A maintenance window is a named, scheduled period tied to one or more configuration items, with alert suppression switched on by default — so planned work like a server reboot or a patch cycle doesn't light up your incident queue with expected noise.
Setting one up
From Maintenance Windows, create a window with a start and end time, attach the CIs it covers, and leave alert suppression on (or turn it off if you specifically want alerts to keep firing through the window). While it's active, alerts against those CIs are muted at the source rather than arriving and needing to be manually dismissed.
Example
A team schedules a 2-hour maintenance window on a database cluster for a planned failover test. Alerts on that cluster are suppressed for the duration, so the expected connection blips during the failover don't open a dozen incidents — while an unrelated issue on a different system still alerts normally.