ZFS, from hell's heart I stab at thee!

So, I used to be a "storage architect" at $job[-1], which was my last
sysadmin gig at current employer (I'm currently nominally a Fortran
wrangler).

I say "storage architect" because it basically came down to "how do we
set up a $2k/TB system for $700/TB?" The system, as usual, looks good
on paper: LUNs exported from former NAS boxen to Solaris box via
iSCSI, with ZFS providing the FS and storage virtualization services.

So, ZFS is a nice system. 256-bit checksums of data, all sorts of
"data safety" features. What this system is sorely lacking is
metadata safety.

At any point where the metadata might look a little green, the pool is
marked failed, and you're pointed towards a URL telling you to flatten
and re-install from backups. The ZFS devs say they don't have a fsck
program because fsck's can only work on "known failure pathologies",
which their WunderFS lacks.

Except for, you know, when it needs to try and fix its metadata.

I can assume all of you know where this is going. And I can assume
you all know how happy we are currently with entrusting 45 TB of data
with profs too cheap to have proper backups to ZFS.

However, I'm currently getting a wonderful crash-course in ZFS
internals.

--
Matt Erickson <peawee@xxxxxxxxxx>
BASIC is the Computer Science equivalent of `Scientific Creationism'.
.