We are experiencing a critical issue with internal service-to-service communication in our production environment. Identical configuration works flawlessly in our acceptance environment, but fails in production.

Environment Details:

Project ID: a7e5849c-cbde-48f7-a765-fd62d670758b

Production Environment ID: 665f0795-45b5-4e00-a9b0-f67545b41232

Acceptance Environment ID: e2d00610-2b29-40db-9983-753e53ddb627 (working)

Problem:

Internal service-to-service communication fails with IPv6 connection refused errors. The Tracking proxy (Caddy reverse proxy) cannot reach Rybbit Client and Rybbit Backend services via their internal Railway DNS names.

Error Log:

{ "message": "dial tcp [fd12:28b7:3311:0:1000:1a:7cae:56cc]:3002: connect: connection refused", "severity": "error", "attributes": { "err_trace": "reverseproxy.statusError (reverseproxy.go:1626)", "logger": "http.log.error" }, "status": 502}

Services Involved:

Tracking proxy (Caddy 2.11.4) → attempting to reach:

backend-ws3y.railway.internal:3001 (Rybbit Backend)

client-q3wd.railway.internal:3002 (Rybbit Client)

Key Observations:

All services are running and healthy (status: online)

Configuration is identical between acceptance and production environments

Internal routing works perfectly in acceptance environment

Production environment shows IPv6 connection refused errors

Services cannot communicate internally despite being in the same private network

This affects all internal service-to-service communication (tested with both Caddy proxy and simple Hono scripts)

Expected Behavior:

Services should be able to communicate via internal Railway DNS names within the same environment's private network, regardless of IPv4/IPv6 protocol.

Actual Behavior:

IPv6 connections to internal service DNS names are refused in production, while the same setup works in acceptance.

Impact:

Complete service-to-service communication failure in production. The application is non-functional.