Use async model streaming for chat

ai_knowledge_base_v2/apps/backend/app/api/chat.py

@@ -1,7 +1,8 @@
 from __future__ import annotations
 
+import asyncio
 import json
-from collections.abc import Iterator
+from collections.abc import AsyncIterator
 
 from fastapi import APIRouter, Depends, HTTPException, Query
 from fastapi.responses import StreamingResponse
@@ -87,7 +88,7 @@ def stop(
     return api_success()
 
 
-def _chat_stream(payload: ChatCompletionRequest, db: Session, current_user: User) -> Iterator[str]:
+async def _chat_stream(payload: ChatCompletionRequest, db: Session, current_user: User) -> AsyncIterator[str]:
     config = load_chat_queue_config(db)
     queue_request = chat_queue_manager.request_slot(config)
     acquired = queue_request.status == "acquired"
@@ -111,7 +112,7 @@ def _chat_stream(payload: ChatCompletionRequest, db: Session, current_user: User
             activeCount=queue_request.active_count,
             waitingCount=queue_request.waiting_count,
         )
-        queue_request = chat_queue_manager.wait_for_slot(queued_token, config)
+        queue_request = await asyncio.to_thread(chat_queue_manager.wait_for_slot, queued_token, config)
         acquired = queue_request.status == "acquired"
         if queue_request.status == "timeout":
             yield _sse_event("error", message="排队等待超时，请稍后再试。")
@@ -125,10 +126,12 @@ def _chat_stream(payload: ChatCompletionRequest, db: Session, current_user: User
             activeCount=queue_request.active_count,
             waitingCount=queue_request.waiting_count,
         )
-        for chunk in ChatStreamService.stream_answer(db, current_user, payload.sessionId, payload.message):
+        async for chunk in ChatStreamService.stream_answer_async(db, current_user, payload.sessionId, payload.message):
             yield _sse_event("content", content=chunk)
     except HTTPException as exc:
         yield _sse_event("error", message=str(exc.detail))
+    except asyncio.CancelledError:
+        raise
     except Exception:
         yield _sse_event("error", message="AI 回复生成失败，请稍后再试。")
     finally: