Uptime
Maintenance window
Maintenance window is a published time slot in which you expect to break or restart things. Work outside it counts as a surprise.
How it is measured
Record planned start, end, scope, and whether monitoring is paused or flagged. Compare actual duration to the announced slot. Overruns are the metric: a 30-minute window that takes 70 has a 40-minute overrun.
Decide how it affects SLO math. Some contracts exclude announced windows up to a cap, such as 4 hours per month with 72 hours of notice. Probes should keep running and tag results rather than switch off.
Worked example
A school-district parent portal announces a window for Sunday 02:00 to 03:00 local time for a database major version upgrade, with a banner posted 72 hours ahead. The upgrade finishes the data move at 02:48, but the app fails its smoke test until 03:25.
Announced window: 60 minutes. Real: 85. The extra 25 minutes fall outside the window and count against availability, even though the first 60 do not.
How it differs
Maintenance window is the slot on the calendar. Scheduled maintenance is the work done and the notice given. A window is the permission period. Scheduled maintenance is the activity and its communication, and it can overrun a window.
Common errors
Not announcing. Windows at the busiest local hour. Muting monitoring entirely. Open-ended windows. Not testing afterwards, or not saying when it is done.
In practice
Choose a recurring slot at your lowest traffic hour using a week of hourly pageviews. Post it on the status page and in a site notice. Keep monitoring running and annotate it rather than pausing.