Scribd
Upload
135 views35 pages

Asyncio for Python Developers

This document discusses using asyncio in production Python applications. It begins by explaining why asyncio is attractive due to its explicit and native asynchronous nature. It then covers issues that can arise when migrating existing applications to use asyncio, such as debugging challenges and ensuring dependencies are compatible. Benchmark tests comparing non-asyncio and asyncio versions of applications found significant performance improvements from using asyncio, especially when making database requests. Overall, the document argues that while migrating to asyncio requires effort to address new concepts and potential issues, it can provide substantial scalability benefits for production applications.

Uploaded by

Gabriel Pontes
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
135 views35 pages

Asyncio for Python Developers

This document discusses using asyncio in production Python applications. It begins by explaining why asyncio is attractive due to its explicit and native asynchronous nature. It then covers issues that can arise when migrating existing applications to use asyncio, such as debugging challenges and ensuring dependencies are compatible. Benchmark tests comparing non-asyncio and asyncio versions of applications found significant performance improvements from using asyncio, especially when making database requests. Overall, the document argues that while migrating to asyncio requires effort to address new concepts and potential issues, it can provide substantial scalability benefits for production applications.

Uploaded by

Gabriel Pontes
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 35

Asyncio in production

Hrafn Eiriksson
Why you no asyncio?!
Why you no asyncio?!

Asynchronous programming is different


Why you no asyncio?!

asyncio is relatively new


Why you no asyncio?!

Converting existing Python apps to use asyncio is not simple


Why you no asyncio?!

The community has built multiple concurrency libraries


Why you no asyncio?!

Asynchronous programming is not always what you want


My goal today
1. Discuss why I went all in on asyncio (and try to convince you to do the same)
2. Migrating to asyncio and the inevitable issues you run into
3. Asyncio in production: A before/after comparison

Not my goal today

4. An introduction into asyncio


Part 1: Why bother with asyncio?
My software already works!
A bit of background...
A bit of background...

Credit: https://www.destroyallsoftware.com/talks/wat
A bit of background...
How we (typically) scale our services

Instance 1
Gunicorn
worker 1
Instance 2
Gunicorn
Clients Gunicorn worker 2
Instance 3
master
Load
Gunicorn
balancer
Instance 4 worker 3

... ...
Gunicorn
Instance x
worker y

flow
TLDR

What makes asyncio so attractive is that it’s:

- Explicit
- Part of the language
Part 2: Migrating to asyncio
The asyncio ecosystem

Previously people relied on monkey patching

Now it seems to be becoming quite mature:

- Dozens of web frameworks (aiohttp, Sanic, Quart)


- Loads of database drivers (asyncpg, aiomysql, aioredis, etc)
- ...and way more [1]

[1] https://github.com/python/asyncio/wiki/ThirdParty
A microservice migration
Based on a true story
Asyncio web frameworks

Sanic
aiohttp
WebSockets
support

Mature
Flask-like

Fast

Flask compatible

HTTP 2.0
An example: Quart
Flask Quart [2]

[2] https://gitlab.com/pgjones/quart
aiohttp
A migration example: Sentry
Then problems hit...
Issues with asyncio
- A lot of new concepts to wrap your head around
- async/await everywhere
- Debugging asyncio code can be problematic
- Be wary of running synchronous code in async functions
TLDR

1. Map out your dependencies to see if asyncio-compatible versions exist


2. Experiment with asyncio versions of your dependencies
3. Watch out for asyncio gotchas
4. Profit!
Part 3: asyncio in production
Before/after comparison

Before: Flask + psycopg2 + eventlet

VS

After: aiohttp + asyncpg + uvloop


Methodology
- Use wrk (https://github.com/wg/wrk) to do HTTP benchmarking.
- Ran each configuration of the benchmark:
- For 30 seconds
- 10 times
- Using a variable number of open HTTP connections
- Noted the median and the 25%/75% latency for each run
- 10 seconds of sleep between runs
Local comparison: Simple ping

Requests/sec
Local comparison: Database access

Requests/sec
In production comparison

Requests/sec
Conclusion

So is asyncio worth the effort?


Thank you!

You might also like