Status
Is the check service ready right now?
This page reads the same two endpoints the external monitor and the Fly proxy read. It is live, not a dashboard of every past minute.
Live checks
- Liveness (
/health) - Loading
- Paid checks ready (
/ready) - Loading
- Replay authority
- Loading
- Transparency log
- Loading
- Catalog, history and admission
- Loading
- Writer lease
- Loading
- Checked at
- Loading
When the replay authority reports not ready, paid checks answer HTTP 503 before any payment is verified and the free paths keep serving. That is the designed behaviour after a database restart or failover until an operator confirms continuity; it is described in the trust overview.
How it is monitored
A scheduled job outside Fly requests /health, /ready, the unpaid /route challenge, the transparency checkpoint and /rails every ten minutes. A failure opens or updates a public incident issue in the repository; recovery closes it. Fly's own proxy checks liveness every fifteen seconds and readiness every fifteen seconds and stops routing to a machine that fails liveness.
Separately, the hosting provider's metrics feed alert rules evaluated every minute: no running instance, the share of server-error responses, no traffic at all, memory running low and the data volume filling up. Each one pages a person by email. Thresholds and the paid-check volume are not published.
What "ready" does and does not promise
Ready means the admission policy is loaded, the private storage is writable, the replay authority answers as a durable primary with the pinned instance identity, and the transparency log can append. It does not promise a service level, a throughput figure or that every seller endpoint answers.
The writer lease is reported beside ready, not inside it. One process at a time holds the lease and is the only one that appends to the transparency log, crawls the catalog and runs housekeeping; the lease itself lives in the replay authority. A machine without the lease still answers ready: it serves every free path and completes plain paid checks by queueing their public leaf bytes for the writer, while session, binding and transparency requests answer HTTP 503 writer_unavailable until a writer holds the lease again.
The replay authority replicates asynchronously. A database failover can lose the newest acknowledged admissions; the fence stops paid checks until an operator reconciles them. A settled checking fee is durable on the primary the moment the answer is returned; the receipt in the answer is the buyer's own copy of that fact.