Setup:

- Two services in the same project and environment (production): a

Python/FastAPI backend and a Railway-managed PostgreSQL database

- Both services have private networking enabled

- Backend DATABASE_URL points to postgres.railway.internal:5432

- Backend runs alembic upgrade head (database migrations) on startup

before starting the web server

Problem:

The backend container cannot resolve the Postgres private hostname.

Every deploy crashes with:

socket.gaierror: [Errno -2] Name or service not known

DNS debug output from inside the backend container:

resolv.conf:

search internal railway.internal

nameserver fd12::10

Trying to resolve postgres.railway.internal...

IPv6: [Errno -2] Name or service not known

IPv4: [Errno -2] Name or service not known

Any: [Errno -2] Name or service not known

The Railway internal DNS nameserver (fd12::10) is configured in

resolv.conf but returns no records for the Postgres service hostname on

any address family.

What I've tried (all failed):

1. Multiple redeploys of the backend service

2. Redeploying the Postgres service, then the backend

3. Renaming the Postgres private hostname (e.g. postgres → db),

redeploying Postgres, then redeploying the backend with updated URL

4. Renaming back to postgres, redeploying Postgres, then backend

5. Adding a DNS retry loop with backoff in the startup script (retries

5 times over 15 seconds — never resolves)

What changed:

This was previously working with a public proxy URL. I removed the

public proxy on the Postgres service to use private networking only.

Since then, the private hostname has never resolved.

Container info:

- Base image: python:3.12-slim

- The DNS debug runs Python's socket.getaddrinfo() directly — same

resolution path the database driver (asyncpg) uses