Asynchronous Methods¶
What are the different ways of asynchronously executing code within Python? What are their pros and cons? which one should I use in my testscript?
This section of the documentation attempts to answer the above questions systematically in an introductory manner, providing comparisons, insights and explanations.
Tip
the quick answer? For all intents and purposes, use multiprocessing. Not convinced? Read on.
asyncio¶
asyncio is a newly added module since Python 3.4. The main goal of this module is to provide native Python support for coroutines, event loops, tasks and asynchronous I/O.
asyncio, coroutines and event-loops are not true parallelism. They are concurrent in the sense that their states are kept independent of each other, but are not run in parallel: at any given time, only one of the coroutines is running (whilst the others are suspended), irregardless of the number of CPU/cores. Coroutines are not threads: whereas threads are run in parallel at the same time, coroutines collaborate and only run one at a time.
Despite not being true “parallelism”, where asyncio and coroutines truly shine is when operations are I/O bound: the program consistently waits for I/O instead of waiting for the CPU. It allows libraries to perform at “parallel like” performance without the limitations & overhead of multi-process and multi-thread.
Should you use asyncio? Most likely not. This module is intended not for the general user: it’s more designed for developers to write libraries on top of it for other developers to use, such as Gevent, Tornado, and Eventlet. For your day-to-day test automation libraries and scripts, using asyncio will only add further unneccesary complexity.
Note
asyncio is so new that it is still receiving updates, and is therefore provided on a provisional basis (eg, no backwards compatibility guarantee).
Threading¶
Python Threading module provides a very user-friendly interface to create, handle and manage threads within the current process. Threads are very lightweight and also natively shares memory and objects within the same process.
On the surface, threading may look promising. However, there are a few downsides to threading that makes it very dangerous and bug-prone:
cPython - Global Interpreter Lock
not interrutable and/or killable
thread safety is a major concern: all shared object accesses needs to be done using lock acquire/release.
race condition prone, drastically increases code complexity
A hanging thread, hangs forever in python: threads cannot be killed. The use of threads in Python requires careful designs & coding, revolving around the use of object locks & such in order to make your code thread-safe.
Should you use Threading? Most likely not. Everything you can do with threads, can also be done using Multiprocessing with much less code changes required.
Warning
pyATS infrastructure is not thread-safe. If you must, use threads only on your own libraries and functions.
Multiprocessing¶
Python Multiprocessing module allows users to spawn/fork child processes using an API interface similar to that of Threading (and thus also very easy to use). Here’s some highlights of the benefits of using multiprocessing:
separate memory space: no race conditions (except with external systems)
very simple, straightforward code
no Global Interpreter Lock, takes full advantage of multiple CPU/cores
child processes are easily interruptable/killable
Typically speaking, processes are slower to spawn than threads, and tends to also use a larger memory footprint (due to separate memory spaces). However, on Linux systems (which pyATS is designed to run on), process forks are extremely lightweight, and parent-child memory is mostly shared until it starts to differ (due to memory writes).
The main benefit of multiprocessing is that it doesn’t require much speciality code: because each process runs in its own, separate memory space, virtually all functions can be run in parallel. This, combined with the light-weightness of process forking makes it the perfect candidate for easy-parallelism in test automation.
Thus, for all intents and purposes, users of pyATS should be mostly focused with using Multiprocessing module and functionalities.
Final Thoughts¶
Each system exists for a reason: they each have their advantages and disadvantages. However, for the purposes of test-automation, weighing the pros and cons of each async method, Multiprocessing comes out easily ahead of the bunch.
Does this mean you should avoid asyncio and Threading altogether? No. It only means that unless you know exactly why you need to use them, stick with Multiprocessing.