Does Scheduled Maintenance Count Against SLA Uptime? The Exclusion Test
Usually not. Almost every service level agreement excludes properly announced scheduled maintenance from its availability calculation, which means a provider can take the service down for a maintenance window and still report 100% uptime for the month. The exclusion is standard and generally reasonable. The arguments happen over the conditions attached to it, because an exclusion with weak conditions lets a provider reclassify unplanned downtime as planned.
This article covers why the exclusion exists, the four conditions that determine whether a window genuinely qualifies, how it distorts published uptime figures, and what to check before signing. Your customers experience maintenance windows as downtime whatever the contract says, and that gap is the thing to manage.
Why the exclusion exists
The reasoning is sound. Systems need patching, databases need upgrades, and hardware needs replacing. Some of that work cannot be done without interrupting service, particularly on older architectures. Counting planned, announced, low-traffic maintenance the same as an unplanned failure would penalise providers for doing necessary work and push them toward deferring it, which produces worse availability over time rather than better.
So the standard formulation excludes maintenance from the availability calculation, provided it happens within the terms the agreement defines. Those terms are the entire substance of the clause. An exclusion reading scheduled maintenance is excluded with no conditions is effectively an unlimited exemption, because nothing stops the provider from declaring a window at short notice for as long as they like.
Our guide to managing planned downtime covers the operational side of running maintenance windows well, which is the other half of this.
The four conditions that make an exclusion legitimate
Call this the exclusion test, and apply all four when reading an agreement.
Notice period. How far in advance must the provider announce the window? Anything from 24 hours to a fortnight is common, and shorter notice periods are worth pushing back on. An exclusion with no stated notice requirement is not really conditional at all.
Duration and frequency caps. Is there a limit on how long a single window can run, and how much maintenance is permitted per month or quarter? Without caps, a provider could hold weekly four-hour windows and remain contractually perfect while being unusable for sixteen hours a month.
Timing constraints. Is maintenance restricted to defined low-traffic hours, and in whose timezone? This matters enormously for services with users in multiple regions. A window that is three in the morning for the provider is the middle of the working day for a third of the world.
The emergency maintenance carve-out. Most agreements permit emergency maintenance with little or no notice, and this is the clause worth reading closely. It is legitimate in principle, since a critical security patch should not wait a fortnight. It is also the clause through which unplanned downtime gets reclassified. Look for a definition constraining what qualifies as an emergency, and for whether emergency windows are capped or counted separately.
How the exclusion distorts published uptime
This is the practical consequence, and it is why comparing two providers' uptime figures is harder than it looks.
A provider running four hours of excluded maintenance monthly and experiencing 20 minutes of unplanned outage can publish 99.95% availability. From the customer's side, the service was unavailable for four hours and twenty minutes, which is worse than a 99.5% target allows. Both statements are accurate. They are measuring different things, and only one of them describes the customer's experience.
This is why maintenance policy is worth reading alongside any uptime number, and why the strongest position a vendor can take is publishing availability that includes maintenance windows. Very few do, and the ones that do are usually running architectures that no longer require downtime for routine work, which is itself the signal buyers should be reading.
From your own side, if you publish an uptime figure, decide explicitly whether it excludes maintenance and say so where the figure appears. Publishing an excluded number without labelling it is the kind of omission that reads as deliberate when a customer discovers it, even when it was not.
What to check before you sign, and what to do about it
Read the exclusions clause before the availability commitment, because a strong-looking percentage with a weak exclusions clause is worth less than a modest percentage with a tight one. Specifically: the notice period, whether duration and frequency are capped, whether timing is constrained and in which timezone, and how emergency maintenance is defined and limited.
Where the terms matter to your business, they are negotiable more often than people assume, particularly around timing. A provider will frequently agree to schedule windows outside your peak hours if you ask during procurement, and almost never if you ask afterwards.
Then measure it yourself regardless. Your own monitoring records unavailability without distinguishing planned from unplanned, which gives you the number that matches your customers' experience rather than the contractual one. Keeping both is useful: the contractual figure tells you whether a credit claim exists, and the raw figure tells you what your users actually got. If you build on a provider whose maintenance windows land badly for your users, that gap is a real cost and it is invisible in their reporting.
Common mistakes in handling maintenance exclusions
Reading the availability percentage without the exclusions clause. A 99.99% commitment with unlimited, short-notice maintenance is weaker than 99.9% with tight conditions. The exclusions determine what the number means.
Ignoring the emergency maintenance carve-out. It is the route by which unplanned downtime becomes excluded. Check whether emergencies are defined, capped, or counted separately, rather than left open.
Assuming maintenance timing suits your users. Windows are typically set in the provider's timezone or their largest market's. Confirm when they fall for your customers, and negotiate during procurement rather than after.
Publishing your own uptime figure without stating whether maintenance is excluded. Customers assume the number reflects their experience. An unlabelled excluded figure reads as concealment once discovered.
Measuring only the contractual number. Track raw availability too. The contractual figure tells you whether you have a claim; the raw figure tells you what your users experienced, and only one of those affects renewals.
FAQ
Does scheduled maintenance count against SLA uptime?
Usually not. Nearly all agreements exclude maintenance that was announced according to the terms they define, which means a provider can take the service down within a window and still report full availability for the period.
What conditions make a maintenance exclusion legitimate?
Four: a defined advance notice period, caps on duration and frequency, constraints on when windows may occur and in which timezone, and a properly bounded definition of emergency maintenance. An exclusion with none of these is effectively unlimited.
Can a provider claim 100% uptime while having downtime?
Yes, and it can be entirely accurate. If the downtime fell inside a properly announced maintenance window, the availability calculation excludes it. That is why maintenance policy should be read alongside any published uptime figure.
Should my own uptime figure exclude scheduled maintenance?
Either is defensible, but say which you have done wherever the figure appears. Customers reasonably assume a published number reflects their experience, and an unlabelled excluded figure damages trust more than a slightly lower honest one.
Closing thought
The gap this clause creates is the thing worth holding onto. Contractual availability and experienced availability are different measurements, and the maintenance exclusion is the largest single reason they diverge. A provider can be entirely honest, entirely compliant, and entirely unavailable to your users for four hours on a Tuesday morning, and every number in the reporting will look fine. Nobody is being deceived; the contract is simply not measuring what your customers felt.
Closing that gap requires a measurement that does not know about your contract. Odown checks from seventeen global locations at intervals down to one minute on every plan, including the twelve dollar tier, and records unavailability whether or not it was scheduled. Keeping that raw figure alongside the contractual one is what tells you when a maintenance policy that looks acceptable on paper is landing squarely on your busiest hours.
