Planned downtime that does not count against you

What a window actually changes

A maintenance window is a promise to the system that failures during a time span are expected. While it is active, a monitor that fails does not open an incident and does not send alerts. Four switches decide the rest, and each window has its own: announce it on the status page, email subscribers, exclude the time from uptime figures, and keep checking. Announcing and exclusion are on by default.

Keep checking is off by default, which means the monitor simply stops running checks for the duration and nothing is recorded. Turn it on when you want the checks to carry on quietly: the window then accumulates the seconds of downtime that really occurred inside it, one check interval per failed check. That number is the useful one afterwards. A planned 30-minute migration that logged 24 minutes of real downtime tells you the plan held; one that logged 118 minutes in a 120-minute window tells you the migration was rougher than it looked.

How the uptime math changes, with numbers

Take a 30-day month, which is 43,200 minutes. Suppose a site is down for 20 minutes unplanned and 90 minutes during a migration. If the migration window is set to exclude downtime, uptime is computed against the unplanned 20 minutes only: 20 divided by 43,200 is 0.046 percent, so 99.954%. The migration still appears on the status page and in the monthly report as a maintenance event, just not as lost availability. If the window did not exclude downtime, the total would be 110 minutes, or 99.745%.

The difference between three nines (99.9%, about 43 minutes a month) and 99.7% is the difference between meeting and missing a typical SLA, so be honest about which downtime is genuinely planned. A window that stretches to cover a bad afternoon is a quick way to make reports lie. If your contract only excludes maintenance announced 48 hours in advance, schedule accordingly. The uptime calculator shows what each target allows.

One-off, weekly and monthly windows

A one-off window has a start and an end in a time zone you choose. A weekly window repeats every seven days from its first start, and a monthly one repeats on the same day of the month; when a month is shorter than the chosen day, the window falls on the last day of that month, so one set for the 31st still happens in February. After a daylight-saving change, glance at the next occurrence and check it still starts at the hour you expect.

A window with no monitors selected applies to every monitor in the account, which is convenient for a host-wide patch night and dangerous if you forget it exists. Name windows descriptively so the list is readable. Three patterns that work:

• Weekly patch night: Tuesday 02:00 to 03:00 Europe/Tallinn on the monitors of one server, announced, excluded.
• Monthly backup job: the 1st at 04:00 for 45 minutes with keep checking on, so you can see how long the site was really affected.
• One-off migration: Saturday 22:00 to 23:30, announced to subscribers a few days ahead.

Deploy scripts: what the API can and cannot do

Windows are created in the app. The REST API does not have a maintenance-window endpoint today, so a deploy script cannot open one. What a script can do, with a write-enabled key on Pro or above, is pause a monitor before the deploy and resume it afterwards, using the pause and resume endpoints. That silences alerts, but it differs from a window in two ways: the time is not announced to subscribers, and a paused monitor leaves a gap in its history instead of a recorded maintenance event. Always resume in a trap or finally step, or a failed deploy leaves the monitor silent.

For recurring deployments, schedule a window instead; for ad hoc ones, pausing is the pragmatic route.

What windows do not do

A window does not protect against unplanned failure: if the site is still broken after the window ends, a confirmed failure opens an incident as normal. It is also not the same as pausing, and it does not retroactively edit history: a window created after an outage cannot hide it. Certificate and domain-expiry notices are separate from outages and are not what a window is for.

Free, Starter, Pro and Agency all have maintenance windows, so there is no upgrade needed. Public visibility follows your status pages, and the effect on what goes to clients is covered in client reports.