Stewart Stremler wrote:
So... you just give up, and let the drive die, so you can pull the one that isn't spinning?
How can I avoid drives dying if the drive is going to die? Usually they give no warning. Sometimes smartd can clue you in ahead of time but most often not. So yeah, we let the drive die and pull the failed one. What choice is there? I'm letting a drive die right now in fact. I don't know which one. Or when it will die. But I guarantee you I have one dying out there somewhere. What do you suggest I do? RAID/LVM report the failure but cover up the consequences and take care of re-establishing order when I put the new drive in.
It's important to beat engineers about the head and shoulders for recommending excessively complex solutions that mostly demonstrate how clever they are.
Someday you may have to admin a thousand spindles like I often do. At that point you will not be able to just keep track of which filesystem is on which disk. You will need a more complex solution.
It's all about the replacable unit -- I don't care what platter my data is on, I'll just replace the whole disk. That's the appropriate level of abstraction.
Er...isn't that what I just said? I don't care what platter my data is on. I do know that *this* partition is on *that* array. Sometimes those arrays contain a hundred disks, but I still know. And that is Good Enough.
I've watched people who play with the mega-massive disk arrays that function on a "higher level of abstraction", and it's very impressive, amazingly flexible, and way cool... until they lose a weekend rebuilding the whole damn thing because something went haywire and now there's no way of determining what data is where (all that data is on a disk, but which one, nobody knows...).
I have never spent more than an hour or two waiting for a rebuild. Why does it matter which disk a piece of data is on? Why would anyone care? Just knowing which array it is in has always been good enough.
Granted, when you start talking about petabytes of data, there's going to be a disk management problem. But that's not where most folks are.
In 2000 a petabyte required a hundred spindles and was a disk management problem without RAID/LVM. Nowadays people still have hundreds of spindles but they have hundreds of terabytes of video or whatever. It's not where most folks are, true. Those folks can have one or two spindles in their machine and always know where their data is. But as a system administrator I often see situations where it goes way beyond that and a greater complexity is needed.
-- [email protected] http://www.kernel-panic.org/cgi-bin/mailman/listinfo/kplug-list
