zfs discuss - Dec 2005 - Probability of future compatibility with present-day pools?

How likely is it that a pool I create today (say, with build 28) will be
importable (and fully feature-compatible) with the eventual
''production'' implementation of ZFS?

Anyone uninterested in my ramblings can pretty much skip the rest of this post.

I''ve been putting off building a home NAS server for a few months now,
waiting for a few things, such as the perfect silent SFF case with 4 3.5"
bays and the horsepower to hit 50M-60MB/s (I''ve pretty given up on
finding such a case / motherboard, though
[url=http://www.mashie.org/casemods/udat1.html]this[/url] is gorgeous, and if I
could buy one with a decent CPU, I''d pay dearly - [i]dearly[/i], but
now I''m getting off topic).

The other thing I''ve been waiting for, to get back on track, is ZFS,
which is finally here.  I''d preferred to wait until things shook out a
little more (I''ve seen my share of ZFS-related panics tooling around at
work with 27a on a SB1000 and an old multipack with its SCSI menagerie of 4GB,
8GB, and 16GB relics), but I''ve recently had various disks around the
house begin to fail, so I need to get started now.

If there were a really good chance that a ZFS pool I build today will use the
same data/metadata format (i.e. be fully usable) in ''official''
Solaris releases down the road, I think I''d take the risk of riding out
SE bugs for the time being, as my other option would be to go Linux, and then
I''d be pretty much stuck with that.  I don''t expect to be able
to do a wholesale array migration at any point in the forseeable future, so what
I build now is very likely what I''ll have for awhile, and I''d
much rather that be Solaris / ZFS.

Thanks for the awesome work, guys.
This message posted from opensolaris.org

At the moment, I know of no on-disk format changes, but
they are possible. We did leave quite a lot of padding fields in each
meta data structure to allow for unknown various future changes.
You can see this in the source code, or from a soon-to-be-released draft
on-disk format document. We can also use versioning, but this gets ugly
after a while, with code conditional on various versions.

As far your ZFS crashes, please tell us about them.

Thanks: Neil.
This message posted from opensolaris.org