[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [rdiff-backup-users] Incremental Restore
From: |
Kosta Welke |
Subject: |
Re: [rdiff-backup-users] Incremental Restore |
Date: |
Tue, 7 May 2013 09:40:49 +0200 |
Hi!
On May 7, 2013, at 24:30 , Edward Ned Harvey wrote:
> As I understand it, at present, "df" doesn't fully understand btrfs raid-1.
I'll see how it works when I get there. I tried to get a better understanding
of this by reading the btrfs FAQ on the wiki and the man pages, but now I'm
only more confused :)
>> Doesn't the `rdiff-backup --force -r $INC /rdiff-backup-dir/
> /tmp/staging/`
>> step do a full checkout?
> Yes. That's why I said I'm sure there's a more space and time efficient
> solution possible... But ...
>
> Trying to parse through the "increments" subdir is tricky. If you get it
> right, you can save some time and space (but you'll probably waste that much
> time figuring out how to get it right, and you'll have higher risk of
> getting something wrong.)
Yeah but those full checkouts take forever. However, the actual reason why I
didn't want to do full checkouts is that I feared that this would not allow
copy-on-write and the btrfs deduplication sounds rather... experimental.
So I think full checkout to temp restore directory && rsync -a --delete is the
way to go here.
> It's also very easy to accidentally forget about the final timestamp, or
> handle it wrongly, because (a) it's not displayed directly by the -l switch,
> and (b) if you "ls" the rdiff-backup-dir, you don't directly see it there
> either.
In any case, after all rsync backups are through, I will do an rsync -a
--delete from the actual directory (NOT the rdiff-backup storage) to the new
disk. But thanks for the heads-up: This means I'll have to do a restore from
`now` after all increments. I wouldn't have done that otherwise.
On May 7, 2013, at 08:38 , Adrian A. Baumann wrote:
> I'd do an "rsync -a --delete /tmp/staging/ /destination" - otherwise, you'll
> end up with all the deleted files restored as well.
Thanks! That's actually the big reason to go for "full restore" of each
increment. I feel that I could easily write that code, but my gut feeling is
that that would reach a complexity where I would screw up somewhere and only
notice after it's too late.
Cheers,
Kosta