Recovery
What holds state, how to back it up, and how to roll back or recover a broken deploy.
What holds state
| Store | Holds | Backup / restore |
|---|---|---|
D1 emitkit-db |
users, sessions, orgs, API keys, projects, channels, identities, push subscriptions, 25-hour event buffer | Time Travel: any minute in the last 30 days (Paid plan) |
Analytics Engine emitkit_events |
events, ~90 days | Not restorable; export with the SQL API if you need a copy |
Alchemy state (alchemy-state-store Worker in the account) |
resource ids, generated BETTER_AUTH_SECRET and VAPID key |
Lives in Cloudflare; pnpm exec alchemy state read -r EmitKit/prod shows it |
.env.local (local, gitignored) and the Workers Builds secret |
deploy token | Recreate the token in the dashboard |
Common procedures
Roll back code: check out the last good commit, then pnpm release. Migrations are forward-only. If a migration broke something, write a new one that fixes it instead of editing the applied file.
Restore D1 to a point in time
pnpm exec wrangler d1 time-travel info emitkit-db
pnpm exec wrangler d1 time-travel restore emitkit-db --timestamp=2026-09-24T10:00:00ZExport D1: pnpm exec wrangler d1 export emitkit-db --remote --output=backup.sql
Lost Alchemy state (the state-store Worker or its Secrets Store was deleted): pnpm state:check fails and deploys stop. Run pnpm exec alchemy deploy --stage prod --env-file .env.local --adopt. Alchemy re-adopts the existing Worker, D1, queue and domains by name. The generated secrets can’t be recovered, so new ones are created:
- a new
BETTER_AUTH_SECRETsigns everyone out; - a new VAPID key stops existing push subscriptions: users re-enable notifications once in Settings → Notifications;
- the old analytics token (
emitkit-analytics-read) stays in the account; delete it in the dashboard.
Rotate secrets: delete the matching Random entry (pnpm exec alchemy state delete EmitKit/prod/BetterAuthSecret, or VapidPrivateKey), or change its id in alchemy.run.ts, then deploy. The consequences are the same as above.
Rotate the deploy token: create a new token, update .env.local and the CLOUDFLARE_API_TOKEN build secret, delete the old token. Nothing else changes.
Custom domain stuck / DNS conflict: the hostname has a record managed outside the Worker. Delete that record in the dashboard (DNS → Records), then deploy again.
Everything is down: pnpm exec wrangler tail emitkit shows startup failures. A ConfigError means a variable or secret is missing or malformed. A SCHEMA_MISMATCH means a migration hasn’t been applied (deploy again). Previous Worker versions are listed under Workers → emitkit → Deployments and can be rolled back from there as a stopgap.