async-python-patterns

Use asyncio.gather() to run multiple coroutines concurrently and collect their results as a list. For long-running work, create tasks with asyncio.create_task() and await them later, which lets the event loop interleave execution.

Create an asyncio.Semaphore with a maximum count and wrap each operation in 'async with semaphore'. This bounds how many coroutines run simultaneously, preventing resource exhaustion when calling external APIs at scale.

Use asyncio for I/O-bound work like network and database calls with many concurrent connections. Use multiprocessing for CPU-bound computation, since asyncio runs on a single thread and cannot parallelize CPU work.

Yes, wrap blocking calls with asyncio.to_thread() (Python 3.9+) to run them in a thread pool without stalling the event loop. Calling sync code directly inside a coroutine blocks all concurrent tasks.

Hangs usually come from forgetting await, blocking the event loop with time.sleep() or requests, or operations without timeouts. Use asyncio.wait_for() to enforce deadlines and asyncio.sleep() instead of time.sleep().

Use the pytest-asyncio plugin and mark test functions with @pytest.mark.asyncio. Tests can then await coroutines directly, including asserting that asyncio.TimeoutError is raised for slow operations.