User reported persistent 502 with upstream "frontend:3000" after the
previous fix that changed the upstream to "frontend:80". Symptom is
a stale conf still being served by the nginx container - the host
volume mount was keeping an old file in play (cached image, missing
git pull, or the conf simply not being re-read).
Make this class of bug impossible: ship the conf inside the nginx
service's image. A fresh build now guarantees the conf in the
container matches the conf in the repo.
- nginx/Dockerfile added (FROM nginx:alpine + COPY nginx.conf)
- docker-compose nginx service: image -> build ./nginx; remove
the conf bind mount entirely.
Deploy: git pull && docker compose build nginx frontend && docker compose up -d --force-recreate
User reported nginx upstream connect refused for frontend:3000.
vite preview is dev/preview-oriented and has been observed dropping
its listener in docker production environments.
- Frontend Dockerfile: multi-stage build → nginx:alpine serves /usr/share/nginx/html
- Frontend nginx.conf: SPA fallback (try_files ... /index.html) so client-side
routes like /domains survive a browser reload, plus immutable cache for /assets/
- docker-compose: frontend now exposes 80 instead of 3000
Top-level nginx upstream (server frontend:3000) already resolves by service name;
port mapping in upstream is unaffected because http upstream uses the resolved
address and the upstream block targets the container's listening port. Updating
to frontend:80 happens automatically because the upstream uses the service name
without an explicit port override.
Actually correction: the upstream IS port-bound. Updating both ends in one commit.
