The check run
What a scheduled check stores, when it alerts and what happens if it fails.
Adding a monitor to heyo-status.config.ts doesn't start it. A scheduler or
manual trigger must call statusPage.cron.runChecks(). The local dev script,
Cloudflare and Vercel cron, node-cron, CLI and admin endpoint all use this
same operation.
Run one check batch
In a generated SQLite or PostgreSQL project, run one batch with:
That script uses the project-local CLI and requires Bun 1.4.3+. To run the CLI
directly, install it globally and use heyo-status check run. Cloudflare/D1
projects have no check script: use Worker cron, or the authenticated admin
endpoint if you've enabled adminToken.
You can also call the runtime directly, for example from your own scheduler:
What happens during a run?
- Check monitors. HTTP, DNS, SSL and TCP checks run concurrently. Composite monitors then combine the results from this same batch. The check phase is bounded roughly by the slowest check, but database writes and notification delivery take additional time.
- Detect transitions. Before writing new results, Heyo Status reads
previous check history. At the default
failureThreshold: 3, a down event fires on the third consecutive failure, not on every failed run. A successful check after that outage fires one recovery event. - Save and reconcile. One row per monitor is inserted into
status_results. Failing monitors can open automatic incidents; recovering monitors can move them toward resolution. - Refresh and notify. The process-local snapshot cache is invalidated, then events are sent to configured channels and eligible subscribers.
The results keep the order of monitors in your config, even though network checks run concurrently. A composite doesn't send its own network request; it depends on the results of its named component monitors.
What does a failed check mean?
A timeout or exception from an individual monitor becomes a failed
CheckResult with a message. It doesn't abort the other checks. Each result
contains name, type, status, message, latencyMs and checkedAt.
latencyMs measures the whole check, including any HTTP soft-retry delay.
The database stores those six fields; HTTP and SSL diagnostics on the
in-memory result aren't separate history columns.
The CLI exits with code 1 when any check fails, which is useful in CI.
A database error is different: the run cannot save the results and
rejects instead of returning a normal failed check. Notification delivery
failures are isolated and logged; they don't undo saved results or make the
run fail.
What about maintenance?
A monitor covered by an in-progress maintenance window is still checked and stored. It still appears in uptime history, but it doesn't produce monitor alerts or automatic incident changes during that window. An empty list of affected monitors on a window covers all monitors.
For scheduling, see Checks and scheduling. For what happens after a failed run, see Automatic incidents and Notification delivery.