2 months ago
I’m running a production backend on Railway deployed in asia-southeast1 (Singapore).
Most of my users are in India, and I’m consistently seeing a major issue:
First request takes 6–7 seconds
After that, all subsequent requests are fast (100–200 ms).
I confirmed the issue:
My backend is not slow
I have logs for any request >1s --> no slow logs at all
So the delay happens before the request reaches my server
It looks like cold-start–like behavior or network/edge warm-up issues
Happens randomly but very frequently for new connections from India
All users impacted are from India → Railway Singapore route
This is hurting user experience badly because:
First API call from India sometimes freezes for 6–7 seconds
Subsequent requests are instant (so server processing is fine)
It feels like Cloudflare → Railway routing or region warmup latency
Request
Can someone from the team look into this routing / cold-edge behavior?
Also, is there:
Any known issue for India → Singapore latency spikes?
Any network warmup or region handshake on first request?
Any recommended config to avoid slow first request latency?
If possible, please consider adding India region (asia-south) in the future
since the majority of Railway users are from India and this would eliminate the high initial latency.
Thank you!
4 Replies
2 months ago
Make sure serverless is disabled in your service settings.
0x5b62656e5d
Make sure serverless is disabled in your service settings.
2 months ago
It's already disabled.
2 months ago
I have this same issue, but in USA east, its like after 30 minutes inactive, requests are extremely slow and time out
Pro plan, US East
6-11 second first requests from external IPs
81ms from your own machine
Pattern: first request slow, subsequent fast
Container is warm, not traditional cold start
2 months ago
If you're using *.up.railway.app default domain, ISPs sometimes deprioritize or flag traffic to shared generic domains like railway.app, herokuapp.com, or onrender.com . You need to use a custom domain to get this resolved in this case.
This issue happend to a user before, and he said they got it resolved by using a custom domain. Check the issue here.