A pre-announced period when degraded service is acceptable, the dying art that zero-downtime deploys are replacing.
A maintenance window is a scheduled, customer-announced time window during which a service may be partially or fully unavailable while the team performs disruptive work, database migrations, infrastructure upgrades, security patches. The window is communicated in advance, falls in low-traffic hours, and excludes the time from SLO accounting. Modern SRE practice is shifting away from maintenance windows toward zero-downtime techniques (rolling upgrades, online schema migrations, blue-green deploys), but they remain necessary for some classes of work.
Maintenance windows that go long and creep into business hours destroy SLO budgets and customer trust faster than any single outage. Rigorous pre-window rehearsal (in staging, on a copy of prod data) is what turns a 4-hour planned window into a 90-minute one. The discipline of doing fewer, shorter windows over time is also a forcing function to invest in zero-downtime tooling.
See the part of the platform that handles maintenance window in production.