11 hours ago
Hi,
Our services run in europe-west4-drams3a (Amsterdam), but every request from Armenia and Georgia enters Railway's network in São Paulo. The response header shows x-railway-edge: gru1. This adds about 200–250 ms to every connection. Our users are in Armenia, so the whole app feels slow.
- Services: project mivo, services backend and web
- Domains:
- api.mivo.am, which is a CNAME to 481o9wdk.up.railway.app (69.46.46.93)
- order.mivo.am
- Example request ID: gjTh5xdVQC-Ovbp-YqdHTg
We measured with Globalping probes on 11 September 2026, using HTTPS to /health:
┌─────────────────┬──────────────┬─────────────────┐
│ Probe │ Network │ Time to connect │
├─────────────────┼──────────────┼─────────────────┤
│ Yerevan, AM │ Ucom │ 243 ms │
├─────────────────┼──────────────┼─────────────────┤
│ Yerevan, AM │ Viva Armenia │ 199 ms │
├─────────────────┼──────────────┼─────────────────┤
│ Yerevan, AM │ G-Core Labs │ 241 ms │
├─────────────────┼──────────────┼─────────────────┤
│ Armavir, AM │ House Net │ 199 ms │
├─────────────────┼──────────────┼─────────────────┤
│ Tbilisi, GE │ Cloud 9 │ 245 ms │
├─────────────────┼──────────────┼─────────────────┤
│ Tbilisi, GE │ TradeZone │ 254 ms │
├─────────────────┼──────────────┼─────────────────┤
│ Falkenstein, DE │ Hetzner │ 13 ms │
└─────────────────┴──────────────┴─────────────────┘
For comparison, the same Armenian probes connect to other providers' European servers in 1–60 ms. Examples:
- Frankfurt apps on Render: 1–32 ms
- Vercel: 30–67 ms
- a Hetzner server in Germany: about 60 ms
So the network path from Armenia to Europe is fine. The problem is that Railway's anycast routing sends this traffic to gru1.
Could you make traffic from Armenia and Georgia (and the region generally) enter at a European edge? If that isn't possible, is there a way to pin our domains to a European edge?
Thanks!
2 Replies
Status changed to Awaiting User Response Railway • about 11 hours ago
2 hours ago
Hello,
We've withdrawn from a global transit provider (Lumen) in Sao Paulo which has now fixed this issue. Thank you for the report!