On Thu, 21 Sep 2006, Gene Heskett wrote:
> On Thursday 21 September 2006 05:09, Geert Uytterhoeven wrote:
> >On Wed, 20 Sep 2006, Paul Bijnens wrote:
> >> On 2006-09-20 11:56, Geert Uytterhoeven wrote:
> >> > On Mon, 11 Sep 2006, Geert Uytterhoeven wrote:
> >> > > On Sat, 9 Sep 2006, Josef Wolf wrote:
> >> > > > On Tue, Sep 05, 2006 at 03:34:42PM -0400, Jean-Louis Martineau 
> wrote:
> >> > > > >    * Works with GNU tar 1.15.91 - work with new gtar state file
> >> > > > > format.
> >> > > >
> >> > > > Can someone please explain what this exactly means?
> >> > >
> >> > > The format to store information about the incrementals was changed.
> >> > > Since Amanda made some assumptions about this format (while she
> >> > > shouldn't have cared,
> >> > > and just considered it as opaque files), this broke Amanda.
> >> > > After the fix, Amanda just treats the files as opaque files.
> >> > >
> >> > > But be careful, at least the tar 1.15.91-2 from Debian is broken:
> >> > > it ignores
> >> > > the --one-file-system option when doing incrementals, causing
> >> > > exorbitant backup
> >> > > sizes for any level > 0. I don't know about the upstream version,
> >> > > but since
> >> > > this bug has been reported almost 2 months ago, I'm afraid that one
> >> > > is broken,
> >> > > too.
> >> >
> >> > Apparently the problem is more subtle. Thanks to the Debian bug
> >> > tracking system, I noticed this:
> >> >
> >> > http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=384508
> >> > tar: -l option changed meaning, without any warning!
> >>
> >> OK. But AFAIK (grep *.c in the sources), Amanda does NOT use the
> >> -l option, but only the --one-file-system option, since a very long
> >> time already.
> >>
> >> So I think this option name change has nothing to do with the
> >> use of gnutar by Amanda.  (AFAIK the format of the incremental-state
> >> files has changed, and Amanda assumed they were in some line-oriented
> >> format instead of handling it as opaque objects.)
> >
> >Indeed, thanks for reminding me! I just send a clarification to the
> > Debian BTS: - 384508 is about -l no longer meaning --one-file-system
> >  - 377124 is about --one-file-system breaking when combined with
> >    --listed-incremental (Amanda does pass --one-file-system (not -l) to
> >    tar)
> >
> And how does this breakage manifest itself again?  Is it by not following 
> and counting out-of-filesystem links in the estimate phase, but including 
> them during the backup?  This would of course result in "small estimate" 
> notations.

I noticed 2 things when doing non-level-zero backups:
  1. Warnings about weird files in /proc, while tar shouldn't have entered
     /proc as it's a different file system
  2. Backups being way too large, as tar escaped from the file system it was
     backing up.

Gr{oetje,eeting}s,

                                                Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert AT linux-m68k 
DOT org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                                            -- Linus Torvalds