zfs discuss - Aug 2006 - Find the difference between two snapshots

Hi everyone,

Is there an easy way to find out which files has changed between two
snapshots? Currently I''m  doing a

# rsync -arvn <snapshot1> <snapshot2>

and it creates a list. But rsync needs to go through the whole
filesystem and compare files. It would be nice if zfs would have this
option builtin.

Regards,
Nickus

On 8/31/06, Tim Foster <Tim.Foster at sun.com>
wrote:
> Hi Nickus,
>
> On Thu, 2006-08-31 at 15:44 +0300, Niclas Sodergard wrote:
> > Is there an easy way to find out which files has changed between two
> > snapshots? Currently I''m  doing a
> >
> > # rsync -arvn <snapshot1> <snapshot2>
> >
> > and it creates a list. But rsync needs to go through the whole
> > filesystem and compare files. It would be nice if zfs would have this
> > option builtin.
>
> Nope, unfortunately not - you''re interested in bug
> http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6425091
>
>  - it''d be pretty handy for the zfs desktop integration stuff I
was
> playing with too...
Thanks for the quick answer. It looks like the problem is much more
complex than I first thought.

cheers,
Nickus

Hi Nickus,

On Thu, 2006-08-31 at 15:44 +0300, Niclas Sodergard
wrote:
> Is there an easy way to find out which files has changed between two
> snapshots? Currently I''m  doing a
> 
> # rsync -arvn <snapshot1> <snapshot2>
> 
> and it creates a list. But rsync needs to go through the whole
> filesystem and compare files. It would be nice if zfs would have this
> option builtin.
Nope, unfortunately not - you''re interested in bug 
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6425091

 - it''d be pretty handy for the zfs desktop integration stuff I was
playing with too...

	cheers,
			tim
-- 
Tim Foster, Sun Microsystems Inc, Operating Platforms Group
Engineering Operations            http://blogs.sun.com/timf

On Thu, Aug 31, 2006 at 02:55:27PM +0100, Tim Foster
wrote:
> On Thu, 2006-08-31 at 15:44 +0300, Niclas Sodergard wrote:
> > Is there an easy way to find out which files has changed between two
> > snapshots? Currently I''m  doing a
> 
> Nope, unfortunately not - you''re interested in bug 
> http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6425091
See also:

6370738 zfs diffs filesystems

The idea is that SEEK_HOLE[*] could be used to find block-level diffs
for individual files that have changed.  [*] Or SEEK_DIFF, to avoid
aliasing holes and diffs.

Dealing with renames/links/unlinks is... harder.  The idea is that a
diffs filesystem would only show directories and files that have changed
or are part of a path where some directory/file has changed, and all
unchanged things would be hidden.  You''d still have to do a fair bit of
work at the app level to deal with renames/links/unlinks generally.

Partly what is difficult here is that ZFS tracks differences between
snapshots/filesystems at the block level, so you can find out that some
dnode changed, but mapping that dnode to its name(s) in the filesystem,
if it can have hardlinks (i.e., it''s not a directory) is as hard as
it''s
ever been to map inode #s back to paths.

It''d be nice if there was a background dnode->paths indexer that
could
asynchronously maintain such an index (synchronously at snapshot time,
so a snapshot could capture this index as it should be).  Asynchronous
so as not to slow down meta-data operations unnecessarily.  If that''s
too hard, then make it synchronous but let running it be optional
per-filesystem.

Nico
--