[CentOS-devel] Timeline for C5 beta / final?

Mon Feb 19 10:51:00 UTC 2007
Nils Toedtmann <centos-mail at nils.toedtmann.net>

Am Montag, den 19.02.2007, 04:42 -0200 schrieb Rodrigo Barbosa:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On Sun, Feb 18, 2007 at 07:14:16PM -0600, Les Mikesell wrote:
> > Rodrigo Barbosa wrote:
> > 
> > >>As a safety mesurement you could clone the machine ("cp -a") to a
> > >
> > >"cp -a" for cloning ? Errr, bad mistake.
> > >
> > >"dump -0f - / | ( cd /newlocation; restore -f - )" is a much better idea 
> > >on these cases.
> > >
> > 
> > The results should be pretty much the same on anything with GNU cp 
> > except that cp can copy to different filesystem types.  It is a good 
> > idea to use the --one-file-system option and repeat for each partition.
> 
> Are you sure ? Last I remember, "cp -a" does not preserve some of the inode
> data. Also, it does some file reordering.
> 
> Yes, most of these are inconsequential on a practical pov, but it is
> still farther from cloning than using dump/restore.
> 
> Of course, if you want real cloning you would have to use "dd", but
> that is _way_ beside the point here.

My mistake, i should have provided my definition of "cloning an
OS" (not: "cloning a partition"): "making a copy of the OS that is
similar enough to its original that it works the same way with no
perceivable difference". 

"cp -a[x]" does exactly that (at least CentOS' GNU cp; it preserves
labels like selinux contexts, too) provided that your apps do not
operate directly on the blockdevs ("hexedit /dev/sda1"). I consider it
being superior to dd to do this job as it "defrags" the filesystems. I
copied several systems and never observed any breakage due to cp
forgetting or changing low-level details.

That may have been false back in '91, but things change ;)

/nils.