Health checks & errors

Two small hooks for running Keel in production: a health endpoint for uptime monitors, and one place every uncaught server error passes through.

/api/health

app/api/health/route.ts runs select 1 against the database with a 3 second timeout and answers:

{ "status": "ok", "database": "ok", "latencyMs": 12 }

with a 200, or { "status": "down", "database": "down" } with a 503 when the query fails or times out. Point any uptime monitor at it and alert on anything that isn't a 200. It's public on purpose and never includes error messages, hostnames or versions, the real error goes to the server log. Responses are Cache-Control: no-store, so a CDN can't keep serving a stale "ok".

Server errors

Next calls onRequestError in instrumentation.ts for every uncaught error in a route handler, Server Component render or proxy.ts. Keel hands it to reportServerError in lib/observability.ts, which:

• Logs one JSON line to stderr with the message, the error digest, the route and the method. The digest is the same "Reference" the error pages show, so a user's screenshot maps straight to a log line.
• POSTs the same JSON to ERROR_REPORT_WEBHOOK_URL when that's set (a log drain, an incident tool, your own collector), with a 3 second timeout. A failing webhook is logged and never throws.

Query strings and request headers are left out of every report, they can carry reset tokens, session cookies and API keys.