So we have a service on Railway that automatically redeploys from GitHub. An Nginx proxy sits in front of it and forwards requests from /api to the internal service endpoint.

Everything works fine initially. However, after the service redeploys:

• API requests start hanging
• Eventually they time out
• No response is returned from the upstream service

Interestingly, if we open the Nginx configuration UI and simply click “Save” (without making any changes), everything immediately starts working again.

What’s actually happening

When Railway redeploys the service, the underlying container is replaced and its internal IP address changes.

Nginx, however, resolves the service hostname only once when it starts and continues using that cached IP.

So after a redeploy:

• Nginx is still sending requests to the old (dead) IP
• Connections hang because that container no longer exists
• When we “Save” the config, Nginx reloads → DNS is resolved again → correct IP is used → things work

Solution

We need to make Nginx re-resolve DNS dynamically, instead of caching it forever.

Fix

Add a DNS resolver and force runtime resolution:

resolver 1.1.1.1 8.8.8.8 valid=10s;
resolver_timeout 5s;

set $backend "your-service.internal";
proxy_pass http://$backend;

Why this works

• resolver tells Nginx how to resolve DNS
• valid=10s ensures it refreshes periodically
• using a variable ($backend) forces Nginx to resolve the hostname at request time instead of startup

Result

After this change:

• Nginx automatically picks up the new IP after redeploys
• No more hanging requests
• No need to manually reload or “save” config