In the world of corporate computing—whether the actual number-crunching takes place on a company's own servers or out there on somebody else's—the work has to keep running pretty much all the time. Minutes of downtime, whether scheduled or unscheduled, can be a big deal. Hours of downtime? Forget about it.
So what are we to make of this post on the "Oracle Cloud Enterprise Hosting and Delivery Policies" site that says the company's "Cloud Service" may be subject to scheduled maintenance for about 10 hours every month? Ten hours is a very long time for important services to be missing in action.
Hence the interest in this document. The pertinent language is below:
Oracle reserves specific maintenance periods for changes that may require the Cloud Service to be unavailable during the maintenance period. Oracle works to ensure that change management procedures are conducted during scheduled maintenance windows, while taking into consideration low traffic periods and geographical requirements. The typical scheduled maintenance period is once a month on Friday, initiating at approximately 20:00 data center local time, lasting around 10 hours.
There is a related link that talks about exceptions, which is not publicly accessible. Fortune contacted Oracle Friday for comment and will update this story as necessary.
Having some services offline late on a weekend night may be trivial or a huge deal depending on which services are involved and what business you're in.
For certain applications in some businesses, it's not unusual to build six or seven hour scheduled maintenance windows into the weekend or overnight. There are some jobs, for example, that can be slated easily around maintenance. A payroll system, can go offline over the weekend to update tax tables, without a lot of angst. But, for a global high-traffic retailer, a payment system going offline over the weekend is unacceptable, said an IT consultant in the Washington D.C. area who works with Oracle government accounts and who requested anonymity because of those relationships.
One thing that's not crystal clear here is whether "Cloud Service" necessarily means the Oracle (orcl) database service. Or is it perhaps less critical but still important other application services. If it's the database, this could be an especially big issue
The consultant was so confounded by the document and he assumed it was an error or that something big was missing.
Get Data Sheet, Fortune’s daily newsletter on the business of technology.
"This seems wildly off," he wrote in an email. "I can't believe I could spend $2,000 a month for a two-zone Oracle Enterprise RDBMS and have 10 hours outages a month—plus any unscheduled time." RDBMS means relational database management system, known to most people more simply as a database.
Amazon (amzn) Web Services has its own relational database service or RDS, and schedules 30 minutes of maintenance weekly as outlined here. RDS users told Fortune that the way these updates are slated makes it easy to avoid downtime.
AWS, which started out as a favorite among software developers at startups 10 years ago, has gotten serious about winning business at big companies, where Oracle is a huge presence. Last year AWS launched a migration tool to move databases from customers' servers to AWS. The customers can then elect to keep running their databases on Oracle or Microsoft SQL Server, or (Amazon hopes) move that work to an AWS alternative like its Aurora database.
Amazon pioneered the model of creating huge data centers for data storage and networking that it rents to customers who are sick of maintaining their own data centers. That "public cloud" model poses an existential threat to old-line hardware and software companies like IBM (ibm) and Oracle, which made money selling pricey software and hardware into lots of customer data centers.
Now that more customers are renting AWS or Microsoft (msft) or Google (goog) data centers, the model has changed and Oracle has upped its cloud computing game to compete better in this new world. Its ace in the hole is that the Oracle database runs in many, many large companies, and a database is a hard thing to dislodge, migration tools notwithstanding.
“Oracle does make a great product, but this [maintenance window] is certainly bad optics from an enterprise perspective," said Kenneth White, senior vice president of research and development for Dovel Technologies, a technology consulting firm.
For more on cloud, watch:
Indeed, Oracle has a perception problem among cloud enthusiasts who think the company still doesn't "get" cloud. This 10-hour maintenance window may stoke that perception.
On Monday afternoon, Oracle emailed a statement about the Cloud Service post referenced above. That document, according to the company, is "a statement of policy that reflects the worst-case scenario and not the actual performance of Oracle's Cloud Platform."
Most of Oracle's "critical services like compute, Database as a Service, Storage Service, Java Cloud Services have zero planned downtime in the foreseeable future."
This story was updated at 4:28 p.m. with an additional quote and again at 5:38 p.m. on June 27 with Oracle's statement.