Serverless to resume takes 10 secs
talosdeus
HOBBYOP

22 days ago

Hi,

All my projects are taking a ridiculously long time to go from serverless (sleeping) state to resumed state. Any ideas as to why? I hope someone from Railway can look at it. It's not just 1 project, it's for all of my projects. It wasn't this bad before. Now it is taking 5-10 secs for my site to load. Even simple standalone service websites. Region deployed: Singapore. Request source: India.

$10 Bounty

6 Replies

I think your website latency = cold start delay+app's own startup time +India to Singapore network latency


dharmateja

I think your website latency = cold start delay+app's own startup time +India to Singapore network latency

talosdeus
HOBBYOP

22 days ago

No it can't be that much. It wasn't this bad before (also this latency is for all my projects, even the most basic ones). 2-3 secs is acceptable but anything more than that and the user experience degrades a lot. I also have a fly deployment in parallel which is also serverless (because I am not sure if I am using Railway or Fly.io as my main hosting before making my site prod and public) and I get a wakeup time of around 500ms-1sec which is really good. It's just that on Fly the infrastructure is not so well rounded as Railway and there's no vertical scaling, only horizontal. So I do want to stick to Railway but they need to improve their networking a lot.


Have you tried checking the load times while serverless is disabled?


0x5b62656e5d

Have you tried checking the load times while serverless is disabled?

talosdeus
HOBBYOP

21 days ago

Yes with serverless disabled it's quite good, like <300ms for most requests. Railway is still not as good as Fly for Singapore to India latency. I am using Neon's Singapore region so I need my servers to be in Singapore too. Otherwise they would be in India as that's my main customer base. But I ideally wouldn't want to have serverless disabled as my main service still consumes 300-400MB which if billed for a month continuous is like $3-4 and that's kind of ok because I get really good scaling, but still. Fly's wakeup is almost instant, and also Fly's horizontal scaling is different than Railway's, it doesn't route randomly to all active machines, only start routing based on number of incoming requests at scale.


I think it’s safe to assume that serverless is causing the extreme load time then.


0x5b62656e5d

I think it’s safe to assume that serverless is causing the extreme load time then.

talosdeus
HOBBYOP

21 days ago

Yes that's what I said to begin with


Loading...