Python (FastAPI) vs Go (Golang) Performance Benchmark

[@AntonPutra]

🔴 To support my channel, I'd like to offer Mentorship/On-the-Job Support/Consulting ([e-mail]) 🍿 Benchmarks: kzbin.info/aero/PLiMWaCMwGJXmcDLvMQeORJ-j_jayKaLVn&si=p-UOaVM_6_SFx52H 👋 AWS is expensive - Infra Support Fund: buymeacoffee.com/antonputra

[@my_name_is_ahad]

FastAPI vs ExpressJs please

[@LW12386]

Hey, what software do you use to run those benchmarks?

[@MDFireX5]

Go vs Deno2 vs Php

[@vamshiaruru5494]

this is like coughing baby vs atomic bomb

[@AntonPutra]

haha

[@Yusuf-ok5rk]

@@_vk03 not really

[@DavidSmith-ef4eh]

Spoiler: Go wins.

[@AntonPutra]

😂

[@christofflinde]

Spoiler: this does not come as a surprise. :)

[@svetlinzarev3453]

spoiler: rust wins :)

[@mrlectus]

I didn't even have to watch the video to know that 😂

[@laas29]

Zig rules over all

[@AntonPutra]

Updated my checklist: 'Use non-blocking async libraries when possible.' 😊

[@minma02262]

Where can I find you checklist?

[@AntonPutra]

Thank you all who helped to improve the Python application. I'll release round 2 in a few days.

[@pedro_alonso]

Can we have a spring boot performance benchmark? I'm curious how it will perform in comparison with other tools/languages and with virtual threads and without it

[@AntonPutra]

yes sure

[@pable2]

@@AntonPutra Spring boot vs symfony + frankenPHP/swoole?

[@mogaclub]

spring boot vs quarkus

[@AlokSingh-rj1gt]

One of the fastest tortoise vs cheetah.

[@AntonPutra]

😊

[@fluxcaruso345]

So the conclusion is use go for any backend service if you dont have any spesific requirements that force you to use another language

[@boburnomonov-d6s]

hell yeah

[@DavidSmith-ef4eh]

nah, dotnet. I rather sacrifice 20% of performance for all the nice things you get in dotnet, like pattern matching, extension methods, linq and of course the most advanced ORM in the world.

[@rodjenihm]

@@DavidSmith-ef4eh Is it even sacrifice in performance? Go might be even slower than C# and even Java nowadays.

[@tpadilha84]

I suggest you to use whatever technology makes you the happiest and most productive. These benchmarks are rarely representative of real world, most apps will never reach the level of concurrency tested here.

[@DavidSmith-ef4eh]

@@rodjenihm nah, he did a benchmark, dotnet vs go. although it's dotnet 8, and go won. I believe go would even beat dotnet 9, but not by much.

[@EddaFredericks]

enterprise-ai AI fixes this. Python vs Go benchmark comparison.

[@ridhotamma6331]

FINALLY!, what I've waiting for so long, thank you Anton!

[@AntonPutra]

my pleasure, i hope someone can improve it and i'll rerun it

[@SAsquirtle]

I understand not everyone knows every language/framework well enough to make a decent app but cmon you could've done some more research. 1. Why are you using a synchronous pg driver with an async endpoint? this blocks the event loop till each query finishes. If you absolutely want to use the sync driver, use plain def routes without async. 2. Why aren't you using asyncpg with async endpoints? the whole point of FastAPI is to make completely async REST APIs which use async packages. If you cannot find an async package or method, use run_in_threadpool with the sync method.

[@muray82]

This is why he ask for PR in repo.

[@SAsquirtle]

@muray82 then what's the point of the video? PRs are fine for minor improvements or enhancements but these are major issues which should've been addressed before the video was made

[@SAsquirtle]

Also want to clarify that even with the changes I mentioned, python and fastapi will still lose terribly but no where close to this extent

[@omlanke3466]

I was looking for this comment. He is not using an ORM in go, but he uses SQLModel in python. ORMs significantly slow down apps too. This comparison should be retested using FastAPI with plain asyncpg.

[@tuanvumaihuynh]

Using asyncpg with proper configuration can indeed improve FastAPI's performance significantly. There benchmark in repo show that it outperforms pgx.

[@pi3ni0]

Based on this and other videos, I didn't realize how efficient Go was! Thanks for sharing

[@egida6486]

its not Go that is efficient here ...

[@pi3ni0]

@@egida6486 What is then?

[@egida6486]

@@pi3ni0 its just python that is very inefficient.

[@mickhuby8786]

Just what I’ve been waiting for! Thanks for doing these benchmarks.

[@ZuvielDrama]

The Switch from C# to Golang was a right one. Golang has the best simplicty / speed ratio. Using rust or c++ is more performant than golang. The costs are complexity in the codebase. Not to mention golang gives you one binary to deploy. This compared with c# dll hell...

[@DavidSmith-ef4eh]

which orm language do you use for go. Most of them require you to write SQL or generate helper code based on exisiting sql files.

[@allanamaral2587]

And that’s a good thing! After spending hours profiling and debugging slow LINQ queries you start appreciating writing just SQL

[@Bob-qz5yj]

C# has excellent aot support. You can often get one binary just like go if you prefer

[@DavidSmith-ef4eh]

@@allanamaral2587 just log the query lol, then fix your linq code

[@123mrfarid]

​@@DavidSmith-ef4ehactually i prefer writing raw sql. Much more flexibility and i can do much more complex data selection or manipulation which is quite difficult to achieve with orms. I use orm of course but only for simple queries.

[@IncomingLegend]

try to compare Go with Gleam

[@AntonPutra]

ok

[@infantfrontender6131]

It will be the same result as with an Elixir because Gleam uses the BEAM and OTP

[@Gruak7]

@@infantfrontender6131 Could be but Gleam has its own ecosystem and a different web framework so there might be some differences.

[@KayandraJT]

@@infantfrontender6131 gleam is currently slower than elixir. the creator of gleam is focusing on improving performance so go slaps gleam atm

[@andreyvershkov4072]

Top video! Thanks. Please make OpenSwoole/Swoole (PHP) comparison vs Go.

[@AntonPutra]

thank you!

[@obamna36]

Go is always a good pick for api server. good dev ux and decent performance. if you want top tier performance but with a longer time to write pick rust.

[@AntonPutra]

i love rust

[@roccociccone597]

I have tried rust for web development. Unless you need the utmost performance Go is going to get you 95% of the way.

[@KayandraJT]

tbh i used to be one of those people who thought like this... then earlier this year, i decided to give rust the love it needs and learned it... my productivity in go and rust is around the same also because i still use both of them. rust for me does not take longer time to write, i'm about as productive with both.

[@jorgetovar621]

It would be great to have a long video setting up all of this and learning how to perform these benchmarks. That way, we could better understand the process and contribute more effectively to this code. Really good job!

[@SnazzieTV]

Python and speed can never go in the same paragraph.

[@DavidSmith-ef4eh]

maybe you could get python like syntax in the fastest language that even beats rust. But don't follow that project right now. It's called mojo, if you want to look it up.

[@AntonPutra]

haha

[@dmitrii_cl]

SlowAPI

[@thegrumpydeveloper]

When they pass things down things work well. Python is hands down the ai winner and pandas and other data science libs are pretty cool. Wish academics had chosen go though because I wonder how much faster things could have been.

[@subbotin.monster]

Could you compare Django and FastAPI?

[@AntonPutra]

yes and flask

[@Imam.Mehdi_]

Also I was curious how much a certain hardware can handle load,,, can you benchmark different VPs or ec2

[@bartoszstasiak237]

As a Django developer I would love to see this

[@sauravagun9751]

no wonder why i have to learn go next year, gonna be my 3rd language yayyyy! multilingual in programming it is 😎😎

[@inithinx]

Please consider doing kotlin(ktor/springboot/native) vs golang(fiber/echo/stdlib)

[@AntonPutra]

added

[@AntonPutra]

► PR by Uday to improve FastAPI - github.com/antonputra/tutorials/pull/375 ► PR by Evgenii to improve FastAPI - github.com/antonputra/tutorials/pull/376 ► PR by Viky to improve FastAPI - github.com/antonputra/tutorials/pull/377

[@neoko7220]

Waiting for php

[@freedomseekerism]

lmao

[@DavidSmith-ef4eh]

@@freedomseekerism you might be surprised that PHP (the swoole version) could be just 10% slower than go.

[@DavidSmith-ef4eh]

of course, laravel will be on level of python lol

[@Oleg77777100]

Waiting for cobol

[@AntonPutra]

i'll do swoole soon

[@theuser384]

The real competitor for Go is Crystal, both compiled and tiny footprint.

[@AntonPutra]

ok i'll take a look

[@alemao182]

awesome content man! would be nice a python vs python comparison as well, something like fastapi vs django-ninja vs litestar

[@franklynd]

Does this improve if you use the pypi JIT with FastAPI ?

[@skippednote]

I'm assuming you are computing most of the metrics using nginx ingress. If so can you please share your configurations and the dashboard queries?

[@ramonpereira4460]

I see that your are using C instances! Thanks a lot, Anton. Maybe try C instances vs M instances! Also, at my previous job, we had 1000 instances of FastAPI for ML inference and 1 Golang instance distributing to all of them. Golang is very fast.

[@BarakaAndrew]

I honestly don't understand people who use python for anything other than ML even that is arguable. Its architecture is so bad

[@AntonPutra]

I was expecting much better results

[@dpdx_05h]

Спасибо, интересные видео!

[@amirhmahmoodi4607]

You did not even use none blocking db driver for python, how can i trust you after this?

[@AntonPutra]

Collecting feedback and rerunning a test in a few days

[@stefanhoromnea6064]

Hi, nice comparison! What did you use to make the graphs?

[@denysmelnyk3969]

grafana

[@VladimirS-h9o]

Why in on 7:28 Connection Pool Size of Go increased and after that more than Python?

[@appsky7982]

I see you have received lots of PRs for fastapi. I can't wait for the second benchmark

[@EricSouzarys]

Hey Anton! Any chances we can get a new comparison with the java quarkus reactive code?

[@ferdynandkiepski5026]

Now this is actually kind of surprising to me. Python libraries all call into C/C++/Rust. I'd really expect it to be better. I guess the overhead from using such a slow language to call into library functions can't be hidden.

[@live4wap]

Not that fast

[@DavidSmith-ef4eh]

well, it was a fast video, since fast-api couldn't keep up half way through. Otherwise it might have lasted 20 minutes.

[@AntonPutra]

haha

[@gibzrival1565]

Go has become the standard language for benchmark comparisons

[@AntonPutra]

► PR by Uday to improve FastAPI - github.com/antonputra/tutorials/pull/384 ► PR by Paweł to improve FastAPI - github.com/antonputra/tutorials/pull/386 ► PR by Uday to improve FastAPI - github.com/antonputra/tutorials/pull/387

[@trsdesrn5246]

Just a newbie question: if Python is so slow compared to Go in almost all the frameworks, what is the advantage to adopt Python over Go for such applications like backend development?

[@DavidSmith-ef4eh]

well, go lacks language features and is very verbose. which means you have to type a lot, but the code is easy to understand. personally not a fan of typing a lot. But even then, I'd still use go over python on gunpoint.

[@MrBeltalowda]

pyhon is much cheaper and faster to iterate.

[@OlegKorsak]

it runs the same as Django because of sync nature of your app. IO is getting bottlenecked. waiting for a true async version

[@StingSting844]

This was brutal. Yes, the python version can be really fast too but the main point is that the speed in Go is something you get by default. If you're building for release speed then use python. If you are building for performance then use Go

[@BlackistedGod]

Its better to do this comparison with another interpreted language like PHP (Laravel, Symfony) or JS frameworks (Express, Next). Python is the slowest language and we would already know the result if its being compared to any compiled languages.

[@AntonPutra]

true, but fastapi uses a lot of new features, including coroutines and type hints, and on the official website, they compare fastapi with go and nodejs

[@BlackistedGod]

@@AntonPutra then their official website could be full of BS they even dare mention G hahaha I doubt that SlowAPI could even beat Swoole/RoadRunner or NodeJS. I hope you can run benchmark on either one of them or both

[@saikirannadipilli7831]

Hi @AntonPutra can you please do benchmark for go vs c++(Poco library)

[@DhavalAhir10]

Anton, can you please tell me more about the company "Juniper Networks"? I am Python Dev and have plan to apply there. I heard the company is quite good with decent work culture.

[@AntonPutra]

I was part of the MistSys startup when Juniper acquired us. Right now, those buildings in Sunnyvale look like ghost towns after covid

[@DhavalAhir10]

@AntonPutra ohhh that's the case. Btw I am from India. They offices in India.

[@kostan3v]

Would we see Dart (with shelf) any time soon?

[@hooyah]

OOT# golang chi is only a package for routers. that means chi perfomance is similar to std lib?

[@qwerty6713]

That's interesting that difference was only 7-10 times. But if we add business logic, than difference will be more significant

[@AntonPutra]

sounds about right

[@roccociccone597]

oh it is. As soon as you have any form of looping in your code python shits itself.

[@j-p-d-e-v]

@@roccociccone597 agree, I parse large data if I have time I use rust to do it and its x10 to x20 faster than python. I just use python if the team needs an immediate output.

[@SergeyBevzuk]

Swoole/OpenSwoole vs Go/Rust/C#

[@DavidSmith-ef4eh]

swoole 6 is about to be released, and we didn't even get swoole 5. At this point, I'd rather wait for him to do swoole 6.

[@AntonPutra]

yes, i'll do swoole soon

[@GiacomoMiolo]

Axum vs FastAPI?

[@hughesabsalom526]

Come on bro that will be a bl**d bath😂😂

[@AntonPutra]

well maybe

[@GiacomoMiolo]

@@hughesabsalom526 just giving Anton a challenge to show in the same graph drastically different scales eheh

[@AndrewTikhonov]

Why are you refreshing model after insert? it is useless and only make additional requests. I am also not sure sqlmodel is a good thing to use. Try mapping sqlalchemy to dataclasses directly. Can you please measure how fast will be golang vs python implementing facet search over 10million table with (unique fitlter combinations on each request, of cause). Who will be faster?

[@omlanke3466]

Why not use asyncpg directly? ORMs slow down the app

[@AntonPutra]

i'm refreshing to return it back to the client, i do the same thing for go

[@MusKel]

When Kafka vs Nats?

[@BORNINSPACE]

Try Robyn, Rust based python framework

[@AntonPutra]

Ok, I'll take a look, but it has fewer than 5k stars

[@BORNINSPACE]

@@AntonPutra its new yes, but from my tests when you need speed in python it's Sanic and Robyn, they are like 2-4 times faster then FastAPI. FastApi is nice becouse it has good support for async, does OpenAPI docs automaticaly and has build in type checking with Pydantic. But becouse of pydantic it does aditional checks for each request, if you remove Pydantic you will see an easy 2x r/s improvement) Its fast for development speed though)

[@PTsanti5010]

Can you please show how to create the clients ?

[@AntonPutra]

Sure, well, I have a bunch of them in my repo

[@PTsanti5010]

@@AntonPutracan you share the code of the client ?

[@PouriyaJamshidi]

Any progress with Nim, Anton?

[@AntonPutra]

not yet

[@bebus7017]

Which ASGI server is FastAPI running on in this comparison? It would also be helpful to compare different ASGI servers (e.g., Nginx Unit, Uvicorn, Daphne, etc.) using the same Python app for a broader perspective. Additionally, consider using a lighter framework like Starlette (the foundation of FastAPI) for such benchmarks, as it avoids the overhead introduced by FastAPI’s data validation features

[@bebus7017]

After checking the source code, I noticed that the comparison runs FastAPI using fastapi run, which defaults to Uvicorn as the ASGI server. However, my point about comparing different ASGI servers remains relevant. A broader analysis with different servers could provide deeper insights into performance variations

[@garveyebsmith]

Thanks Anton!

[@carlos_rijo]

Looking at these benchmarks, I guess "Go" is top 1 right?

[@poet9638]

when I use python to code llm, I use fastapi. This is because i want to use codes created by a large number of scholars. I like Python so much~

[@pekhov21]

Hey, I have an idea for the next video. What if compare interpreted languages. PHP, JS(node), RUBY, PYTHON And run PHP with Roadrunner. What do you think about this?

[@WebNinja-r2b]

Python Emmett vs Rust Actix you will LOVE this

[@toragodzen]

When Python surpasses Go, it will truly become the best programming language ever! )

[@roccociccone597]

so never then? Python can't even match JS and JS sucks balls for performance.

[@AntonPutra]

i'll be waiting

[@AB-ms7my]

Hello again! I guess that CPython is your primary topic. Python is a language. Python has more than one implementation.

[@saadhabashneh5587]

FastAPI vs expressjs

[@AntonPutra]

got it

[@Yue2u]

In first test you create data every time in endpoint, python isn't fast working with memory operations like this, it makes app a lot slower than it could be. In second test using synchronous sqlalchemy inside async function slows fastapi a lot, it works like django in this case, executing code sequentialy. So, python is slower than go, thats obvious. However the benchmark isn't really fair.

[@MrLOPIU22]

They should change the name for slowapi

[@AntonPutra]

😊

[@oxXPMXxo]

Now I wanna see Java 21 virtual threads if it's legit

[@AntonPutra]

will do, probably 22

[@pickle1987]

Comparing python to go is like comparing some random generic Coke with CocaCola

[@carlos-rian]

Amazing ❤

[@AntonPutra]

thank you!

[@rida_brahim]

at this point it's pretty obvious that python backend isn't the go to tbh, I don't care if you got 10 users or 100m users you just avoid python. the amount of time and energy you would spend on python can be spent on elevating your skills on a better language ( talking web development only).

[@huuphong3657]

could you compare NestJs (express) vs Go (Gin) one day plz

[@iddiiddrisu5971]

Golang is definitely winning here!!! 😂 FastAPI vs Spring might be more fair. I can't think of any golang lib that would be as slow a FastAPI in terms of latency. Let me watch now and see.

[@kamikaz1k]

DSQL vs RDS?

[@gayalmn]

For when an elixir comparison ?

[@Mr.BinarySniper]

0:10 people can easily understand which is golang and which is fastapi.

[@AntonPutra]

yeah i know

[@RayDenis]

Yeah all you have to do is look at the chart

[@pi3ni0]

Waiting for round 2 with PRs for FastAPI

[@AntonPutra]

will do soon

[@pi3ni0]

@@AntonPutra Yeah, no pressure, really appreciate you doing the benchmarks

[@alexbrik1802]

Love your work, mate! I personally see Go as the perfect choice for balance between performance, development speed, and maintainability Glad to see your tests confirm this somehow 😁

[@RealEngineer]

Soooo what we need now is C# vs Java 😃

[@АрианШапиро]

Антон, позязя питон сравни с нодой и bun

[@AntonPutra]

sure, i'll wait for someone to improve fastapi a little and then i'll compare it with node/deno/bun

[@saiphaneeshk.h.5482]

7:15 3500 or 35000?

[@AntonPutra]

i misspoke, 3,500 rps

[@kvs7720]

can you plan to do one DevOps end-to-end project frontend and backend your all videos is clear content @TQ

[@leularia]

considering instagram use python, Javascript for Backend is not that much bad idea !

[@ArcticPrimal]

Please test Golang Hertz and Bun router

[@akinshipeolaoluwa761]

This did not test fastAPI at all. You need to use non blocking driver for the database and gunicorn server so as to be able to use more process. Please redo this video as it very important. Thanks

[@brunocascio]

comparing compiled vs interpreted does not make sense. Before the test is completed we would know who is going to win xD

[@techinsider3611]

Please do also test litestar vs go lang vs fast api.

[@leandroapariciodoprado7072]

Compare AWS and Azure and give us a request per dollar report.

[@AntonPutra]

i was thinking about this as well, using thanos to compare different cloud providers in a single dashboard

[@cinos3911]

C++ vs Rust, please 🙏

[@jholansabaria6001]

try Go vs PHP (using Open Swoole)

[@AntonPutra]

yes soon

[@DavidSmith-ef4eh]

@@AntonPutra I'd use swoole, swoole-src github. Open swoole seems to be lagging behind. Ideally swoole version 6 which will be soon released.

[@shreyaspatange8653]

do django next please

[@AntonPutra]

yes i'll do django as well

[@snippletrap]

Try Litestar, a FastAPI competitor that claims to be 2x as fast

[@ranggatohjaya5471]

RabbitMQ Stream vs Kafka vs Redis Stream vs Pulsar

[@vivekchaudhary5728]

SpringBoot (Highest Optimisation) vs GO please

[@moochalshrek6519]

FastAPI and Node.Js vs Bun!

[@ysomad]

NATS vs KAFKA / RabbitmQ please

[@iddqd2048]

Those test on AWS are nonsense. The same type instance performance can be different by 40%!!!!

[@j-p-d-e-v]

Python hogs a lof of memory and cpu thats why Im slowly rewriting my web app to Rust.

[@mohagungnursalim8219]

Php swoole/frankenphp vs go

[@WaneyMamahit]

Please use Pypy for FastAPI

[@AntonPutra]

ok but they don't use it in the "prod" example - github.com/fastapi/full-stack-fastapi-template/blob/master/backend/Dockerfile#L1