What maintenance should a dedicated server have?

From running a fleet of dedicated servers, we would open tickets to the provider for disks when they had read errors or (later) when SMART values indicated likely failure [1] or if the disks just disappeared (like SSDs like to do), we'd open ram tickets for excessive ECC correctables (more than a couple a day) or the second ECC uncorrectable, and for network issues which were sometimes diagnosed as NIC issues, we got a couple CPU cache errors too. Sometimes we'd get maintenance tickets for overheating CPUs (they'd need to unrack the machine and reapply thermal grease and the heatsink) and maybe for failing power supplies and fans.

If you don't do this proactively, it's not the end of the world, it just means you don't set the maintenance window:

If you don't manage disks to catch failures early, the problem(s) should be obvious when you look at system logs because your host is offline.

If you don't alert on ECC correctables and the ram gets really bad, you may get a large enough volume of errors that the machine check exception processing brings your application to a crawl, or you may get uncorrectable errors, and most OSes will halt (I think Solaris? and some other more enterprise OSes have more granular responses; blacklisting the page with the error if it's not in OS critical memory, and only killing process(es) that try to read or write the broken page).

Network issues are probably similar to network issues faced in VMs. It's a little easier to diagnose them, because there tends to be less layers, but there's plenty of weird stuff, if you push a lot of traffic.

If you run your stuff long enough, you'll need to move to new machines from time to time; either because you want newer hardware, or the old hardware breaks, or the provider wants to clear out older gear (but that probably happens at 10-15 years old; most likely you'll want to move before they force you).

[1] (10 bad sectors in a day or 100 total would be a decent threshold, but maybe your provider wants to wait for it to show failed)

- You will want to do regular firmware updates, say once or twice a year. For example, to prevent this "SSDs locking up after 40K hours" issue that may have taken the HN primary and backup down recently.

- Regular OS updates, say weekly or biweekly.

- RAID array monitoring to you know when the first drive dies rather than the second.

It's probably pretty similar to your VPSes otherwise. Make sure you have backups.

Oh boy this reminds me

7:26:28 up 622 days, 12:39, 1 user, load average: 0.30, 0.31, 0.32