Is async worthwhile?

One of the main motivations for building Piccolo was the lack of options for an asyncio ORM.

But is async worthwhile? This is what this article will explore.

What is asyncio?

asyncio is a library added in Python 3, to provide an event loop implementation in the standard library.

Prior to this, each framework that implemented non-blocking IO via an event loop had their own event loop implementation, limiting interoperability (Twisted and Tornado being by far the most well known).

Does my website need asyncio?

Most small websites can deliver a perfectly acceptable user experience without using async.

However, for building websites or APIs at scale, there are real benefits to using non-blocking IO.

Asyncio will help improve the throughput of a Python application. This means that a given server can handle more traffic, which can result in real cost savings.

Is asyncio all about speed?

Non-blocking IO won't make your website faster when under small load. For example, when only dealing with one sequential request at a time.

However, it does improve throughput, so under high load a user's request will be queued for less time, and they'll receive a response faster.

An interesting side effect of asyncio is it got library authors thinking about performance. By using efficient HTTP parsing, and Cython-ising slow parts, many asyncio libraries are actually faster than synchronous alternatives, but this isn't due to asyncio itself.

How much time does Python spend waiting on a database?

Even a simple database operation takes in the order of milliseconds (10^-3) to execute. Even though databases are highly optimised, they still have to parse and execute the SQL, which involves interacting with the disk.

In addition, there is also network lag when talking to a remote database, and overheads such as authentication and encryption.

Python isn't a fast language, but basic Python operations take in the order of microseconds (10^-6).

So there is time for Python to do meaningful work when waiting for a database response. The question becomes how much?

This is dependent on the overhead that asyncio imposes. If the asyncio event loop, and associated Python code required to schedule coroutines, is slow then it'll defeat the purpose.

Libraries such as uvloop are important in this regard, since they offer a faster event loop implemention, which is still compatible with asyncio.

Conclusions

The performance benefits of asyncio are real. The asyncio ecosystem is maturing fast, and for websites where performance is critical, asyncio makes a lot of sense.