Could u reply with N/A so the thread doesn't get closed?

Also, take a look at these docs:
https://docs.railway.com/reference/scaling

as far as I know there's no "auto instance creation" (auto horizontal scaling)

But I think I've seen some users implement some strategy to programatically deploy replicas using the graphql endpoint

cc @Brody (user doesn't have the pro plan cargo)

@DEOGEE

N/A

My bad , i was away for a while

Should i open a new thread?

nope this thread is good

I'll look into graphql thanks. Is this the industry standard of solving this issue or am i approching it wrong?

I apologize for the basic questions.

One more thing , If i have 2 users using one railway replica VS 2 users using 2 seprate replicas , how does the cost compare?

I'm not a well versed devops guy, but I know you can setup auto scaling on services like GCP, AWS and Kubernetes but I don't think i've saw something like this on other PaaS like Fly, Render or Railway

That depends entirely on the resource usage of your service (since that's what railway charges you)

you can have a 3x Go server replicas that run in a idle of 20mb each and don't increase that much on higher loads or 1x ExpressJS server that has an idle of 120mb and goes to ~200mb on high load, you'll get billed for the resource they use

idk if that made sense

Thank you , i'll look into these

but depends on what you're running

It's an ffmpeg server that converts images to video for my startup. It's pretty gpu intensive. I guess i'll try both and compare the results.

Thank you so much @Medim

Are you using a serverless gpu provider?

Railway doesn't have gpus for now™️

maybe with metal..

Yes i am , runpod.

nice

Railway doesn't support auto horizontal scaling based on demand/usage (I'd really like this though)

It does support manual horizontal scaling

If you desperately wanted auto-scaling, you'd likely have to create an app that consumes signals (requests/usage) & shoots out scale requests to the railway API. But (I think, unsure) that would cause all instances to restart

all replicas will stop / start when downscaling / upscaling.

best just to set a fixed amount of replicas, enough to handle your expected peak.

!s