Is there room for improvement in rescue mode? (was Re: Goodbye, Fedora)

[Bill Rugolsky Jr.]

On Thu, Feb 22, 2007 at 02:20:10PM -0500, Jeremy Katz wrote:
> If someone were to start looking at it, we could probably even get some
> simple stuff done for F7 since we're not talking about huge amounts of
> stuff.  Also, once we have the basics, we can think about things like
> integrating the functionality into the live CD as well.

The last time this came up, IIRC, was Jeff Layton's mkinitrd rescue mode
patch.  I haven't looked at anaconda since then, so I have no idea where
things were taken for FC6+.

My firm is currently rolling our own initramfs that we use for PXE
installs, as well as providing a ramfs-based rescue mode.  I've encouraged
my colleague to package this as a mkinitrd replacement that might be more
generally useful.

As I mentioned last go-round, GRUB (and perhaps the other boot loaders)
can load multiple initramfs payloads.  That means that one can have
a small standard *kernel-independent* payload, a kernel-dependent payload
(modules, probably udev at the rate we are going :-|), 
perhaps an install-specific image (e.g., localization info -- fonts,
keyboard maps), and a large, static rescue payload.

With suitable hooks in the standard payloads to test for the existence
of various files (just like /fastboot, /forcefsck, etc.), task-specific 
initramfs payloads could be dropped into /boot and do their bit automatically.
E.g., one of the things that we are doing is failing out and doing an
upgrade on half of a RAID1; if the upgrade fails, we boot back into the old
image.  Another oft-requested task is running parted to move partitions
around.

Regards,

	Bill Rugolsky