I just checked that last 90 days of ncidents for eastern US for Azure. And I saw 19 incidents. None of them accounts for a full outage, only a particular feature may be down.
I have worked with mainframe maintenance. For one this "24/7/365 for decades at a time" isn't really true. As even the best maintained system will have scheduled outages. Otherwise you may have a "God help us if this system ever fails" issue, as a failure over such time could show off all the other systems that have failed.
I have seen such long running systems, up for years, without being maintained. The system just did its thing and did it well. So much so that the application ran completely in active RAM, and once the system went down, we found out the drives failed a long time ago. Why didn't we see it in the logs? Well the logs were attempted to be written to the disk. And the operator at the time had a tendency to ignore and clear out console errors on the terminal.
Also much like Azure, while the System may be running, there are often parts of it that may fail, and would need to get restarted. Stop and restart the database. Unmounted the drive and replace a new one.