Agent Skills: Python Async

Python Async

Overview

Asyncio is a concurrency model designed for IO-bound and high-level structured network code. It uses cooperative multitasking on a single thread, allowing routines to pause while waiting for I/O, yielding control back to the event loop.

When to Use

• Network IO: Web scraping, API requests, and database queries.
• Web Servers: Handling thousands of concurrent connections (e.g., FastAPI).
• Concurrent Tasks: Running multiple independent IO-bound operations simultaneously.

Decision Tree

1. Is the task CPU-bound (e.g., heavy math, image processing)?
   • YES: Use multiprocessing instead.
   • NO: Is the task waiting for external resources (API, Disk)?
     • YES: Use asyncio.
2. Are you using time.sleep?
   • YES: Replace with await asyncio.sleep to avoid blocking the event loop.

Workflows

1. Concurrent Task Execution

1. Define multiple coroutine functions with async def.
2. Initiate the tasks using asyncio.gather(*coros) to run them concurrently.
3. Await the results to aggregate outputs efficiently.

2. Asynchronous Resource Management

1. Implement an asynchronous context manager using __aenter__ and __aexit__.
2. Use the async with syntax to ensure resources (like network connections) are opened and closed without blocking the loop.
3. Perform I/O operations inside the context using await.

3. Processing Async Streams

1. Create an asynchronous generator using yield inside an async def function.
2. Iterate over the generator using async for to process data as it becomes available.
3. Avoid materializing the entire sequence in memory to keep memory overhead low.

Non-Obvious Insights

• Cooperative Multitasking: Async is NOT parallelism; it is single-threaded. If one coroutine blocks (e.g., time.sleep), the entire program stops.
• Modern Entry Point: Always use asyncio.run(main()) for the main entry point; avoid manual event loop management in modern Python.
• Materialization Risk: Using async for is essential for large datasets to prevent OOM (Out of Memory) errors by processing items one by one as they arrive.

Evidence

• "asyncio is often a perfect fit for IO-bound and high-level structured network code." - Python Docs
• "Async I/O is a single-threaded, single-process technique that uses cooperative multitasking." - Real Python
• "The await keyword suspends the execution of the surrounding coroutine and passes control back to the event loop." - Real Python

Scripts

• scripts/python-async_tool.py: Examples of asyncio.gather and async for generators.
• scripts/python-async_tool.js: Equivalent JavaScript Promise.all and async iterator examples.

Dependencies

• asyncio (Standard Library)
• aiohttp or httpx (Recommended for async HTTP)

References

• references/README.md