Building block: frontend¶
lyndrix-ui is nginx with static files and a same-origin reverse proxy to the core
(LYNDRIX_BACKEND_URL). It has no state and scales to any number of replicas today.
Contract¶
LYNDRIX_BACKEND_URLpoints at the load balancer / Service, never at one core instance.- The ingress health check for the frontend is
GET /(nginx); core readiness is probed separately on/api/ready. - SSE passes through nginx without buffering (
proxy_buffering off,proxy_read_timeout 3600s— already indocker/lyndrix-proxy.conf). - Plugin React bundles come from the core (
/api/plugins/<id>/static/) → identical plugin artifacts on every core instance.
Options¶
| Option | Notes |
|---|---|
N lyndrix-ui containers behind the same balancer |
stateless; the shell reconnects its SSE stream by itself after an event: reconnect |
React served by the core (LYNDRIX_UI_ENGINE=react, /app) |
one container type less; the core instances serve the SPA |
| Legacy NiceGUI | process-bound websocket; needs sticky sessions — unsupported in cluster mode (preflight warns) |
What the shell does in a cluster¶
- The boot screen shows the cluster rows (coordinator, leader, peers, storage) from
GET /api/boot. - Runtime-status broadcasts carry
instance_id; the warm-up panel shows whichever instance is warming up. - A plugin pinned elsewhere renders a "runs on another instance" panel instead of a broken UI.
- Settings → Cluster shows instances, leader, leases, placements and the preflight, refreshed
over SSE (
cluster:changed). - The Android app (TWA) points at the public hostname → the VIP.
Known limits¶
- The shell has no per-instance view of plugin runtime status (it shows the latest broadcast).