Threads vs Async: Has Asyncio Solved Concurrency? by Jacob Unna

In recent years asyncio has garnered much attention as a faster and easier alternative to threads. So do threads still have a place in the world? And what of other concurrency tech such as greenlets?

The fundamental problem solved by both async frameworks and threads is one of scheduling. This talk will put aside nitty gritty implementation details and go back to basics, examining how a cooperative scheduler (used by async frameworks) differs from a preemptive scheduler (used by threads). This basic understanding will clarify what is often considered a tricky subject and shed light on the pros and cons of each approach.

We will also see that even having decided on the async model, asyncio is not the only tool available. We will explore the pros and cons of greenlets, curio and trio as alternatives to the standard library's solution.

This talk is suitable for all levels from beginner to expert. The format will be simple and intuitive, with basic live examples of how to use threads and asyncio to do everyday tasks. You will come away not with an encyclopaedic knowledge of a specific toolchain, but with a deeper appreciation of the core concepts that power concurrency.

Transcript

1. 1 PyCon ZA 2020 Has Asyncio Solved Concurrency?
2. 2 ​ Jacob Unna ​ Software Engineer @ Deloitte Analytics & Cognitive ​ junna@deloitte.co.uk @JacobUnna
3. 3 What is Concurrency?
4. 4 Concurrency in Python − Threads − Asyncio
5. 5 Concurrency in Python − Threads − Asyncio − How do they work? − Which is better? − Does asyncio spell the end for threads?
6. 6 Scheduling
7. 7 Scheduling
9. 9 Cooperative Scheduling Asyncio
10. 10 Cooperative scheduling • Can be implemented at the application level • Everything must be non-blocking
11. 11 Cooperative scheduling • Can be implemented at the application level • Everything must be non-blocking Cooperation
12. 12 Preemptive Scheduling Threads
13. 13 Preemptive scheduling • Run any two bits of code at the same time • Handled by the operating system − Wouldn't be possible using pure Python − Operating system preempts running code i.e. freezes it mid-flow and switches the CPU's attention to something else • Doesn't get stuck on uncooperative code
14. 14 Today's Agenda − How do threads and asyncio work? − Which is better? − Does asyncio spell the end for threads?
15. 15 Which is better? Threads Asyncio Overhead when switching task High Low
16. 16 Which is better? Threads Asyncio Overhead when switching task High Low Easy to reason about No Yes
17. 17 Which is better? Threads Asyncio Overhead when switching task High Low Easy to reason about No Yes Handle blocking code Yes No
18. 18 Which is better? Threads Asyncio Overhead when switching task High Low Easy to reason about No Yes Handle blocking code Yes No
19. 19 Today's Agenda − How do threads and asyncio work? − Which is better? − Does asyncio spell the end for threads?
20. 20 Which is better? • Cheap • Safe • Easy to use • Can cut through solid wood
21. 21 Can I do this with asyncio? Decision flow Use asyncio Use threads Yes No
22. 22 ​ Asyncio ≠ Cooperative Scheduling
23. 23 ​ Because cooperative scheduling takes place at the application level, it is easy to create new frameworks
24. 24 ​ Because cooperative scheduling takes place at the application level, it is easy to create new frameworks ​ (Relative to creating a new operating system)
25. 25 Asyncio
26. 26 Gevent Tornado uvloop Twisted Asyncio PyZMQ Curio Trio Diesel
27. 27 Gevent Tornado uvloop Twisted Asyncio PyZMQ Curio Trio Diesel
28. 28 ​ Questions? ​ Jacob Unna ​ junna@deloitte.co.uk @JacobUnna Code: git.io/JU6FT