Optimize `asyncio.get_running_loop`

[kumaraditya303]

asyncio.get_running_loop is one of the most performance critical functions in asyncio. With #66285 the running loop can be reset in the fork handler thereby avoiding the need for the getpid checks. The whole running loop holder object is unnecessary now and adds up unnecessary dependent memory loads.

Benchmark:

import asyncio
from pyperf import Runner

async def main():
    for i in range(10000):
        asyncio.get_running_loop()


runner = Runner()
runner.bench_async_func("main", main)

Result:

Benchmark	base	patch
main	5.03 ms	328 us: 15.33x faster

The numbers speak for themselves.

Linked PRs

• GH-100363: Speed up asyncio.get_running_loop #100364

[gvanrossum]

asyncio.get_running_loop is one of the most performance critical functions in asyncio.

Do you have data about this? Does it show up in profiles of real functions, or do you have call counts or whatever? Usually I imagine the call is followed by a much more expensive computation.

[kumaraditya303]

This is based upon call counts and the fact that most functions now internally call this function as loop argument was removed from most asyncio APIs.

[gvanrossum]

Okay, makes sense.