019d11e1-619b-49aa-9f60-379506b03485

hobby users now deploy exclusively to metal, we have found any slowdowns can be attributed to using the public network to connect to the database or inefficient code paths.

Theres only a flask API in my project. my DB is a supabase instance

my app just has one endpoint which generates an embedding and calls a supabase RPC function. was ~6-8s response times before which im fine with but thats jumped to me having to increase my RAM/CPU limits as well as 25s response times now

what region is your supabase database in

West US (north california)

what region is your app deployed into?

then you have them both in the same zone

yeah so if anything metal should be faster than orgeon

yep it has newer CPUs

yeah so im not understanding how my RAM/CPU usage almost increase by 2x and my response times 3x

i was running 4gb ram and 4vcpu before and it was all fine but after metal i get out of memory issues. after increasing both to 8/8 my vcpu usage is 7 during the request

perhaps the new CPUs don't perform well for your specific task?

yeah it just sucks because now ill most likely have to switch over to a new cloud provider. huge fan of railway but i figured we wouldve had some email warning or something instead of just waking up monday to find out my app was broken all weekend and also that there is nothing i can do about it except pay more money to go back to the old server

paying more money wouldn't help long term, pro users will eventually be moved to metal too, the end goal being for Railway to leave GCP.

sorry that this didn't work out for you!

some interesting customer support. theres numerous people complaining so clearly theres some underlying issues. all this without warning too. lol

I'm really not sure what we can do about this unfortunately, we have actively been working on moving off of GCP

i am having the same issue, my postgres db on railway can only be set in oregon, but my app deployment ar restricted to metal which are in the west causing huge slowdowns and connection timeouts

I'm also seeing a noticeable slowdown (increase of +- 100ms per request on the private network)

If the aim is to move entirely off GCP, is there a timeline of when services with volumes will be moved across as well?

The slowdowns are generally caused by using the public network to communicate with the database.

I am using the private network and everything was perfect before switching to metal

^^

!s