Verify and troubleshoot

Deployment checks

Migrations applied.

db:migrate ran clean against the Neon database from packages/db.

HTTP service answers.

Its public /health URL returns a response, and the scenes API works behind authentication.

WebSocket service upgrades.

A room page connects, presence appears, and commits reconcile across two tabs.

Flush worker running.

Room edits become durable scene revisions in Postgres — not just Redis state.

Web app wired.

The deployed editor points at the deployed API and WebSocket URLs, and sign-in plus invites work end to end.

Symptom	Likely cause and fix
First request slow	Cold start — the service was spun down and takes about a minute to wake. Narrow it with a keep-alive window.
WebSocket won't connect	Wrong port (`WS_PORT`, not `PORT`), `wss://` vs `ws://` mismatch, or origin not allowed — check `WEB_ORIGIN` and `WS_ALLOWED_ORIGINS`.
CORS or cookie errors	`WEB_ORIGIN` does not match the deployed web URL, or `BETTER_AUTH_URL` points at the wrong origin.
Changes not persisting	The flush worker is not running — room state is sitting in Redis only.
`NEXT_PUBLIC_*` change not applied	Those values bake in at build time — redeploy the web app.

Last verified: 2026-10-11.