Infrastructure & Operations › Working in Production

Maintenance Window

A scheduled, announced time for risky changes.

Backend Engineer · Mid-level Data Engineer · Mid-level

Also known as: maintenance window, scheduled maintenance, planned downtime

A maintenance window is a scheduled period for work that’s too risky or disruptive to do silently: a major database upgrade, a storage migration, a network change that may interrupt traffic. Announcing it in advance sets expectations, and choosing the time lets you do the work when the least harm is possible — often off-peak.

A well-run window has a few parts:

Announce it. Post on the status page, email affected customers, and tell support and on-call. People handle planned downtime far better than surprises.
Scope it. What changes, what will be unavailable, and how long it should take.
Have a rollback or stop point. Know how to back out, and set a “go/no-go” moment if it’s running long (see backups).
Communicate during it. Post when it starts, when it’s done, and if it’s extended.
Verify afterwards. Confirm the service is healthy, not just “the change ran”.

announce (days ahead) → start notice → do the work → all-clear → follow-up

The classic mistakes:

No communication. Customers discover the outage at the worst moment. Always announce.
Overrunning without an update. A window that silently runs long makes people assume the worst. Update as soon as you know.
No rollback. Even planned work fails. Have a way back before you start.
Using windows for routine changes. If deploys are safe (see continuous delivery), you don’t need downtime for them. Windows are for genuinely disruptive work.
Choosing a bad time. Scheduling over a peak or a known event turns a controlled window into an incident.

The trade-off is availability: a window deliberately accepts some downtime in exchange for doing risky work safely rather than in the dark. Where possible, prefer designs that let you migrate with no downtime at all — failover to a replica, rolling changes behind a load balancer — and keep windows for the cases that truly need them. The commitment you’re temporarily stepping on is your SLA.