fastapi-async-trap

The document is generated by co-pilot with GPT-4o

1. /fake-async Endpoint

   • Behavior: This endpoint is marked as async, but it directly calls the synchronous function services.heavy_computation_work(). Since heavy_computation_work() is synchronous and uses time.sleep, it blocks the event loop while it runs.

   • Why It Blocks: Even though the endpoint is async, the synchronous function (heavy_computation_work) runs in the main event loop, preventing other tasks from being processed until it completes.

   • Impact: This endpoint will block other requests (both async and sync) because the event loop is occupied by the blocking synchronous function.

2. /real-async Endpoint

   • Behavior: This endpoint is marked as async and calls the asynchronous function services.heavy_computation_work_async() using await. The heavy_computation_work_async() function uses await asyncio.sleep, which is non-blocking.

   • Why It Does Not Block: the await keyword allows the event loop to continue processing other tasks while waiting for the heavy_computation_work_async() function to complete. This ensures that the event loop remains free to handle other requests.

   • Impact: This endpoint will not block other requests because it uses non-blocking asynchronous code.

3. /sync Endpoint

   • Behavior: This endpoint is synchronous and calls the synchronous function services.heavy_computation_work(). However, FastAPI automatically runs synchronous endpoints in a thread pool executor.

   • Why It Does Not Block: FastAPI offloads the execution of synchronous endpoints to a separate thread in the thread pool. This means the synchronous function runs in a background thread, leaving the main event loop free to handle other requests.

   • Impact: This endpoint will not block other requests because it runs in a separate thread.