zfs discuss - May 2008 - Deletion of file from ZFS Disk and Snapshots

This is a common problem that we run into and perhaps there''s a good
explanation of why it can''t be done. Often, there will be a large set
of data, say 200GB or so that gets written to a ZFS share, snapshotted and then
deleted a few days later. As I''m sure you know, none of the space is
returned to the pool since the bits on disk are still being referenced by the
snapshot. Is there any way to delete a large set of data both from the disk and
from any snapshots that may reference this large data set?
 
 
This message posted from opensolaris.org

>From my understanding, when you delete all the snapshots that reference the
files that have already been deleted from the file system(s), then all the space
will be returned to the pool.
So try deleting the snapshots that you no longer need. Obviously, be sure that
you don''t need any files referenced by the snapshots first ;-)
 
 
This message posted from opensolaris.org

See, that''s the problem. I don''t want to delete the entire
snapshot, I just want to remove a particular file from the snapshot.
 
 
This message posted from opensolaris.org

OK, I understand.

I don''t think it''s possible though, as snapshots are read
only, as far as I know.
If you like voodoo you could try going into the .zfs directory, then into the
snapshot directory and see what you can do there, but I wouldn''t try
it, as it might be possible to do harm that way.

Instead, perhaps you could:

1. do a full backup from the initial snapshot, then do incremental backups from
the subsequent snapshots, if they are any. This should give you a file system
containing all of the files in the current snapshots.

2. Then you could delete the file you want to remove, but this will cause some
snapshots to take ownership of this deleted file.

Here is an example of how to make a backup of all the data from the snapshots to
a new file system. Once done, you can see about deleting the file:

zfs send pool/fs at 1 | zfs recv pool/newfs  (full initial backup)
zfs send -i pool/fs at 1 pool/fs at 2 | zfs recv -F pool/newfs (inc. backup
between snaps 1 and 2)
zfs send -i pool/fs at 2 pool/fs at 3 | zfs recv -F pool/newfs (inc. backup
between snaps 2 and 3)
zfs send -i pool/fs at 3 pool/fs at 4 | zfs recv -F pool/newfs (inc. backup
between snaps 3 and 4)
etc for each snapshot until you have done all the snapshots

rm /pool/newfs/file_you_want_to_delete

I think there''s probably a quicker way using a snapshot range, but I
didn''t try that yet.

I actually wrote up some stuff on snapshots and full & incremental backups
yesterday. |f it''s useful for you, I don''t know :)

See: 
http://breden.org.uk/2008/05/12/home-fileserver-zfs-snapshots/
http://breden.org.uk/2008/05/12/home-fileserver-backups-from-zfs-snapshots/

Good luck, and let us know if you manage to make it work.
 
 
This message posted from opensolaris.org

Simon Breden wrote:
> OK, I understand.
> 
> I don''t think it''s possible though, as snapshots are read
only, as far as I know.
> If you like voodoo you could try going into the .zfs directory, then into
the snapshot directory and see what you can do there, but I wouldn''t
try it, as it might be possible to do harm that way.
There is no voodoo you could do and you can''t do any harm.

You can create, rename, delete and view snapshots via the .zfs/ 
directory but not do any harm to ZFS.

-- 
Darren J Moffat

Thanks Darren, that''s good to know.

If multiple snapshots reference (own?) the same file, what''s the
quickest
way to zap that file from all snapshots?
I say ''own'' because apparently the original file system that
the snapshots
were created from, has been deleted, apparently, if I understood the initial
post correctly.

2008/5/13 Darren J Moffat <darrenm at opensolaris.org>:
> Simon Breden wrote:
>
> > OK, I understand.
> >
> > I don''t think it''s possible though, as snapshots are
read only, as far
> > as I know.
> > If you like voodoo you could try going into the .zfs directory, then
> > into the snapshot directory and see what you can do there, but I
wouldn''t
> > try it, as it might be possible to do harm that way.
> >
>
> There is no voodoo you could do and you can''t do any harm.
>
> You can create, rename, delete and view snapshots via the .zfs/ directory
> but not do any harm to ZFS.
>
> --
> Darren J Moffat
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20080513/5dc21eea/attachment.html>

Can''t you turn the snapshot into a clone (kind of an editable
snapshot)? Or does the existence of a clone created from this snapshot prevent
from removing the snapshot afterwards?
 
 
This message posted from opensolaris.org

On Tue, May 13, 2008 at 10:02:01AM -0700, Marc Glisse
wrote:
> Can''t you turn the snapshot into a clone (kind of an editable
> snapshot)? Or does the existence of a clone created from this snapshot
> prevent from removing the snapshot afterwards?
You can create a clone from the snapshot, but it does not turn the
snapshot into a clone.

Any clone created and the original filesystem are tied together through
the snapshot.  The snapshot can''t be destroyed as long as both exist.
You can pick either one of them to be the parent, allowing you to delete
the other and the snapshot.

-- 
Darren

On Tue, May 13, 2008 at 04:33:29PM +0200, Simon Breden
wrote:
> Thanks Darren, that''s good to know.
> 
> If multiple snapshots reference (own?) the same file, what''s the
quickest
> way to zap that file from all snapshots?
There isn''t a way to do that at all, not short of deleting the actual
snapshots (which you''ve already said you don''t want to do).

I doubt there ever will be a way to do that (though that''s just an
opinion informed by public information -- I don''t work for the ZFS
team).

Personally, I don''t think there should be a way to do that.  If
you''re
running out of space, then add more or delete snapshots that you no
longer need (backing them up if need be).

Nico
--

On Tue, May 13, 2008 at 04:33:29PM +0200, Simon Breden
wrote:
> 
> If multiple snapshots reference (own?) the same file, what''s the
quickest
> way to zap that file from all snapshots?
There is no way.

If you could do that, then they wouldn''t really be
"snapshots".

I''m not saying that the ability wouldn''t be useful, but it
would
conflict with certain expectations about what a "snapshot" is today.

I had a hard enough time dealing with the ability to create and destroy
sanpshots via normal filesystem semantics.  The ability to modify a
snapshot would throw me for a loop!

-- 
Darren

Thanks for the info folks, it''s what I expected but I just wanted to
verify that this is not possible.
 
 
This message posted from opensolaris.org