[rdiff-backup-users] Atomicity of backups?

[Nathaniel Smith]

I've just discovered rdiff-backup, and it looks _very_ cool; just what
I've been looking for.  I have a worry, though, that I don't see
addressed in the website or mailing list archives.  What I want to
know is, what happens if my disk crashes half way through a backup?

rdiff-backup seems to modify the entire mirror directory during the
process of the backup, and if the mirror is corrupted then the diffs
it keeps may become useless.  Modifying the mirror directory is
presumably a lengthy operation; my worry is that if the disk I'm
backing up picks some time in the middle of that to go south, then
I'll be left with no disk, and an inconsistent mirror directory...
which would then prevent me using the diffs to get at the previous
days backup.  Which would, worst case, leave me with no accessible
data at all.

Does rdiff-backup have any provisions to prevent this happening?
Conceptually, it seems like you need something like journalled
commits, where after a crash it's always possible to revert to the
last-good-version.  Even just making a full copy of the mirror
directory before modifying it would be a good simple solution, albeit
somewhat hard on the disk space :-).  What does it do now?

Thanks,
-- Nathaniel

-- 
Eternity is very long, especially towards the end.
  -- Woody Allen

This email may be read aloud.