9 hours ago
Hi Railway team,
Since around Nov 18, 2026 (JST morning), some HTTPS requests to our service (stylemind-proxy-production.up.railway.app) fail with "Recv failure: Connection reset by peer" (curl error 35) during the TLS handshake. Requests that succeed are served by edge "hnd1" (server: railway-hikari).
What we checked: The app itself is healthy (logs show no errors, and requests that get through return normally). Restarting the service did not change the failure rate. From the same network, other sites (including railway.com) show 0 failures.
Failure rate: about 20-30% on Nov 18, about 2-3 in 20 on Nov 19, and 2 in 20 on Nov 21. It has not gone away.
Could you check whether there is an issue with the hnd1 edge? Thank you.
1 Reply
9 hours ago
This thread has been opened as a bounty so the community can help solve it.
Status changed to Open Railway • about 9 hours ago
4 hours ago
I checked your hostname from a second network that also reached hnd1. On Nov 21 at 16:26:24–37 UTC, nine fresh curl HEAD requests to / completed TLS and returned HTTP 404. All reached 153.52.255.23. One successful request ID for Railway to correlate is kY28CpZ0QYO05q9xn6XIxQ (upstream railway/us-west2). That short sample leaves an intermittent reset entirely possible.
For the next failure, the most useful comparison is a failed attempt and an adjacent success from your original network. Record the UTC time, destination IP, IP family and curl error for both, plus X-Railway-Request-Id on the success. If possible, repeat once via a mobile hotspot at the same time. The successful response identifies its POP; a reset before HTTP has no response header to identify the failing POP. Railway support can use the paired times to check the edge logs.
AI-assisted reply; the measurements above are from live requests. https://docs.railway.com/networking/edge-networking