The nginx config written by install.sh proxied waiter.*, manager.* and the
IP default_server without proxy_http_version 1.1 or Upgrade/Connection
headers, so /api/ws/connect never upgraded and live events (new orders,
KDS status, chat, phone calls) never reached waiters or the manager. The
repo's nginx-proxy/nginx.conf had the same gap on the manager block.
Both configs now use a $connection_upgrade map, 1h read/send timeouts and
proxy_buffering off (SSE) on every proxied location. Verified with nginx -t
and a header-echo upstream: old config strips Upgrade, new one forwards it.
Existing sites: copy the new install.sh, re-run it, restart the proxy.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bonaminandClaude Opus 5.5
0f9946e6ca