fix(infra): nginx upstream DNS staleness — 502 after backend restart

- Replace upstream blocks with variable-based proxy_pass + Docker DNS
  resolver (127.0.0.11) so nginx resolves hostnames dynamically on
  each request instead of caching at startup only.
- Add explicit 'docker compose restart proxy' in CI deploy step
  (belt-and-suspenders: also forces nginx restart after deploy).
- Remove no-op sed commands from CI (docker-compose.yml already uses
  GitLab registry, no ghcr.io replacements needed).

Root cause: docker compose up -d only recreates containers whose image
changed. When backend container is recreated (new Docker IP), nginx
still caches the old IP → 502 Bad Gateway on /api/*

.gitlab-ci.yml

@@ -115,15 +115,19 @@ deploy-vps:
         echo '$CI_JOB_TOKEN' | docker login $CI_REGISTRY -u '$CI_REGISTRY_USER' --password-stdin
 
         echo '→ Updating docker-compose image references...'
-        sed -i 's|ghcr.io/\${OWNER:-mytheclipse}|$REGISTRY_PROJECT|g' infra/docker/docker-compose.yml
-        sed -i 's|\${OWNER:-mytheclipse}|$CI_PROJECT_NAMESPACE|g' infra/docker/docker-compose.yml
-
         echo '→ Pulling latest images...'
         docker compose -f infra/docker/docker-compose.yml pull
 
-        echo '→ Restarting containers...'
+        echo '→ Restarting containers (recreates if image changed)...'
         docker compose -f infra/docker/docker-compose.yml up -d --remove-orphans
 
+        # Force restart proxy to pick up new upstream DNS IPs.
+        # Docker's DNS changes when backend/frontend containers are recreated,
+        # but nginx only resolves upstream hostnames at startup. Without this,
+        # nginx keeps pointing to stale container IPs → 502 Bad Gateway.
+        echo '→ Ensuring proxy container is restarted (nginx upstream DNS refresh)...'
+        docker compose -f infra/docker/docker-compose.yml restart proxy
+
         echo '→ Cleaning up...'
         docker image prune -f
         echo '✓ Deploy complete'

infra/docker/nginx/nginx.conf

@@ -1,10 +1,10 @@
-upstream backend {
-    server backend:3000;
-}
-
-upstream frontend {
-    server frontend:3000;
-}
+# Docker DNS resolver (127.0.0.11 = Docker's embedded DNS).
+# Required for variable-based proxy_pass below to resolve upstream
+# hostnames on each request instead of caching them at startup.
+# Without this, when backend/frontend containers are recreated (new IP),
+# nginx keeps pointing to stale IPs → 502 Bad Gateway.
+# Valid=10s re-resolves at most every 10 seconds to avoid excessive DNS queries.
+resolver 127.0.0.11 ipv6=off valid=10s;
 
 map $http_upgrade $connection_upgrade {
     default upgrade;
@@ -17,7 +17,8 @@ server {
 
     # API proxy — matches /api and /api/*
     location ^~ /api {
-        proxy_pass http://backend;
+        set $backend_url "http://backend:3000";
+        proxy_pass $backend_url$uri$is_args$args;
         proxy_set_header Host              $host;
         proxy_set_header X-Real-IP         $remote_addr;
         proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;
@@ -26,7 +27,8 @@ server {
 
     # WebSocket proxy — matches /ws and /ws/*
     location ^~ /ws {
-        proxy_pass http://backend;
+        set $backend_url "http://backend:3000";
+        proxy_pass $backend_url$uri$is_args$args;
         proxy_http_version 1.1;
         proxy_set_header Upgrade              $http_upgrade;
         proxy_set_header Connection           $connection_upgrade;
@@ -43,7 +45,8 @@ server {
 
     # Frontend SPA fallback
     location / {
-        proxy_pass http://frontend;
+        set $frontend_url "http://frontend:3000";
+        proxy_pass $frontend_url$uri$is_args$args;
         proxy_set_header Host              $host;
         proxy_set_header X-Real-IP         $remote_addr;
         proxy_set_header X-Forwarded-For   $proxy_add_x_forwarded_for;