[Dirvish] Dirvish/rsync thinking partition is 'new'

[Paul Slootman]

On Sat 05 Mar 2011, Dale Amon wrote:
> On Sun, Mar 06, 2011 at 01:35:44AM +0100, Paul Slootman wrote:
> > 
> > The owner is (has become?) different. Hence dirvish needs to retransmit
> > those files as the metadata is different.
> 
> But if the contents has not changed, the actual transmittal
> should be nothing by a few bytes of directory data here
> and there. In this case, I would expect (hope!) that the
> update would run very fast. 

Very true.

Unfortunately it seems that dirvish only passes --link-dest and not also
--compare-dest to rsync.  I might look into seeing if this can be fixed
easily.

> I do see this as a problem though... does he end up with
> a (locally copied) version of the file instead of a hard 
> link? Seems a bit on the assinine side to duplicate say, an
> unchanged 100GB iso image, just because the user 'touch'd, 
> chmod'ed or chown'ed it. Really a rather serious waste of
> resources.

Dirvish's mission is to keep a completely correct backup of the source
tree. If attributes have changed, they may have been changed because
things didn't work correctly before, so it's important to have the
latest image reflect those changes. It's also possible that the changes
were in error, hence it's also important to preserve the previous image,
so simply hard-linking the files and updating the owner is wrong.
Unfortunately the only correct solution is therefore to create a new
copy of all those files.

I don't know whether the COW implementation on top of btrfs might help
in that respect.


Paul