After the api container restarts it gets a new IP; nginx cached the old one
at startup and kept proxying /api/* to the stale address — which by then
belonged to another service (casdoor), so login returned Casdoor's error
envelope and the frontend crashed with "Cannot read properties of null
(reading 'user')". proxy_pass now uses a variable + resolver 127.0.0.11 so
the upstream re-resolves per request. Verified: login works, and still works
after restarting the api container.
ali_mohaghegh