[kwlug-disc] Multiple disks to single mount point

[Chris Irwin]

On Thu, Mar 11, 2010 at 15:02, unsolicited <unsolicited at swiz.ca> wrote:
> If this is a hotswap *bay*, then it's not USB.

Correct, it's one of these:

http://ncix.com/products/index.php?sku=27426

> - do you currently notice that the /sdX letter changes?

Not typically. As Khalid pointed out, if the drive is released
correctly, the system notices and removes the node (say sde). The new
drive is assigned correctly as sde. I prefer not to trust this though
as it is coincidence. If I were to plug in a USB disk in the middle of
that for some reason, it would become sde instead, and my backup drive
would become sdf. Or if I was missing a disk from my RAID array at
boot my backup drive would be sdd. I can't trust a variably assigned
name.

>  - all the sata ports should be enumerated at boot, drive or not, I'd
> expect. This isn't your experience?

It doesn't seem to, it seems to enumerate devices rather than ports.
There are six sata ports, and if I hot-plug the drive into port 6
instead of 5, it still becomes the lowest name (sde).

> - perhaps it doesn't change if there is always a drive in it at boot time?
> - can you put a file, say, .thisisabackupdrive, at the root of each such
> drive, and your backup script can look for the file? (This would prevent
> hosing a non-backup drive.)

That is a possibility. I was thinking of keeping a copy of the backup
log on each drive anyway. But then I would need to mount every
connected device and filesystem to probe for that file, which I don't
really want to do. I want to mount, backup, umount, all in one swift
move.

> I could see the advantage to duplicate UUIDs. Accidentally put two in at the
> same time, and all sorts of flags should fire off, warning you. Not
> necessarily a bad thing. (-:

Actually, I didn't test what would happen if I have two devices with
the same UUID connected and I try to mount based on that. I would
imagine it would mount the first-probed device, but I'm not sure. I
suppose I'll have to test.

-- 
Chris Irwin
<chris at chrisirwin.ca>