A grace period before a service-down notification, so a quick reboot or update does not page you.
A service that loses its servers is now held for a configurable window — 30 seconds by default, on Notifications — before it is reported. A reboot or an update that takes it down for a few seconds passes silently.
- The clock starts when the service first goes bad and is not reset by a down → degraded → down flap, so a genuinely persistent outage still fires once the window has elapsed.
- A service that recovers inside the window says nothing at all — no down alert, so no "recovered" message either. A bounce is completely quiet, not a matched pair of pages.
- Once the alert has fired, its recovery follows as before.
- Set it to 0 to alert on the first check, the previous behaviour.
The setting is shared across the cluster, like the rest of the notification settings. It covers the health-check "servers down" alerts — the near-instant ones; the once-a-minute URL probe is a separate, slower path and is not grace-gated.
Upgrading from 1.91.x: nothing to do; existing installs get the 30-second default and can change it on the Notifications page.
Packages are attached below; the container image is at ghcr.io/avandeputte/haproxy-manager:1.92.0.