Documentation

Check statuses & lifecycle

A heartbeat check has five statuses. Understanding the transitions explains exactly when an alert arrives — and why sometimes it (correctly) doesn't.

Status	Meaning	Alerts
new	the check exists, no ping has arrived yet	none
up	pings arrive on time	the up event on recovery after down
late	the deadline (+ up to 30 s of slack) was missed, the grace period is running	the late event (if enabled)
down	grace expired, or the job explicitly reported failure	the down event + reminders
paused	monitoring switched off manually	none

Why new never alerts

A check in the new status is not tracked by the scheduler: it has no deadline yet — one appears only with the first ping. So you can create checks for future jobs in advance (or roll the config out with a deploy): until the code ships and pings, there are no false downs. The first successful ping flips the check to up and sets the first deadline.

The deadline: period and cron

The deadline is the moment the next ping is due. It is computed from the actual time of the last ping:

• period — last ping + period_sec. The schedule "floats": a job that starts with drift does not accumulate error;
• cron — the next run of the cron expression in the check's timezone after the last ping. DST transitions are handled by the timezone;
• OnCalendar — like cron, using a systemd expression.

Grace

Grace is the allowance after the deadline — insurance against schedule jitter and long runs: cron rarely fires to the second. The check becomes late not at the very second of the deadline but after a small slack — deadline + min(grace, 30 s): otherwise the minute-precision scheduler would outrun a ping arriving a second behind schedule, and minute-level cron jobs would blink up↔late. The slack never moves the down moment — that stays at deadline + grace_sec:

up ──(deadline + min(grace, 30 s))──▶ late ──(deadline + grace)──▶ down
 ▲                                                     │
 └──────────────── any successful ping ◀───────────────┘

With grace ≤ 30 s the late window collapses: the late threshold coincides with the down moment and the check goes straight to down. Grace is set per check: 0 to 30 days. Time-based transitions are executed by the scheduler with minute precision. The down alert arrives within ~1–2 minutes of grace running out: the overdue check is found by the minute-precision scheduler, the delivery itself takes seconds.

Instant down

An explicit failure signal bypasses grace: a ping to /<uuid>/fail or with a non-zero exit code (/<uuid>/42) flips the check to down immediately — the job itself said it failed, there is nothing to wait for.

Recovery, pause and reset

• from late and down the check returns to up with the very next successful ping. The "up" alert goes out only when recovering after down; a late → up transition is just a slightly delayed ping — no alert, nothing in the delivery journal;
• paused — a manual pause: pings don't change the status, deadlines aren't tracked, no alerts. For HTTP checks paused means "don't probe";
• reset (the "Reset" button on the check page) returns the check to new, as if it had just been created: the last ping, duration and deadline are cleared. The ping and flip history is kept — the reset itself shows up in the timeline as its own event. Counting starts over with the first ping; handy after a long idle spell or when a job moves.

Unstable checks (flap damping)

A check that keeps bouncing between up and down ("flapping") can send dozens of identical alerts within an hour. CronAlive dampens that storm automatically:

• alertable status changes — down and the down → up recovery — are counted in a sliding 1-hour window;
• on the 6th change within an hour a single "check is unstable: N status changes in an hour, alerts paused" alert goes out instead of the regular one — to every channel subscribed to down events;
• further down/up alerts are suppressed but journaled: the delivery journal shows them as suppressed with the reason "the check is unstable (flap damping)" — it is always clear why things went quiet;
• the check page shows an amber "unstable" badge next to the status;
• after 30 minutes without a single status change a "has stabilized, current status: …" alert arrives — it states since when the alerts had been paused — and regular alerting resumes.

late alerts neither move the threshold nor get suppressed — they are an early signal. Manual operations (pause, resume, reset) do not count as flips. Defaults: threshold 6 changes, window 1 hour, 30 minutes of silence to exit.

Pause, resume, delete and tag assignment also work in bulk — tick the checkboxes in the list. From the check page you can clone a check (config without history) and move it to another project of the account.

Statuses are visible on the dashboard, in the API and on public badges; HTTP checks use K confirmations instead of ping timing.