preventing deadlocks in snapshot directories - unexplained

[Eric Anderson]

Oliver Fromme wrote:
> user <user at dhp.com> wrote:
>  > I have asked this before, but nobody answered ...
>
> Many people tend to be reluctant answering questions
> from anonymous users (or they don't bother reading their
> questions at all).
>
>  > If you have multiple snapshots, how do you segregate them in order to
>  > avoid the deadlocks that the ".snap" directory is supposed to fix ?
>
> As far as I know, using that directory is just a matter of
> convention.  You can place snapshots anywhere else, and it
> doesn't make a difference.  The only thing special about
> ".snap" is that it is created by default by newfs(1), and
> it is expected to exist by the snapshot support of the
> dump(8) and fsck(8) tools (for dumping live file systems
> and background fsck, respectively).
>
>  > I understand why a snapshot is created in (mount)/.snap - but what if I
>  > have multiple snapshots running simultaneously ?
>
> It doesn't make a difference.
>
>  > cd /.snap
>  > rm -rf snap3
>  > 
>  > the _entire_ /.snap directory locks up until that command completes.
>
> Well, yes, certain file system operations are blocked
> until the removal of the snapshot is complete, which
> can take quite some time, depending on the size of the
> file system, because a lot of data has to be updated.
> So, that blocking is to be expected (unfortunately).
>
>  > So, this leads me to conclude that actually, I need to do this:
>  > 
>  > /.snap/snapshot_file
>  > /.snap2/snapshot_file
>  > /.snap3/snapshot_file
>
> No, it doesn't make a difference.  As far as I know, the
> snapshot code doesn't care about the directory name where
> the snapshot files are placed.
>   

Well, its ok to put snapshots anywhere, but when operations on a 
snapshot file are in progress, stat calls on the snapshot file will 
block.  I think this is his problem.  I wonder how painful it would be 
to have stat's on snapshot files return with (something) when a snapshot 
operation is in progress, instead of blocking, or even if that is possible.


Eric




-- 
------------------------------------------------------------------------
Eric Anderson        Sr. Systems Administrator        Centaur Technology
Anything that works is better than anything that doesn't.
------------------------------------------------------------------------