On Fri, 2006-05-05 at 15:34 -0500, Aaron Konstam wrote:
On Fri, 2006-05-05 at 14:56 -0400, Mauriat Miranda wrote:
> On 5/5/06, Les Mikesell <lesmikesell(a)gmail.com> wrote:
> > > I question if this necessary. You can have entries in fstab with or
> > > without labels. I've had 3 single Fedora installations on single
drive
> > > across multiple partitions and never have run into a problem in the
> > > installer or usage nor have I had to manually edit fstab for this. I
> > > think in Anaconda the labels will start shifting to /1, /2, /home1
> > > etc. Since at runtime you don't deal with partition labels, just
their
> > > mount points, it really is not a serious concern.
> >
> > Assume you are in the IT dept for some group and you are used
> > to being able to re-use disks in different machines and to
> > recover data from any disk by installing/mounting in any working
> > machine. Now you find that any combination of disks from
> > default fedora/RH/Centos installs won't boot... It is a
> > problem.
>
> So you boot with a LiveCD or some Rescue disk/media and fix it - use
> the device ID or relabel it. Isn't that the way to fix it? I don't
> understand the problem. Your assumed scenario is too vague.
>
> -Mauriat
We are tying to produce a distribution, are we not, that the uninitiated
can install and it will just work. A lot of things can be fixed but we
want to avoid that.
The uninitiated will not likely be moving drives between machines. They
would do an install on already existing drives and as such the Fedora
install/labeling scheme works well. The conflicts occur when previously
labeled drives are mixed in a machine as mentioned above.
I agree that it needs to be more capable of recovering but the problem
conditions do not appear to be with the new/uninitiated but rather with
the users who are moving hardware between machines after the
installation has completed for whatever reason.
Some OSes write a PVID on the physical device that is unique (similar to
the way identifiers on LVM logical volumes and volume groups are
unique). This may be a better way since a unique identifier of this
sort (physical volume plus logical volume/partition) is guaranteed to
not conflict the way the current labels do.