async/await Under the Hood

Python's asyncio uses a single-threaded event loop (similar to Node.js):

import asyncio

async def fetch(url):
    async with aiohttp.ClientSession() as session:
        async with session.get(url) as response:
            return await response.text()

async def main():
    # Run 10,000 requests concurrently
    tasks = [fetch(url) for url in urls]
    results = await asyncio.gather(*tasks)

asyncio.run(main())

The Event Loop

1. fetch(url1) → starts HTTP request → awaits response → YIELDS control
2. fetch(url2) → starts HTTP request → awaits response → YIELDS control
3. ... 10,000 requests in flight, one thread handling all
4. Response arrives for url1 → resume fetch(url1) → return data

When to Use What

ApproachBest ForConcurrencyMemory/thread

asyncioI/O-bound (HTTP, DB, files)10K+ concurrent~1KB per coroutine
threadingI/O-bound + legacy code~100 concurrent~8MB per thread
multiprocessingCPU-bound (parsing, ML)CPU cores~30MB per process

Why asyncio Was Faster for Scraping

Threading: 50 threads × 8MB = 400MB overhead, OS context switching

asyncio: 10K coroutines × 1KB = 10MB overhead, no context switching

Result: More concurrent connections, less overhead, faster completion

Common Pitfalls

❌ Blocking the event loop
async def handler():
    time.sleep(5)  # BLOCKS THE ENTIRE EVENT LOOP
    # No other coroutines can run for 5 seconds!

✅ Non-blocking
async def handler():
    await asyncio.sleep(5)  # Yields to event loop

❌ Running CPU work in async
async def process():
    result = heavycomputation(data)  # Blocks event loop

✅ Offload to process pool
async def process():
    loop = asyncio.geteventloop()
    result = await loop.runinexecutor(None, heavycomputation, data)