api: Improve Gradio client stability.

app.py

@@ -16,6 +16,7 @@ from fastapi.responses import StreamingResponse, JSONResponse
 from pydantic import BaseModel, Field, ValidationError
 from gradio_client import Client
 from contextlib import asynccontextmanager
+import httpx
 
 logging.basicConfig(
     level=logging.INFO,
@@ -71,20 +72,31 @@ class SessionManager:
 
 session_manager = SessionManager()
 
+async def create_gradio_client_with_retry(url: str, max_retries: int = 3, backoff_factor: float = 0.5) -> Client:
+    for attempt in range(max_retries):
+        try:
+            client = Client(url)
+            _ = client.config
+            return client
+        except (httpx.ReadTimeout, httpx.ConnectTimeout, httpx.HTTPError, Exception) as e:
+            logger.warning(f"Attempt {attempt+1}/{max_retries} failed to create Gradio client: {e}")
+            await asyncio.sleep(backoff_factor * (2 ** attempt))
+    raise RuntimeError("Failed to create Gradio client after retries")
+
 async def refresh_client(app: FastAPI):
     while True:
         await asyncio.sleep(15 * 60)
         async with app.state.client_lock:
-            if app.state.client is not None:
-                try:
+            try:
+                if app.state.client is not None:
                     old_client = app.state.client
                     app.state.client = None
                     del old_client
-                    app.state.client = Client("https://hadadrjt-ai.hf.space/")
-                    logger.info("Refreshed Gradio client connection")
-                except Exception as e:
-                    logger.error(f"Error refreshing Gradio client: {e}", exc_info=True)
-                    app.state.client = None
+                app.state.client = await create_gradio_client_with_retry("https://hadadrjt-ai.hf.space/")
+                logger.info("Refreshed Gradio client connection")
+            except Exception as e:
+                logger.error(f"Error refreshing Gradio client: {e}", exc_info=True)
+                app.state.client = None
 
 async def clear_terminal_periodically():
     while True:
@@ -99,6 +111,11 @@ async def lifespan(app: FastAPI):
     app.state.session_manager = session_manager
     app.state.client = None
     app.state.client_lock = asyncio.Lock()
+    try:
+        app.state.client = await create_gradio_client_with_retry("https://hadadrjt-ai.hf.space/")
+    except Exception as e:
+        logger.error(f"Initial Gradio client creation failed: {e}", exc_info=True)
+        app.state.client = None
     app.state.refresh_task = asyncio.create_task(refresh_client(app))
     app.state.cleanup_task = asyncio.create_task(session_manager.cleanup())
     app.state.clear_log_task = asyncio.create_task(clear_terminal_periodically())
@@ -208,7 +225,7 @@ async def get_client(app: FastAPI) -> Client:
     async with app.state.client_lock:
         if app.state.client is None:
             try:
-                app.state.client = Client("https://hadadrjt-ai.hf.space/")
+                app.state.client = await create_gradio_client_with_retry("https://hadadrjt-ai.hf.space/")
                 logger.info("Created Gradio client connection on demand")
             except Exception as e:
                 logger.error(f"Failed to create Gradio client: {e}", exc_info=True)
@@ -420,7 +437,10 @@ async def router(request: Request):
         elif endpoint == "history":
             return await get_history(body.user, body.session_id)
         elif endpoint == "responses/cancel":
-            return await cancel_response(body.user, body.session_id, body.tool_choice if isinstance(body.tool_choice, str) else None)
+            task_id = None
+            if isinstance(body.tool_choice, str):
+                task_id = body.tool_choice
+            return await cancel_response(body.user, body.session_id, task_id)
         else:
             raise HTTPException(status_code=404, detail="Endpoint not found")
     else: