What To Do When A Hard Drive Fails
server.dzone.com
server.dzone.com
In the event of an actual head crash, you're usually screwed. If the head crash left a stripe all the way around the platter, the head likely won't seek past it. (I don't know why; it should be able to, but it doesn't -- I tested this recently on a Toshiba drive.)
Head crashes seem to have a tendency to occur close to the spindle, and if that happens, the data on the drive is completely irretrievable.
Occasionally the armature will develop a mechanical fault and will collide with the edge of the platters when the drive powers up. If this happens, you can usually either "adjust" the armature slightly, or replace it with an armature from a matching donor drive.
There are a few other tricks and things too, but those are the most common for internal failures. Sometimes just replacing the logic board is all that's needed -- you have to be able to diagnose what's wrong with the drive. Simply sticking it in the freezer isn't likely to do a darned thing for you.
And I'm a little skeptical that this guy has ever even cracked a drive open. They don't have a "bunch" of platters, they almost always have 2.
http://www.xbitlabs.com/articles/storage/display/1tb-hdd-rou...
On one occasion, my friend's laptop hard drive kicked the bucket and failed when I tried to use Spinrite.
However, a few weeks ago, a Aeagate hard drive from my brother's Macbook failed. Throwing it in the freezer for a few hours did the trick and we were able to make a full copy of the data.
1. Use ddrescue (http://www.gnu.org/software/ddrescue/ddrescue.html) to copy a complete image of your hard drive onto another drive with lots of space. Let it work for a long time and it will probably get all the data off. If it doesn't you can try the freezer trick, but don't be too hopeful.
2. Unplug the busted hard drive and set it aside. You can perform all the subsequent recovery steps on the image you just copied, which is better than stressing an already-dying drive.
3. Mount the hard drive image (http://tcdb.grinnell.edu/wiki/pmwiki.php?n=Help.Ddrescue).
4. Use testdisk (http://www.cgsecurity.org/wiki/TestDisk) to try to recreate the partition table.
4.5 Alternatively, you can use parted with the rescue command, but that works better if you know exactly what your partition table was.
5. If that fails, use photorec (http://www.cgsecurity.org/wiki/PhotoRec) to pull off all the files it can find. It should be able to get most of your documents, photos, etc, but it might not find everything, and they will be missing file names and folder structure. Remind yourself that this is better than nothing.
I agree with pretty much everyone here, though, in saying that "restore it from backup" is the right answer, and the fact that the parent-linked article doesn't mention it anywhere makes it worthy of ridicule. Fortunately, the second commenter on the article injects a bit of sanity into the situation.
1.5 If the hard drive is so badly failed that you cannot even get ddrescue running, try again tomorrow. And the next day. And the next. I cannot explain why a hard drive would come back to life, but I have seen it happen a number of times. I have seen a hard drive that was clicking and irrecoverable for two solid weeks suddenly show up just fine one day in the device list. It mounted and everything, and we promptly pulled all the data off. I'm not much of an optimist, but when it comes to failed hard drives, I've seen miracles occur.
It doesn't always work. When it does work, you'll get to read data off it just once after the freeze. Sometimes a refreeze allows you to read data again, but never more than that.
Not all RAID-1 has that particular limit. I work with both hardware- and software-based RAID-1 implementations that work just fine across multiple hosts, and that can operate across separations of hundreds of kilometers.
And while you're correct about RAID-1 not being a backup; that a volume corruption, errant delete or sufficient degrees of user disgruntlement can still nuke the data. That written, the software RAID-1 solution I use can be used as a component of a backup procedure; you can pull disks out of the RAIDset, clone or archive, and then merge the volumes back into the RAIDset using a delta of the changes.
FWIW.
edit: added a missing "; that"
The difference being you can get a consistent copy of the whole disk with a very small window where the applications must be quiesced, where the cloning process you've suggested rolls through the whole volume and (with an active volume) the volume state tends to change as the copy process traverses the volume.
As for an alternative and I/O-heavy approach that's possible with the RAID1 implementation, it's also feasible to roll out a member volume and roll in a scratch disk; to replace the volumes using a more traditional scheme. Rolling the member volume back into the RAID1 set after the cloning operation can use the delta of the differences, rather than necessitating a whole-volume copy. Replacing a disk necessitates a whole-disk copy.
If you're working with a fairly quiescent disk, then the difference between these two approaches probably isn't significant. If the volumes are busy, then having a complete and consistent copy of the whole volume with a very short window when the applications have to flush and hold can be advantageous.
Plenty of commenters on HN and dzone are claiming that RAID-1 will provide sufficient protection. Unfortunately, RAID-1 doesn't guard against data corruption or accidental destruction of data. I highly recommend something like Backblaze ($5/mo, unlimited storage, fairly hacker-friendly).
A 70°F (~21°C) room will have a dew point far above the 0°F freezer, so your drive should develop substantial frost from condensation. Unless your relative humidity is around 5%. (Does a glass of ice water drip from condensation? Your drive will be far colder.)
Subjecting data that you care about to potential thermal shock and condensation, well, that doesn't sound like a good idea to me.
A professional repair costs a minimum of $600, so if you know from the beginning that your data is not worth that much, then just go ahead and try it.
Do you have a need to use Linux? I am planning to go OpenSolaris on my next media box.
Of course RAID is not a backup strategy. RAID 1 is great for exactly the mentioned event: when a hard drive fails. If some of your data has been corrupted or lost, you use your backups.