跳到主要内容

并发与异步执行

FastAPI relies on an asynchronous event loop (asyncio). Blocking operations—such as CPU-heavy pandas calculations, synchronous database ORM calls, or synchronous network requests (yfinance, requests)—must never be executed directly on the main event loop thread.

The run_sync Helper (concurrency.py)​

Fonrex provides run_sync() in concurrency.py as a unified abstraction for executing blocking synchronous code safely inside worker threads:

from concurrency import run_sync

# Offloads a synchronous blocking function to the default ThreadPoolExecutor
result = await run_sync(sync_blocking_function, arg1, arg2, kwarg=value)

Internal Architecture​

FastAPI Event Loop (Async)
       │
       ├─► Async HTTP Router Endpoint (async def)
       │         │
       │         ├─► Calls run_sync(DatabaseService.get_asset, ticker)
       │         │         │
       │         │         ▼
       │         │   ThreadPoolExecutor Worker Thread
       │         │   (执行同步 SQLAlchemy ORM 查询)
       │         │         │
       │         │   ◄─────┘
       │         │
       │   ◄─────┘ Event Loop remains unblocked for other requests!
       │
       └─► Processes other concurrent HTTP / WS connections

Guidelines for Developers​

  1. Async Routers: Define router functions with async def.
  2. Synchronous Services: If a service method interacts with DatabaseService or yfinance, invoke it via await run_sync(service.method, ...) from the router.
  3. Async Services: Services that utilize redis.asyncio or httpx.AsyncClient (NewsService, CanaryMonitor) can be awaited directly without run_sync.