Hi Railway team,

We are deploying a project with two services in the same Railway project and region:

• Frontend: Next.js / Node 20
• Backend: Flask / Gunicorn, listening on port 5000

We would like to keep the backend private-only, without exposing a public URL, and have the frontend call it through Railway private networking:

BACKEND_URL=http://backend.railway.internal:5000

However, we are seeing unstable/slow behavior from the frontend container when calling the backend private domain.

Inside the frontend container, we installed curl with:

apt-get update && apt-get install -y curl

Then tested the backend private URL:

curl -4 -w "\nTOTAL=%{time_total}s\n" -o /dev/null -s http://backend.railway.internal:5000/api/auth/me

Results vary a lot. Sometimes IPv4 is fast, but often it takes several seconds:

TOTAL=0.007972s

TOTAL=3.483850s

TOTAL=8.417265s

IPv6 is consistently fast:

curl -6 -w "\nTOTAL=%{time_total}s\n" -o /dev/null -s http://backend.railway.internal:5000/api/auth/me

TOTAL=0.006609s

TOTAL=0.006232s

TOTAL=0.001799s

We also enabled this on the frontend service:

NODE_OPTIONS=--dns-result-order=ipv6first

And confirmed inside the frontend container:

echo $NODE_OPTIONS

--dns-result-order=ipv6first

node -p "process.env.NODE_OPTIONS"

--dns-result-order=ipv6first

DNS resolution is correctly returning IPv6 first:

node -e "const dns=require('node:dns'); dns.lookup('backend.railway.internal',{all:true},console.log)"

Output:

null [

{ address: 'fd12:f284:7247:1:e000:2b:ce14:c87d', family: 6 },

{ address: '10.148.200.125', family: 4 }

]

But Node fetch / Undici still fails or times out when calling the internal backend URL:

node -e "const t=Date.now(); fetch(process.env.BACKEND_URL + '/api/auth/me').then(r=>r.text().then(()=>console.log(r.status, Date.now()-t+'ms'))).catch(console.error)"

Output:

TypeError: fetch failed

[cause]: ConnectTimeoutError: Connect Timeout Error

attempted addresses:

fd12:f284:7247:1:e000:2b:ce14:c87d:5000

10.148.200.125:5000

timeout: 10000ms

Important notes:

• The backend itself is healthy and responds quickly.
• Calling the backend through its public Railway URL works correctly and solved the issue.
• Backend -> Postgres is also fast. A simple SQLAlchemy select 1 test from the backend container gives ~8-12ms after the first connection.
• The issue appears isolated to frontend -> backend over Railway private networking, especially Node/Undici fetch behavior against backend.railway.internal.

Questions:

1. Is Railway private networking expected to be IPv6-first or IPv6-only for reliable service-to-service traffic?
2. Is there a known issue with Node/Undici fetch and Railway internal domains?
3. Is there a recommended way to call private services from a Node/Next.js service without exposing the backend publicly?
4. Should the backend bind differently for private networking? Currently Gunicorn binds to 0.0.0.0:5000.

Our goal is to remove the backend public URL and keep backend traffic private, but currently the public URL is the only stable workaround.

Could you help us understand the correct private networking setup for this architecture?