Appearance
Monitors #
Monitors answer different operational questions. Choose the object that matches the failure you need to detect.
| Monitor | Use it to |
|---|---|
| HTTP endpoint | Poll an HTTP or HTTPS URL and validate status codes or response text |
| CORS baseline | Detect changes to expected CORS response headers |
| DNS delegation | Validate A, AAAA, CNAME, NS, or TXT answers against a baseline or rule |
| User impact | Poll journey and metric JSON produced by your application |
| Heartbeat | Detect a scheduled job that stopped sending pings |
| Single-device offline | Detect loss of connectivity for one critical device |
| Fleet offline | Detect any offline device, an offline count, or an offline ratio |
TCP socket checks, process checks, disk checks, and arbitrary on-device scripts are not current monitor types. Use an HTTP health endpoint, user-impact feed, heartbeat, logs, telemetry, or a fleet job when one of those better represents the outcome.
State and noise control #
Device and fleet monitors evaluate on an interval and count consecutive failures. They trigger only after the configured failure count. Recovery has its own consecutive-success count and can use a lower recovery threshold than the trigger threshold. A cooldown prevents repeated notifications while the same condition remains active.
For a ratio monitor, 0.10 means 10 percent. The minimum impacted-device guard must also be met. A fleet of 20 devices with a 10 percent threshold and a minimum of 3 does not trigger when only 2 devices are offline.
Disabling and re-enabling a monitor clears the start of the previous failure period. A newly created monitor may show no result until its first evaluation.
Monitors, alerts, and incidents #
- Device and fleet offline monitors create alerts. They do not directly open incidents.
- Service, heartbeat, and user-impact signals can feed incident automation.
- Log alert rules evaluate a supported subset of log search and send webhooks. They are separate from device alert rules.
- Status pages display selected monitor widgets. Hiding a widget is different from disabling its monitor.
Test changes against a staging target. Confirm both failure and recovery before relying on a route for on-call response.