With the impending release of Django 3.1, we have access to async web views in a widely deployed web framework with a very rich ecosystem.

What kinds of optimizations can we eke out of the system?

Here’s a common pattern that I think has a really easy speedup. The scenario involves making a fallible network request. If it fails, make another request to a fallback.

You may have seen this before, like when you have a cache in front of a database.

def main():
    result, is_cached = get_from_cache()
    if not is_cached:
        result = get_from_db()

    return result

But there’s a problem. This code doesn’t start the database request, until the cache request has, no matter the result, returned. In the case of a bad caching strategy, a cache miss might be very expensive, itself a long wait.

Here’s what’s now possible.

from asgiref.sync import async_to_sync, sync_to_async
from asyncio import CancelledError, create_task, sleep
from random import SystemRandom


@async_to_sync
async def main():
    # Call two functions at the same time
    from_cache = create_task(get_from_cache())
    from_db = create_task(get_from_db())

    # Read the first result
    result, is_cached = await from_cache

    if not is_cached:
        result = await from_db
        return result

    # Cache was successful, stop the db query
    from_db.cancel()
    try:
        # not using the db response, even if it has finished
        _ = await from_db
        print("db response completed. do you even need a cache?")
    except CancelledError:
        print("db query is cancelled")

    return result


async def get_from_cache():
    if random_bool():
        await sleep(1)

    simulated_failure = random_bool()
    return "from_cache", simulated_failure


async def get_from_db():
    if random_bool():
        await sleep(2)

    return "from_db"


def random_bool():
    return bool(SystemRandom().getrandbits(1))


if __name__ == '__main__':
    # `main()` can be called synchronously
    # because of the magic that is `async_to_sync`
    print(main())

Okay, this entire entry was so that I could play with async_to_sync.