How FastAPI Handles Requests Behind the Scenes

[@moneeshkumar1838]

Great content brother Quick Modification: sync router is Concurrency not Parallelism. In python parallelism is achieved only by multiprocessing

[@vladhaidukkk-learning]

Great video, but I think it’s important to mention that multi-threading in Python is not parallel

[@bakasenpaidesu]

Multi processing*

[@vladhaidukkk-learning]

​@@bakasenpaidesu Actually you are wrong; multiprocessing is parallel because Python spawns an entirely new process with an entirely new interpreter.

[@benshapiro9731]

Multi threading in python is technically also parallel programming whenever a thread releases the GIL, such as during time.sleep or open calls. In those specific instances, there can be two (or more) threads truly executing in parallel in the same python process, because one thread is waiting for the results of a system call, during which time it releases the GIL since reference counts don’t need to be updated, allowing another thread to acquire the GIL and execute a piece of code in parallel.

[@benshapiro9731]

On this topic also: check out the beta of. Python 3.13! There is a flag that can be passed when launching python that removes the GIL, allowing truly parallelized execution with just threads. Been playing around with asyncio and concurrent.futures.ThreadPoolExecutor -> noticeable speed up. Shame that c-extension based libraries like numpy are unusable with this setting

[@vladhaidukkk-learning]

​@@benshapiro9731 This is called concurrency, not parallelism. Parallelism is when two or more tasks can run simultaneously, using the CPU, without waiting for I/O-bound operations. While the GIL doesn't prevent Python from switching context between threads waiting for I/O-bound operations, this is still considered concurrency, not parallelism.

[@AnaskiBaithak]

Thank you for this, I always wondered the difference between async def and def

[@usama57926]

best video on fastapi, straight to the point ❤

[@PriyankRupareliya]

This is very valuable content. New job, starting to work on FastAPI. The whole python paradigm is new for me. This video helped clear one of the most critical concept for me. Thanks brother.

[@hardikvegad3508]

What an excellent explanation. Thank you!

[@saeed-khan-cs]

Precisely on point video! I am new to fastapi & I was looking for exactly this information. Kudos!

[@clearthinking5441]

Can all blocking operations be made non-blocking with asyncio? You advice is to use def for blocking operations, I am wondering wether it would be better advice to ensure that all operations are non-blocking and use async def always ...?

[@codecollider]

You're absolutely right that using only non-blocking operations is ideal, and relying on async def wherever possible is great advice. However, there are cases where blocking operations can't be avoided like when an async version of a library you need doesn't exist, and you're stuck with a synchronous one. It's all about adapting to the tools available!

[@sarvesh6785]

Amazing Video brother. You solved my issue.

[@joeygibli]

It would be more clear if you wrote “Hello 1” and “Hello 2”; “Bye 1” and “Bye 2” for the different functions to see which routine the messages are coming from

[@정하은-h6s]

OMG! This is so helpful and a great video. Thank you and please post more videos like this!

[@vikranttyagiRN]

Nice explanation. Concise and to the point.

[@sany2k8]

Great explanation, you should create more videos bro...

[@valentinmilanesio02]

Great video! I have a question, in the first case, when the second request arrives inmediately, where would be the queue and who is handling that queue?

[@codecollider]

The queue is managed by the ASGI server (like Uvicorn). When the second request arrives, it’s held in the server's queue until the event loop is free to process it.

[@valentinmilanesio02]

@@codecollider thanks!!

[@ishaquenizamani9800]

Thanks for clearing this concept.

[@aguspray]

What if my endpoint have database process (post data) and blocking i/o for timeout process? what should i choose? Async def or normal def?

[@usama57926]

@@aguspray use async querying methods for db that way you can use async endpoint.

[@aguspray]

@ inside of code, there is also blocking process (while loop), then how to make it asynchronously?

[@PyPeak]

Beautifully explained!

[@sticksen]

My question would be how FastAPI then manages workload when it´s handed over to the worker thread. Because I can only see one worker thread running, at the same time it handles 40 'workloads' concurrently.

[@Shane1994322]

Clearly explained!! Thank you

[@hyakuheli]

what happens if the server is created with multiple processes (for multiple cores) with uvicorn main:app --workers 4. The first 4 clients are given a new process? or it can be either a process or another thread if a thread is in a non-blocking task? and what happens when there are more than 4 clients at the same time in this case? the new clients are assigned to threads randomly across the 4 processes?

[@lwangacaleb2729]

I need some help, I want to create a fast api endpoint that calls a synchronous function that has a lot of blocking I/0 operations. But I want the endpoint function to run asynchronously so it can accept many requests at the same time. How should I do this, is there an alternative approach?

[@Praise-rs4mc]

The only way to achieve that is to use multi-threading which I advice against.... instead, make the function asynchronous and try to find the non-blocking function for what you want to do...

[@Praise-rs4mc]

Better still, use the run_in_threadpool function from fastapi to run the process in a different thread so that you don't block the event...better than implementing multi threading on your own.

[@lwangacaleb2729]

@@Praise-rs4mc thanks alot, I will give it a try.

[@fhuadbalogun2997]

Thank you very much for this.

[@rainbowrunner1136]

Hello, great video. I am super new to python. And started using very recently for backend where its being used along with FastAPI. Can you please tell me what are some concepts/ topics which I should be aware for developing efficient code. For eg from this video, I learnt that which function runs concurrently and which doesn't. I had no idea tbh about this before. If you could spare few minutes and share some stuff which I should know it would be great.

[@thanhlongle6276]

Thanks man for the video. I am trying to use fast api for db CRUD, which one do you think i should use for get post put and delete?

[@codecollider]

It depends on whether your database library supports non-blocking queries. Ideally, for endpoints involving database calls, use async def if your library allows awaiting query execution (like await db.execute()). If you're using SQLAlchemy, it provides both blocking and non-blocking methods for queries. It's generally recommended to use the non-blocking approach for better performance.

[@thanhlongle6276]

@@codecollider thank you, everything i write is in normal, non async, and I am using sqlite3 package. I think i will use normal def for all of it, since they are all parallel, and the blocking of read and write on database is performed by SQLite itself

[@JainamParekh-pk6sz]

Hey I actually tried the code and found that all works sequentially for me. I executed the same code which you wrote. could please give me reason ? Please explain.

[@codecollider]

I guess you are testing on a single browser, if you look closely I had used two different browsers to make multiple requests. Try two browsers, hopefully that should resolve the issue.

[@JainamParekh-pk6sz]

@@codecollider Okay thanks that helped.

[@adityahpatel]

2 and 3 look the same. hello of each is printed first then bye from each. how is 2 and 3 different? its not explained crystal clearly

[@ArulMuruganAS]

its working as explained in unix but for some reason its a different behavior in windows

[@arjunc5896]

def endpoint3() is not running parallely for me as supposed to what u said in the video. Instead it is sunning one at a time. Do u know why?

[@codecollider]

I believe you are testing APIs in the browser. Sometimes, browsers like Chrome have limitations on making parallel requests to the same URL. In the video, if you look closely, I am using two different browsers to hit the same API in parallel. You can try the same approach.

[@arjunc5896]

@@codecollider Yes you are right. I tried from different browsers and it worked. Strange though. Thanks

[@myselfriz]

very well explanation.

[@rajeshwariraghuwanshi3852]

gem video ✨

[@nishantthakre8174]

The content is great, but Python's multithreading can't achieve true parallelism because of GIL which is introduced for thread safety. However, there are ways to work around this, such as using multiprocessing or alternative compilers like Jython or IronPython, instead of the default CPython. Also the multithreaded system demonstrated in your video is only achieving perceived parallelism and not actual parallel execution.

[@ChrisHalden007]

Great video. Thanks

[@startup_cult]

this is such an underrated video

[@henilshah.]

coooooooooooooooooooooooooooooooooolllllllllll 🤟thanks man

[@gck2p3education90]

Nice

[@adrianmisak07]

great video

[@lucaspraciano4640]

❤

[@adityahpatel]

It is extremeley easy to demonstrate thius using sleep. Every tutorial on the web uses sleep. the probolem is wvery funciton is not awaitable, hence this example doesnt apply.. in the real worldIt i

[@robertavetisyan8282]

boooozi txeq

[@Rejuyan-Ahmed]

Thanks man

[@udaym4204]

can you make fastapi how run under the hood and how @app.exception_handler work Thanks awesome contentent

[@ashwinabrahamjacob525]

great video