[Dirvish] move/rename files/folders

[Paul Slootman]

On Fri 22 Jun 2007, Eric Wilhelm wrote:
> # from Bernd Haug
> # on Friday 22 June 2007 02:12 am:
> 
> >But if you move files in the backed up directory around, I don't
> >think that rsync goes and checks whether this new name has the same
> >inode as a file currently in backup had when it was on the original
> >FS (where should it get that data in the first place?).
> 
> Yep.  The dirvish scheme is just:
> 
>   1. copy old backup tree to new backup tree, hardlinking everything
>   2. rsync into that

Actually, dirvish does not copy the old backup tree to the new backup
tree, it lets rsync do the heavy lifting there via the --link-dest
option (rsync links unchanged files when it encounters them, it's not a
2-part action).

> Step 2 only looks at filenames, not inodes.  This breaks the links 
> whenever a file has changed and deletes any files that are missing.  An 
> mv on the source (from rsync's pov) equates to delete+create.

rsync now has a --fuzzy option that might help:

    The current algorithm looks in the same directory as the destination
    file for either a file that has  an identical  size and
    modified-time, or a similarly-named file.  If found, rsync uses the
    fuzzy basis file to try to speed up the transfer.

So renaming a file "foo" to "foo.1" might preserve the link (I've never
tested how well the --fuzzy option finds renamed files).

Preserving hardlinks in all cases isn't really an option.
If you're doing a renaming on the source that will have a large impact
on the backup, it's probably advisable to do the same in the latest
backup tree. In other cases, just accept that you lose hardlinks...


Paul Slootman