refac

backend/open_webui/env.py

@@ -565,6 +565,16 @@ try:
 except (ValueError, TypeError):
     AIOHTTP_CLIENT_TIMEOUT = 300
 
+# Optional between-chunks idle cap for streaming aiohttp requests.
+_stream_idle_timeout_raw = os.getenv('AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT', '')
+try:
+    AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT = int(_stream_idle_timeout_raw) if _stream_idle_timeout_raw else None
+except (ValueError, TypeError):
+    AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT = None
+
+if AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT is not None and AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT <= 0:
+    AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT = None
+
 
 # SSL verification for general outbound requests (OpenAI, OAuth, etc.).
 # Accepts "True", "False", or a path to a CA bundle file.

backend/open_webui/routers/ollama.py

@@ -48,7 +48,7 @@ from open_webui.utils.payload import (
     apply_model_params_to_body_openai,
     apply_system_prompt_to_body,
 )
-from open_webui.utils.session_pool import cleanup_response, get_session, stream_wrapper
+from open_webui.utils.session_pool import cleanup_response, get_client_timeout, get_session, stream_wrapper
 
 log = logging.getLogger(__name__)
 
@@ -130,7 +130,7 @@ async def send_request(
             data=payload,
             headers=headers,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
-            timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT),
+            timeout=get_client_timeout(stream=stream),
         )
 
         if not r.ok:

backend/open_webui/routers/openai.py

@@ -55,6 +55,7 @@ from open_webui.utils.payload import (
 )
 from open_webui.utils.session_pool import (
     cleanup_response,
+    get_client_timeout,
     get_session,
     stream_wrapper,
 )
@@ -1344,6 +1345,7 @@ async def generate_chat_completion(
                     part.get('text', '') for part in message['content'] if part.get('type') in ('input_text', 'text')
                 )
 
+    is_streaming_request = bool(payload.get('stream', False))
     payload = json.dumps(payload)
 
     r = None
@@ -1360,7 +1362,7 @@ async def generate_chat_completion(
             headers=headers,
             cookies=cookies,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
-            timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT),
+            timeout=get_client_timeout(stream=is_streaming_request),
         )
 
         # Check if response is SSE
@@ -1587,6 +1589,7 @@ async def responses(
     Routes to the correct upstream backend based on the model field.
     """
     payload = form_data.model_dump(exclude_none=True)
+    is_streaming_request = bool(payload.get('stream', False))
 
     idx = 0
     model_id = form_data.model
@@ -1637,7 +1640,7 @@ async def responses(
             headers=headers,
             cookies=cookies,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
-            timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT),
+            timeout=get_client_timeout(stream=is_streaming_request),
         )
 
         # Check if response is SSE
@@ -1707,6 +1710,7 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
             payload = json.loads(body)
         except (json.JSONDecodeError, ValueError):
             payload = None
+    is_streaming_request = bool(payload.get('stream', False)) if isinstance(payload, dict) else False
 
     idx = 0
     model_id = payload.get('model') if isinstance(payload, dict) else None
@@ -1758,7 +1762,7 @@ async def proxy(path: str, request: Request, user=Depends(get_verified_user)):
             headers=headers,
             cookies=cookies,
             ssl=AIOHTTP_CLIENT_SESSION_SSL,
-            timeout=aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT),
+            timeout=get_client_timeout(stream=is_streaming_request),
         )
 
         # Check if response is SSE

backend/open_webui/utils/session_pool.py

@@ -28,6 +28,7 @@ from typing import Optional
 
 import aiohttp
 from open_webui.env import (
+    AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT,
     AIOHTTP_CLIENT_TIMEOUT,
     AIOHTTP_POOL_CONNECTIONS,
     AIOHTTP_POOL_CONNECTIONS_PER_HOST,
@@ -39,6 +40,13 @@ log = logging.getLogger(__name__)
 _session: Optional[aiohttp.ClientSession] = None
 
 
+def get_client_timeout(stream: bool = False) -> aiohttp.ClientTimeout:
+    return aiohttp.ClientTimeout(
+        total=AIOHTTP_CLIENT_TIMEOUT,
+        sock_read=AIOHTTP_CLIENT_STREAM_IDLE_TIMEOUT if stream else None,
+    )
+
+
 async def get_session() -> aiohttp.ClientSession:
     """Return the shared aiohttp ClientSession, creating it lazily."""
     global _session
@@ -56,7 +64,7 @@ async def get_session() -> aiohttp.ClientSession:
         else:
             connector_kwargs['limit_per_host'] = 0  # aiohttp: 0 = unlimited
         connector = aiohttp.TCPConnector(**connector_kwargs)
-        timeout = aiohttp.ClientTimeout(total=AIOHTTP_CLIENT_TIMEOUT)
+        timeout = get_client_timeout()
         _session = aiohttp.ClientSession(
             connector=connector,
             timeout=timeout,