Summary

We're seeing intermittent request failures where our backend receives the request and returns 200, but the client never receives the response and reports a connection failure

(net::ERR_FAILED / "Network Error"). It appears to be the Railway edge resetting the client-facing leg of reused keep-alive connections.

What we observe

• Client is a React Native Android app using OkHttp (User-Agent: okhttp/4.12.0), HTTPS.
• The failing request reaches our backend — we have the forwarded request (with x-railway-edge: ams1, x-railway-request-id, x-forwarded-for) and the backend responds 200.
• Despite the 200, the client gets net::ERR_FAILED and never receives the response body.
• Clear pattern by connection state, not endpoint or method:
  • The first request on a fresh connection succeeds (e.g. a sign-in POST).
  • A subsequent request that reuses the keep-alive connection (a body-carrying POST shortly after) fails.
  • Plain GETs and fresh-connection requests are unaffected.
• It's not client-network-specific — it reproduces for multiple end users on different networks (mobile data and WiFi), so it's on the shared path through your edge.

What we've ruled out

• Backend errors — it returns 200; the failure is purely on the edge→client response leg.
• Client network / IPv6 — multiple users, multiple networks.
• We confirmed the connection-reuse trigger: disabling HTTP keep-alive on the client (one fresh TCP+TLS connection per request) eliminates the failures entirely. That strongly

implicates how the edge handles reused connections.