On Mon, Mar 27, 2017 at 05:20:25PM +0200, karl@??? wrote:
> Hendrik Boom:
> ...
> > lilo doesn't do RAID assemby, as far as I know. It just starts up
> > with a bunch of blocks at a fixed offset from an identified partition,
> > identified by UUID, and there's two of these. Now I guess at boot
> > time, it may always pick the right one, or always pick the wrong one,
> > or a random one.
>
> If you have /boot on a raid-mirror on all your disks (pref. as first
> partition) there is no problem making lilo work. Since then it
> doesn't matter which disk it is booting from, it will load the kernel
> and possible an initrd/initramfs if you want it to.

Except that the raid-mirror has been missing for some time, and the
two copies of /boot are no longer the same. Reattaching it would
yield two very similar file systems and lilo will thus have a choice
between two different places to boot from. If the two copies of the
RAID were properly sync'd up you would be right -- there would be no
problem.

>
> Since /boot is typically very small, there is not any big problem
> making this happen.

Which is why I put /boot on a small mirror RAID, not realising that
for reasons beyond my control, one drive would become disconnected
for far too long to remain synchronised.

>
> ///
>
> If you have theese set in your kernel:
>
> CONFIG_MD_AUTODETECT=y
> CONFIG_MD_RAID1=y (or whatever raid level you want)
>
> the kernel will autodetect your old style (metadata 0.90) raid
> partitions. So if your / is on such a raid, the kernel will mount
> / without the need of an initrd/initramfs.
>
> The old metatdata limits you to one filesystem per md-raid device.

Indeed, I have one RAID with old metadata with only /boot on it.
Another with new metadata containing many LVM partitions.

>
> Regards,
> /Karl Hammar

Thank you.

-- hendrik