Next-Level Concurrent Programming In Python With Asyncio

[@ArjanCodes]

💡 Get my FREE 7-step guide to help you consistently design great software: arjancodes.com/designguide.

[@MrTrebor2]

Let's go deeper in async. Exceptions handling, pubsub pattern and how to manage back pressure.

[@lawrencedoliveiro9104]

Nothing specific to asyncio there.

[@josephlangford2912]

I'll show you how to handle back pressure

[@infernape716]

​@@josephlangford2912💀

[@jasonmogavero3709]

​@@josephlangford2912 gottem!

[@arthurguz2672]

Regarding your criticism about nested contextmanagers (at 17:35)... Since Python3.10 you are able to write something like this: async with ( aiohttp.ClientSession() as session, session.get(url) as response, ): return await response.json() This is super usefull :)

[@jeroenvermunt3372]

Thanks a lot for this! Why do I have not seen this before. Is it just me or is python so much getting so much cleaner as javascript for these types of thing (I'm a beginner at javascript so totally biased, I think it looks ridiculous)

[@kosmonautofficial296]

woah cool!

[@user-hn6cc2tg2j]

Thanks a lot for this

[@shivampundir1829]

With python 3.11 we can now use Taskgroups as well.

[@AhriaHasegawa]

What! This is so cool! Thank you for this!

[@BladeAndrade]

I'd be great to see a series of concurrency/parallelization in python.

[@TNeulaender]

He should publish it at once.

[@reubenthomas1033]

YES!!!

[@lawrencedoliveiro9104]

15:13 You could leave out both the “await” and the “async”. So “send‗async‗request” becomes a regular function which returns an awaitable.

[@indraxios]

Best video on async, finally someone who shows why , when and how to use async. Cleared all my doubts, Thank you

[@cetilly]

Love, love, love your presentation style; so systematic, well thought out, and articulate.

[@sval4020]

Very helpful video, Arjan! Exactly what I was looking for! I will be looking forward to even more asyncio videos from you! Thank you!

[@ArjanCodes]

Glad to hear it was helpful!

[@CaptainCsaba]

I can't count the amount of times I've tried to find the best method for asnyc requests. I've used aiohttp just like you but it never felt right, but I never understood asyncio enough so this was always something I was scared of more than I should have. I am so happy that you have actually shown the best way to do it, it will help a lot in my everyday work!

[@CharleswoodSpudzyofficial]

Woah... I'm blown away. Another new level for my Python code career

[@ArjanCodes]

Thanks so much Charles, glad the content is helpful!

[@simmonslucas]

I would love more examples using queues and event loops

[@ЕкатеринаФедорова-ь2ч]

Thanks!

[@ArjanCodes]

Thank you so much, Екатерина! I appreciate your support!

[@julienstanguennec9563]

Another video on parallel programming could be very nice ! Thanks for all the material you provide, it's really helpful.

[@ArjanCodes]

Thanks so much, Julien, glad the content is helpful!

[@demolazer]

Just what I needed. I've got an app making a bunch of one by one API and SSH calls to my server which takes forever right now.

[@ArjanCodes]

I'm glad the video was helpful!

[@assaidy]

you provide really good examples in your tutorials. Amazing channel.

[@ArjanCodes]

Thank you so much! Glad you like the content.

[@mentefria98]

this was so hepful, thanks.

[@ArjanCodes]

Thanks so much, glad the content is helpful!

[@MrJjthecronfan]

Thanks so much - Learning a lot from here . A friend of mine recommended your channel , now I am passing it on to others . Good Luck

[@leoff2]

you didn't mention the difference between using `requests` + `to_thread`, and `aiohttp`. Although `aiohttp` is more verbose, it is faster since it involves no threads. You can also reduce indentation by creating the session beforehand and reusing it.

[@majorbadidea]

I dont think he understands..

[@saketkr]

This was really very helpful. Can't be not selfish to ask you for more of these. Thanks! 😌

[@ArjanCodes]

Glad it was helpful!

[@JordanBonnett]

All other videos and tutorials I saw show the same example with asyncio.sleep()... Thanks for create a custom example a explain very well this tricky subject of python asynchronous programming. +100

[@ATXAdventure]

I always check out Arjan when I need to brush up on ways to accomplish things in python.

[@PiletskayaV]

Great! I hope more videos about asyncio are coming ^___^

[@ArjanCodes]

On it 😉

[@sebastianzander87]

Since you said that you "don't like these nested 'with' statements" 17:36, I remember there is a Python feature probably not so well known: you can declare multiple variables in one 'with' statement by delimiting them by commas 🙂

[@eldarmammadov7872]

great explanation with giving right examples

[@ArjanCodes]

Glad you liked it, Eldar!

[@JohnMatthew1]

As always, I always learn something from your videos, THANK YOU. The to_thread() is new to me, very cool Too much sun? Nice color.

[@kunvensoong6047]

Great video! Looking forward for a video that refactors sync+unittest to async+unittest

[@ArjanCodes]

Thank you, glad you liked it!

[@re.liable]

That type union is dope :O Haven't coded in Python for a while; been trying to get into JS frameworks. I love that union type :D

[@Theeoldmann]

IDK how or why, but the subject matter of your videos seem to show up (KZbin suggestion page) exactly when I need a bit more understanding of what I'm actually working with at the time. The providence is well received and please keep making great videos. Thank you

[@ArjanCodes]

Wow, thank you!

[@BlackCloudsAC]

Hey Arjan, thanks for posting this! You've opened my eyes to asyncio. I was wondering if you could explain when you'd use async methods vs when you would use Multi Threading through a ThreadPool? Many thanks!

[@mouhammaddiakhate6065]

Good question

[@draufunddran]

I would like to have more videos about concurrency and asynchronous stuff... Please don't wait too long for the video about this topic.

[@nateriver8261]

Very helpful video, Arjan!

[@ArjanCodes]

Thanks Nate, glad it was helpful!

[@sugo8920]

I'm in love with async python now. Almost all apps will use other services. Almost everything uses REST now. This greatly speeds up by code and it more fun than sync programming. I recommend integrating your applications with Datadog if you can with APM enabled. Very satisfying to see parallel traces and great for optimizations as it can be used to uncover things that one may have thought were running in parallel but were not. Last is, to_thread is not a complete substitute for using aiohttp. Threads will start up slighly slower and I noticed that only 8 threads were running at one time when I used, where as if I used aiohttp everything ran at the exact same time and I was was able to run 100s of request in parallel were as with to_thread I was limited. Perhaps there is some env var i could have set to prevent the 8 thread limit but I could not find any docs on the topic.

[@lawrencedoliveiro9104]

12:20 By the way, forgetting the “await” in front of “asyncio.sleep()” is a mistake I have made a few times. How about you?

[@성수-o6h]

Thank you for very helpful video. Mewtwo is one of legendary poketmon, but you got it with just two choices!

[@fartzy]

I think one way to say it - to say parallelism vs concurrency - is to say the system parallelizes. However, the system handles concurrency. The system chooses to parallelize. The system reacts to concurrency.

[@VictorNedilko]

Great tutorial and fantastic way of presenting. Thanks

[@ArjanCodes]

Thank you!

[@walis85300]

Thanks for improving the way the video looks

[@experimentalhypothesis1137]

more async videos please, lets go deeper with u - u can explain this stuff very well

[@ArjanCodes]

Noted! Thanks ;)

[@aashayamballi]

Loving it! Thanks Arjan.. this video was very informative, would like to see more videos on async python 🙏

[@ArjanCodes]

Thank you, glad you liked it! I’ll surely revisit async in the future.

[@Whatthetrash]

Thanks so much! Love to see your videos. Thanks so much for making and sharing them! ^_^

[@sudadal-ajili6205]

Hey Arjan! Thank you for such amazing contents I have learned a lot from your videos. It will be great to see videos about Threading and Multiprocessing. Thank you again.

[@EekaMau5]

Thank you ! this tutorial was awesome !

[@ArjanCodes]

Glad you liked it!!

[@guplex]

Thanks! Would love more asyncio content

[@tobilobaogunware6686]

Hi Arjan, your videos have been of really great help to me. If possible, a video about using the Aiohttp module with asyncio would be really nice. Fingers crossed. Thank you for all you do!

[@AMMullan]

Amazing video and so well explained! I’ve been trying to get a better understanding of how to use asyncio and it’s not easy but you’ve done an incredible job. Will definitely be checking out your other videos. Thanks man!!!

[@ArjanCodes]

I'm really glad it was helpful! Welcome to the channel :)

[@blessedday6808]

1:47 gives a real life matephor of how asyncio computation compares to parallel computation 3:57 explains what async def means I greatly appreciate that you used pokemon in this video. It really helps me to see with clarity the difference between the actual subject matter. Also 10:28 Rage will most likely yield a Cubone or Gyrados! This video makes me want to sharpen up my list comprehension skills. Thank you for explaining the difference between parallel and concurrent code execution. That really made a lot of sense. Especially bringing up the fact that computers use different amounts of cores in a cpu chip. I didn't know that python can manipulate HOW our chips process our code. I learned something new here.

[@huonglarne]

this is incredibly helpful. Thank you so much

[@ArjanCodes]

Thanks so much, Mo Gung, glad the content is helpful!

[@thghtfl]

Thank you so much, man! Brilliant explanation!

[@ArjanCodes]

I'm glad it helped!

[@valentinasinichenko751]

this has been extremely helpful

[@robertmielewczyk4219]

It would be great to see a video about exception handling with asyncio, and forever running tasks using asyncio.

[@PaulSebastianM]

Concurrency and Parallelism are two steps towards multi-threading (green or not). Concurrency is the first step, and means being able to define and start more than one function at a time without waiting for a return value at the call site. The two functions can run in series or in parallel, on one shared thread or two separate threads, it all depends on the system and hardware and load. They also run in the same process that started them. Parallelism is the second step, it is Concurrency but with an explicit request to try and run truly in parallel, by explicitly using an operating system thread for each function call. But it actually does not imply that the functions will truly run at the same time. It's up to the system to schedule and run these threads, which could be in parallel or in series if there is too much load on the system.

[@Krzako]

Just one thing. You compare sync version with print in loop vs async with just one print. Of course it doesn't change the fact that async is much faster but to get more accurate results we have to remember to measure functions without unwanted code like slow prints ;-) Anyway great video. I'm mid developer that switched to python from other scripting language. I recently found your channel and liked it very much. I'm learning a lot from you. Thank you for all your great work!

[@ArjanCodes]

Thank you!

[@dmytrokorbanytskyi1586]

Thanks for the video! I really like your coding style )

[@ArjanCodes]

Thank you - glad you like the video!

[@seeds_of_growth-yi5gx]

I really like the practical approach of exploring the topic, it's my favorite way of learning things. Keep going! just subscribed and signed up for the diagnosis course, eager to start and jump to the next ones. Thanks for sharing

[@michtesar]

That is very useful and interesting video discussing asyncio library. I was about to read the docs, but now I think I don't have to, thanks to your video! Great content! Thanks for that!

[@getpoked101]

Great video. We are doing a bunch of data ingestion from a rest api right now so this is very top of the mind.

[@riptorforever2]

Great lesson. Thanks

[@ArjanCodes]

Thank you, glad you liked the video!

[@anandptyagi5275]

Hello, Thank you for the information. I like your way of teaching, and I would like to see more advanced topics in python explained by you, If possible please do. Have a good day, thanks.

[@tharo20]

This video helped me a lot! thank you so much!💌

[@k98killer]

Comment before watching the full video: I wrote my first bit of code using asyncio (an example implementation of a library I've built) just a few days ago after watching your previous videos on asyncio. Very cool. Building an interactive CLI tool that runs a networking application in the background has been on my to-do list for a long time, so I'm quite happy I figured it out. On a separate topic, have you considered making a video on publishing python packages to PyPI?

[@71sephiroth]

Pfff... nobody knows the struggle of getting Mewtwo on the first try. ;) Thanks for the comprehensive introduction!

[@burakcelik3752]

TL;DR # Excellent, thank you for the great content.# I really appreciate your work as thinking how much time you spent to create high quality contents like this. I watched a lot of videos when I started coding but %90 of them was like demonstrating an isolated feature of Python using a Jupyter Notebook. I accept that I learned so much things from them but here was the problem. Even if I learned so much deep, for instance using generators as coroutines that is what basically what asyncio implements now, but I'd never managed to get that advanced techniques together I later realize that was because I learned all the things in isolation and would stuck when it comes to use them all together, that is, the links were missing. Arjan, you did a great job starting to talk about design patterns that I realized the missing links were not the techniques but the philosophy. Therefore I can't recommend your channel enough for people who stuck in the intermediate level of coding and don't know why can't take a step further regardless of whatever new things they learn. I know there must full of people in that situation and think a great deal of them gives up in the end. Your channel is just a treasure for the people in subject and must be spread. As for concurrency, I've always found using preemptive multitasking far from being elegant because it inherently is a brute force approach and I can't imagine anybody who don't love asyncio s/he learns the philosophy behind it. Since we don't have too many options for cpu bound problems because of GIL even if we tackle the problem by splitting it independent sub methods, we can do much for IO bound problems using asyncio and I would appreciate if you consider to continue on the subject. For instance, how can we develop our own cooperative multitasked classes aside from using the external ones (except sending the couroutines to threads). For instance, think a project that used requests-http for its internal processes but doesn't explicitly support asyncio for its public methods. It's obvious that it has capacity for doing that and we can branch and give it our own asyncio support with a little effort and it would be great to learn how an async class should be designed philosophically. The last word is thank you for your great channel and efforts again.

[@mahdiasadnejad3804]

Thank you, it was great . let's go deeper!

[@ArjanCodes]

On it 😉

[@LittleCrazy2life]

if possible make a video on writing test cases on async functions!

[@ArjanCodes]

Thanks for the suggestion!

[@TheEmrobe]

Awesome tutorial and all, but what really blew my mind was the existence of a Pokemon-API :D

[@ArjanCodes]

I know! After discovering that, my life is complete.

[@therealslimaddy]

Very useful, thanks 🙏🏼

[@holographicsol2747]

really helpful thanks man

[@ArjanCodes]

Glad to hear it!

[@djl3009]

Nice one. Thanks!

[@EricNeilson_]

When you introduce aiohttp and describe the differences, this is hilarious. For a relatively noob like me, it's a lot to digest :) But it's great.

[@Obi-Wan-Kenobii]

Thanks for covering async.. I have once written an aiohttp based webservice to retrieve data from neo4j.. Aiohttp github page have some nice async libraries collections.. I hope more developers will release async libraries as well..

[@kahvelikod]

Very useful , Thank you.

[@rikschaaf]

15:27 This looks a lot like java's fork/join. The create_task is like the fork, the await is like the join

[@MujtabaMuhammadKhushtaba]

awesome video, looking forward to new videos

[@ArjanCodes]

Thank you!

[@lawrencedoliveiro9104]

4:13 “await” is actually an expression, not (necessarily) a statement.

[@jorgev4656]

hi. whats the difference between multiprocesing, and asyncio ?

[@dewciu4904]

Thank you for your movies! You make me better python developer :)

[@ArjanCodes]

You’re welcome - glad you like them!

[@DeepakYadav-ic2uv]

this was helpful!

[@ArjanCodes]

Thanks so much Deepak, glad the content is helpful!

[@ArturLuckyMan]

Simple brilliant)😎👍

[@ArjanCodes]

Thank you, glad you liked it!

Did not know asyncio.to_thread, have been using loop.run_in_executor, which requires more setup. Also, reusing the session will save you quite a bit of time when sending many requests to the same end point, as it will reuse the connection. At work, saving the requests.Session halved the runtime of the application, which was was enough and save us the port to asyncio...

[@leoff2]

to_thread is new from Python 3.8 on, if I'm not mistaken

[@fox123-t8h]

First: Great video as always, Arjan! I would really love to see some advandced asyncio topics on your channel. Some additional thoughts on Álvaro's post and the video: 1. Using a session with aiohttp or e.g. httpx brings the same benefits - especially if you have things like authentication in your requests, too. 2. Using asyncio.to_thread has two downsides: First you have the overhead for creating a thread, which consumes resources. Second Python thread execution is handeled by the OS, so you effectivly have no real control, when execution of the thread is interrupted and continued. This takes away some of the control over the program flow from the programmer - while using native coroutines are a bit more predictable. Apart from fast concurrent programs this is something I like most about asyncio - having so much control. So i personally would always prefer a native async library over using a synchronous one with asyncio.to_thread or loop.run_in_executor. If there's no quality alernative for a synchronous library, using asyncio.to_thread a great way to make you code asynchronous, absolutely agreed. 3. If you don't like deep nested context managers, there are some alternatives: not sure for aiohttp but for httpx you can use: client = httpx.AsyncClient() await client.aclose() With a httpx you also have only have one contextmanager (not two) using a session this way: async with httpx.AsyncClient() as client: r = await client.get('www.example.com/') As context managers are very common in async libraries (e.g. if you use sqlalchemy) you could also an AsyncExitStack to enter many contexts without having to indent every time. 4. You meantioned async Python getting easier with Python 3.10. Was there something added in this version for asyncio? I really can't remember - or did you mean it more generally with the features added in versions before? I hope this post does not sound to negative. I really love your videos and I'm very thankfull for learning so much from them! Just wanted to point out some things to consider. One last question: Are there are payment options aparart from credit card on your homepage or would you consider to add some other options in the future?

[@the1GeniX]

Thanks for this!

[@kosmonautofficial296]

Try out concurrent.futures.ThreadPoolExecutor() it may be even easier. The downside to using .to_thread is that it is using threads to get concurrency and passing the operation to the OS rather than handling it manually. The benefit of asyncio is that you have concurrency in a single thread and you are controlling it but to do that you need to await a function that supports async operations. For that you would need to use aiohttp or httpx instead of requests which relies on the socket library which is blocking and doesn't support async.

[@FEchtyy]

so helpfull! thanks a lot

[@_scourvinate]

17:35 the httpx library does async requests with one context manager no nesting, could be a better option.

[@NostraDavid2]

I hope one of your future videos goes into what the async keyword actually does with your code. Sure, I know that *now*, as I worked with async before out of necessity (which was a great learner for me), but before that I was completely befuddled on what that keyword does and why it made my code stop running unless I used something like async.gather().

[@Hubert4515]

cool as always :)

[@ArjanCodes]

Thank you!

[@MeetJPatel-f5q]

go more deep for async iterators and generators. what are those and how can it be beneficial?

[@tiktuktiktuk]

Would like to see more on async..

[@pypypy4228]

7:40 Could you please explain why would we want this async syntax for doing the task in a pretty much synchronous way?

[@tutobab]

http_get_sync is not an appropriate name for code that executes sequentially, line by line, waiting for all locks. Synchronized skating is when two athletes perform the same movement at the same time, in sync. The better their movements and trajectories match, the more perfect the synchronization. Synchronization in programming is precisely the simultaneous execution of two threads or processes on two different CPU cores. We won't overload this definition by requiring that both processes finish at exactly the same time; I think the previous criterion is sufficient. Thank you for your videos; I enjoy watching them.

[@EnglishRain]

What an excellent explanation! Subscribed!

[@MrEloska]

Could you make some advanced video on GIL? Where you can explain how it works, how to "bypass" it etc?

[@jochinq3116]

Awesome, thanks for this video! maybe you would like to make a video about asyncio and websocket?🤩

[@ArjanCodes]

Thank you, for suggestion, I have put it in my list.

[@kevon217]

I’m just starting to get into async. Have you thought about doing a tutorial on async use in the context of LLM calls and frameworks?

[@goodnewsjohn2482]

really nice video. please can you do something with celery and gevent

[@mtfrsantos]

How to use 'asyncio.to_thread' is different from adding 'await asyncio.sleep(0.01)' before 'response = requests.get(url)' and than calling both from 'asyncio.gather'? 15:56

[@sumaryze]

Thanks for your really useful video. Can you do another video on specifying which part of the program to run concurrently and sequentially?

[@johanvanderploeg1101]

Thanks for the clear Pokemon example! Do you know why at 8:22 the following code won't work: asyncio.run(asyncio.gather(*[get_random_pokemon_name() for _ in range(20)])) The function asyncio.run() takes a coroutine and asyncio.gather() returns a coroutine so that should match. Wrapping asyncio.gather() into a function main() seems unnecessary because when control is passed temporarily back from the main() function to the event loop there are no other functions to be run by asyncio.run().

[@munteanionut3993]

11:08 - "..., but it still run these things in order". If it is like this, then why use the "async" keyword in the for-loop? Does it have any effect then? I presume a syntax error would be thrown if we didn't use it. But if it doesn't have any effect, then why use it?

[@yparwani]

I also wished that this was explained more - I think it has to do with the different calls being added to the "event loop" which gather does automatically.

[@ArjanCodes]

It can still be useful to do this, for example if your loop goes through the steps of a pipeline, and the next step depends on the outcome of the previous one. By making the individual steps async, you can use the await syntax within the step itself and benefit from concurrent programming there.

[@munteanionut3993]

@@ArjanCodes thank you!

[@lawrencedoliveiro9104]

async-for needs the additional ‗‗aiter‗‗ mechanism in order to work, ‗‗await‗‗ on its own isn’t enough. I haven’t found async-for to be all that useful. So far, anyway.

[@kbprojekty]

From what I read, that short answer describing a difference between concurrency and parallelism is incorrect. Parallelism is described correctly, but concurrency is either two lines of customers ordering from a single cashier taking turns, or each line having its own cashier.

[@thorsten9808]

At 16:02 the HTTP response status is not printed, unlike before. What happened there?

[@mlenox]

Excellent!

[@muzafferckay2609]

Could you make a video for parallel programing as well?

[@razean22]

I don't think this really exists in python because of the GIL. I would switch to another language if you need true parallelism.

[@ygorpontelo]

@@razean22 GIL stops thread based parallelism, you can still get processes to execute code in parallel. That said, if you need high performance, probably best to go for another language.

[@razean22]

@@ygorpontelo thanks Ygor, I wasn't aware!