f"Bearer {key}"
+            
+            async with httpx.AsyncClient(timeout=60.0) as client:
+                response = await client.request(
+                    method=method,
+                    url=url,
+                    headers=headers,
+                    json=body
+                )
+                
+                # 检查配额不足
+                if response.status_code == 429 or "insufficient_quota" in response.text.lower():
+                    print(f"Key {key} quota exceeded, trying next key...")
+                    set_key_cooling(key)
+                    current_try += 1
+                    continue
+                    
+                return response
+                
+        except Exception as e:
+            print(f"Request error: {str(e)}")
+            current_try += 1
+            if current_try == max_retries:
+                raise HTTPException(status_code=500, detail=f"API request failed after {max_retries} retries")
 # 添加静态文件支持
 app.mount("/static", StaticFiles(directory="static"), name="static")
 
@@ -243,48 +300,6 @@ async def add_key(request: Request):
         raise
     except Exception as e:
         raise HTTPException(status_code=400, detail=str(e))
-
-# 处理API请求
-async def handle_api_request(url: str, headers: dict, method: str = "GET", body: dict = None, for_chat: bool = False):
-    max_retries = 3
-    current_try = 0
-    
-    while current_try < max_retries:
-        try:
-            # 获取key
-            if for_chat:
-                key = get_chat_key()
-            else:
-                key = first_key
-                
-            if not key:
-                raise HTTPException(status_code=500, detail="No API keys available")
-                
-            headers["Authorization"] = f"Bearer {key}"
-            
-            async with httpx.AsyncClient(timeout=60.0) as client:
-                response = await client.request(
-                    method=method,
-                    url=url,
-                    headers=headers,
-                    json=body
-                )
-                
-                # 检查配额不足
-                if response.status_code == 429 or "insufficient_quota" in response.text.lower():
-                    print(f"Key {key} quota exceeded, trying next key...")
-                    set_key_cooling(key)
-                    current_try += 1
-                    continue
-                    
-                return response
-                
-        except Exception as e:
-            print(f"Request error: {str(e)}")
-            current_try += 1
-            if current_try == max_retries:
-                raise HTTPException(status_code=500, detail=f"API request failed after {max_retries} retries")
-
 # 聊天完成路由
 @app.post("/api/v1/chat/completions")
 async def chat_completions(request: Request):
@@ -313,6 +328,7 @@ async def chat_completions(request: Request):
             
         print(f"Chat completion response status: {response.status_code}")
         
+        # 处理流式响应
         if body_json.get("stream"):
             return StreamingResponse(
                 stream_generator(response),
@@ -320,10 +336,12 @@ async def chat_completions(request: Request):
                 headers={
                     "Cache-Control": "no-cache",
                     "Connection": "keep-alive",
-                    "Content-Type": "text/event-stream"
+                    "Content-Type": "text/event-stream",
+                    "X-Accel-Buffering": "no"  # 禁用 Nginx 缓冲
                 }
             )
         
+        # 处理普通响应
         return Response(
             content=response.text,
             media_type="application/json",
@@ -332,7 +350,10 @@ async def chat_completions(request: Request):
         
     except Exception as e:
         print(f"Chat completion error: {str(e)}")
-        raise HTTPException(status_code=500, detail=str(e))
+        error_detail = str(e)
+        if len(error_detail) > 200:  # 如果错误信息太长，只保留前200个字符
+            error_detail = error_detail[:200] + "..."
+        raise HTTPException(status_code=500, detail=error_detail)
 
 # 代理其他请求
 @app.api_route("/api/v1/{path:path}", methods=["GET", "POST", "PUT", "DELETE"])
@@ -360,6 +381,7 @@ async def proxy(path: str, request: Request):
         )
         
     except Exception as e:
+        print(f"Proxy error: {str(e)}")
         raise HTTPException(status_code=500, detail=str(e))
 
 # 健康检查路由