a month ago
We are evaluating a generic Node service on Railway before any provisioning or deployment. Could you point us to authoritative guarantees or a supported configuration for the following?
-
Is there a stable, authoritative list of immediate source IPs presented to the application's TCP socket by Railway's HTTP edge, including region/redeploy/change behavior? Are these peers exclusively controlled by the edge, or can they also represent internal services/other tenants?
-
How can we ensure that only the intended HTTP edge reaches that listener, excluding raw TCP, alternate public ingress and other services in the project/environment?
-
Does the edge replace all client-supplied X-Forwarded-Proto headers with exactly one https header, including duplicate/chained client values? Does it preserve the original raw Host header unchanged, including a port when present?
-
If these socket-peer guarantees are unavailable, is an authenticated edge-to-origin transport (such as verified TLS/mTLS or an authenticated proxy assertion) supported on Trial with one application process? Please identify any plan/cost requirement before suggesting activation.
-
Can a Trial service use one 0.5 GB persistent volume with a 0700 private subdirectory for SQLite and one Node process, without replicas or overlapping mounted deployments?
We do not want to rely on a sampled IP, public anycast addresses, X-Railway-Edge, broad private CIDRs or forwarded client headers as proxy authentication. We are requesting documentation and configuration guidance only, with no request to provision resources or change our plan/access.
1 Reply
a month ago
We don't publish a stable list or range for the socket-peer addresses your app sees from our HTTP edge. Any observed address, such as one in 100.64.0.0/10, is not a supported contract and may change. We also offer no edge-only ingress switch, no edge-to-origin mTLS or verified-TLS identity, and no signed proxy assertion, on any plan.
The listener is not edge-only. Other services in the same project and environment can reach its port over private networking and send their own forwarding headers, and a TCP proxy, if added, passes raw bytes through unsanitized. Private networks are isolated between projects and environments, so the environment is your trust boundary.
For requests through the HTTP edge, we overwrite client-supplied X-Forwarded-Proto, X-Forwarded-For, X-Real-IP and X-Forwarded-Host rather than appending. Our specs describe X-Forwarded-Proto as https and X-Forwarded-Host as the original host header. Public TLS ends at the edge, and the edge-to-app hop runs in an encrypted WireGuard tunnel. See the public networking specs.
On Trial, a service can have one 0.5 GB volume. Volumes can't be used with replicas, and we prevent multiple deployments from being active and mounted at once, so redeploys have a short gap. Images running as a non-root UID can hit volume permission issues. See the volume reference.
Status changed to Awaiting User Response Railway • about 1 month ago
a month ago
This thread has been marked as solved automatically due to a lack of recent activity. Please re-open this thread or create a new one if you require further assistance. Thank you!
Status changed to Solved Railway • about 1 month ago