Disk Randomly Unmounts Itself

Question

Dual 2.66GHz dual core Xeon
10.5.6 server
Once of the internal SATA HDs randomly unmounts itself, every one or two days. It's a bit of a pain as its the disc with all the user NHDs, so it's very disruptive.
It's only one partition on a disk that unmounts, the disk has another partition that doesn't unmount, so I think that makes it very unlikely to be a hardware issue.
I have absolutely no idea - would be very grateful for any clues!
thanks

Answer 1

Disks do fail.
My first suspicion here would be a disk error, and particularly a disk error located in an area of the disk "underneath" the partition that is dismounting. I'd expect that when the disk error is hit, the disk gets kicked out. (Disks are far better and more reliable than the giant disk drives I started out working with, but all disks can and do eventually fail.)
This might be a case that a block rewrite would fix. Disks can contend with isolated bad blocks; a rewrite can trigger the use of a spare block. This can involve rewriting the file with the block, or performing a disk erase, or a complete disk (or partition) reload from archive.
This assuming that this case is an isolated bad block.
Empirical studies show a roughly 6% failure rate per year; not the commonly-assumed bell curve of disk failures. Related studies have found measured MTBF figures appear to typically be somewhere around 3.4x and can be as much as 15x higher than the MTBF ratings, too. Older disk drives can fail at rates 30x higher than MTBF, too.
And measured latent hard errors (past what disk-level error correction could handle) were between 3 and 6 per terabyte, IIRC.
Put another way, how good is your RAID set (and -- same studies -- double disk failures in RAID sets are surprisingly common) and/or how current is your disk archive? RAID allows a way to retrieve data from a good block or from checksums when a bad block is encountered, of course. It doesn't and can't allow data to be read from a bad block, of course.
Regardless, I'd get an archive of the contents of the disk or the disk partition (using Disk Utility or otherwise), then would check the console logs for any interesting error messages around the time of the dismount, and would look to move to RAID and/or to a replacement disk.
I don't mean any of this to scare you. I do mean to scare you into having backups. And into taking steps to preserve and to recover this disk. And yes, even RAID fails.
Figures quoted here from my memory from reading Google and CMU studies. (And I could have a bit error in my own memory here, too.)