can run with granian by setting REFLEX_USE_GRANIAN var

[Lendemor]

WIP: allow use of Granian instead of the uvicorn/gunicorn for running backend.

[itsmeadarsh2008]

Thanks for this PR.
Edit: This shall improve the performance of API as it is written in rust.

[bertbarabas]

Is it worth trying this currently? I'd love to check if it helps the performance of the file upload since that problem seems to be in uvicorn/gunicorn.

[Lendemor]

Is it worth trying this currently? I'd love to check if it helps the performance of the file upload since that problem seems to be in uvicorn/gunicorn.

We would love to have more people trying it.
It is working, but testing in real condition rather than on basic app will let us know if anything need to be adjusted.

[bertbarabas]

How would I try this out? Is it in vv0.6.0a2 or do I need to do more or perhaps wait for it?

[Lendemor]

How would I try this out? Is it in vv0.6.0a2 or do I need to do more or perhaps wait for it?

If you want to try it right now, you can clone the project and use this branch to run reflex instead of the installed package.

It is not part of 0.6.0a2, and I'm not sure if it will be in 0.6.0, need to discuss it with the team.
At the latest I think it'll be in 0.6.1

[masenf]

How would I try this out?

pip install git+https://github.com/reflex-dev/reflex@lendemor/use_granian granian
REFLEX_USE_GRANIAN=1 reflex run

Is it in vv0.6.0a2 or do I need to do more or perhaps wait for it?

planned for v0.6.1 in ~2 weeks

@Lendemor I tried this out and it seems to work in dev mode, but the websocket doesn't connect in prod mode and /ping gives "Internal Server Error"

[Lendemor]

I'll look into it, prod mode was running fine before my last commit I think

[bertbarabas]

Here are some timings of uploading a file of size: 522,134,318 bytes in my actual application in dev with browser and backend on the same laptop.
(I killed reflex and relaunched between tests)

REFLEX_USE_GRANIAN=0 reflex run
Run 1 took: 127 seconds
Run 2 took: 183 seconds

REFLEX_USE_GRANIAN=1 reflex run
Run 1 took: 63 seconds
Run 2 took: 104 seconds

Looks like Granian is faster although my laptop may not be the best place to measure speed, maybe the laptop gets slower as it gets hotter…

[Lendemor]

I'll look into it, prod mode was running fine before my last commit I think

Turned out the backend never actually worked for prod mode, because I tested it with a stateless app.

It's working now.

[itsmeadarsh2008]

Here are some timings of uploading a file of size: 522,134,318 bytes in my actual application in dev with browser and backend on the same laptop. (I killed reflex and relaunched between tests)

REFLEX_USE_GRANIAN=0 reflex run Run 1 took: 127 seconds Run 2 took: 183 seconds

REFLEX_USE_GRANIAN=1 reflex run Run 1 took: 63 seconds Run 2 took: 104 seconds

Looks like Granian is faster although my laptop may not be the best place to measure speed, maybe the laptop gets slower as it gets hotter…

Try it in a codespace.