Hi Railway team,

Our production e-commerce site is DOWN on its custom domain and we are losing

orders. The deployment itself is healthy — this looks like an edge /

custom-domain routing problem on Railway's side. Please escalate.

Project: CUSTOMERS PROJECTS

Service domain (works): eco-kamen.up.railway.app

(this uniquely identifies the service; its display name uses non-Latin characters)

Custom domains (broken), Cyrillic IDN in punycode:

• xn--80aknalhc1jb.xn--p1ai (apex)
• www.xn--80aknalhc1jb.xn--p1ai (CNAME btmwdzjw.up.railway.app)

SYMPTOMS (verified from multiple networks: a Russian ISP, US-based

infrastructure, and EU monitoring nodes):

1. Requests to the SERVICE domain succeed 100% of the time, instantly, with

   the complete ~467 KB HTML response.

2. Requests to the CUSTOM domain hit the same edge IPs (69.46.46.x) but fail:

   • With DNS pointing directly at Railway (A 69.46.46.120 / CNAME

     btmwdzjw.up.railway.app): TLS connections with the custom-domain SNI get

     a TCP connection reset (browsers show ERR_CONNECTION_RESET). At the very

     same moment, the service domain loads fine from the same machine — so

     this is SNI/domain-binding specific, not the deployment or our network.

   • Behind Cloudflare proxy (orange cloud): the origin returns response

     headers + EXACTLY 20,508 bytes of the body, then the stream stalls

     forever (deterministic, reproduced many times). Captured stalled

     responses include:

     x-railway-request-id: 3AbLxDSITbuaQvVhs_GTAg (x-railway-edge: osl1)

     x-railway-request-id: R7E_Tl8vQheIMh3MrJsmnA (x-railway-edge: ams1)

     (2026-07-13, ~15:04-15:08 UTC)

3. Domain provisioning looks fine on the surface: the domains show as

   configured in the dashboard, the _railway-verify TXT record matches, and a

   valid Let's Encrypt certificate for the apex was issued on 2026-07-10.

WHAT WE ALREADY TRIED (no effect):

• Deleted and re-added BOTH custom domains in the dashboard.

• Redeployed the service.

• Toggled Cloudflare proxy on/off and disabled all Cloudflare features.

  Both paths keep failing: direct = connection reset, proxied = stall after

  the first ~20 KB.

THIS IS RECURRING:

• The same domain previously went down with "upstream error" served by the

  Railway edge while the service was healthy (stale domain-to-service target).

• Another service of ours (custom domain ms.re-ai.ru, project re-ai) silently

  LOST its custom-domain binding: it disappeared from the service's domain

  list while our DNS records and the _railway-verify TXT remained valid.

POSSIBLY RELATED: every deployment of this service triggers a "deployment

failed" email even though the new build actually goes live and serves traffic

on the service domain.

Please inspect the edge routing state for these custom domains, rebuild the

bindings, and share the root cause so this stops recurring. Happy to provide

additional traces.

Thank you — this is production-critical.