[GTALUG] Disks are made of fragilium (was: GnuPG Woes...)

Anthony de Boer adb at adb.ca
Thu Jan 1 16:41:41 UTC 2015


David Collier-Brown wrote:
> On 12/30/2014 04:53 PM, Peter King wrote:
> > Moral of the Story (one moral among many): Keep static time-stamped 
> > backups as well as current redundant copies. Will implement a scheme 
> > to do so this week, a better New Year's resolution than most!
> 
> This is way more common than people expect.
> 
> The ZFS folks found out that they needed to checksum disk files, 
> somewhat to their surprise. ICL used to do it on a sector-by-sector 
> basis as part of the sector footer, and would noisily recover when they 
> detected problems.  Sun 3.5 did too, but you had to do the recovery by 
> booting (!) /stand/diag

I've had ZFS, and recently now also BTRFS, tell me that it was getting
checksum errors from a disk.  In the one case it seemed like a one-off
glitch, but the other seems to be hardware that's failing by flailing,
sometimes returning whatever crap it thinks it might have seen, and in
any event its problems are widespread and that disk's being replaced
and e-wasted.

Defensively, either of those filesystems are a better thing than the
callow trusting filesystems of yesteryear.  Subtle problems you don't
notice til long later waste far too much of one's time.

At the application level, one thing that can help is keeping important
files in git (and pushing a copy to another computer!) so that you
have copies and history of revisions and it keeps checksums and will
squawk at corruption.

Another is cold-storage backups that don't overwrite existing files,
just add new ones, for things like media and other archival files
that shouldn't ever change after their initial creation.

-- 
Anthony de Boer


More information about the talk mailing list