Pause a monitor while you deploy, migrate or reboot. It keeps its history, stops running its policy, and resumes exactly as it was, either when you say so or at a time you set in advance.
The escalation policy does not run, so a planned restart does not wake the on-call engineer or fill a channel.
An outage you caused on purpose does not become a record that has to be explained later.
Past incidents, uptime and response-time data all remain. The monitor picks up where it left off.
The same interval, the same policy, the same connections. There is nothing to reconfigure afterwards.
Anything where you already know the service will be unavailable for a period.
Kernel updates and host restarts that take a machine down predictably.
Schema changes and reindexing that make a backend refuse connections for a while.
Scheduled work by a hosting or network provider that you did not initiate but do know about.
Set the end time when you start, so the monitor is watching again the moment the window closes.
If a monitor is alerting too often, the fix is a confirmation threshold or dampening. Pausing it hides the problem instead.
Pause what the work touches, set an auto-resume time, and let everything else keep watching.