Monitoring

Kept sends nothing about itself anywhere unless you set it up: no telemetry of any kind, not even opt-in install counts.

Health checks

Path	Answers
/healthz	200 while the process serves HTTP. For a liveness probe.
/readyz	200 when both of the web’s database connections answer, else 503. For readiness, and the image’s own HEALTHCHECK.
/version	the version, the commit it was built from, the source repository, and where the third-party notices are (/notices.txt, in the image).

A container running only the worker (KEPT_ROLE=worker) serves no HTTP. The image’s HEALTHCHECK follows KEPT_ROLE: for the worker it checks that the job loop has polled the database within the last two minutes (it touches /tmp/kept-worker-alive as it goes), so a split deployment needs no healthcheck override.

The status page and alerts

Admin → Status shows what needs attention: the recovery kit, mail, failed background jobs, the reminder scan, and backups: the backup’s age, the last restore drill, bucket versioning, disk space, the release and its migrations, the update check, HTTPS and the connectors. Admin alerts reach instance admins in the app, by mail and by push; each is resolved when its cause clears.

Metrics

Set KEPT_METRICS_TOKEN and /metrics serves Prometheus text to requests that send it as a bearer token (Authorization: Bearer <token>); without the variable, the path doesn’t exist.

Metric	What
kept_build_info{version, revision}	always 1; the running build
kept_process_uptime_seconds	seconds since the process started
kept_process_resident_memory_bytes	the process’s memory
kept_db_pool_connections{pool, state}	connections per database pool: total, idle, waiting
kept_backup_last_success_timestamp_seconds	when the last good backup finished (0 before the first)
kept_backup_last_run_status{status}	1 for the last finished run’s status, 0 for the others
kept_disk_used_ratio{volume}	how full the data disk and a directory backup target are
kept_update_available	1 when the update check has seen a newer release

No label ever holds a path or a target. A database that can’t answer leaves the backup, disk and update lines out; the scrape itself still succeeds.

Logs

JSON lines on standard output (KEPT_LOG_FORMAT=json, the default), or pretty for reading docker compose logs by eye; KEPT_LOG_LEVEL sets the level. Every request has an id, sent back in the x-request-id response header, so a report can be matched to its log line. Tokens in URLs are redacted from the log, and secrets and passwords are never logged.

Tracing and error reporting

Both are optional and off by default.

• Tracing: with OTEL_EXPORTER_OTLP_ENDPOINT set, Kept sends OpenTelemetry traces of its HTTP requests and database queries to your collector. Unset, none of the tracing code even loads, and a collector that is down never slows a request.
• Error reporting: with a Sentry-compatible DSN in KEPT_ERROR_DSN, unhandled errors are reported with the request id and route, and never a request body, header, query string, name or anything stored in a location.