Bug#401916: initramfs-tools: [powerpc64] i-t tries to mount RAID/LVM stuff before the disk are up -> unbootable
On Sun, Dec 10, 2006 at 10:39:45AM +0100, maximilian attems wrote:
> On Wed, 06 Dec 2006, Sven Luther wrote:
>
> > Package: initramfs-tools
> > Severity: critical
> > Justification: breaks the whole system
>
> hmm yes i know of that situation it affects a certain range of roots.
>
>
> > Today i went to the datacenter to reboot the xserve G5 i have there, for some
> > random reason, and what was not my surprise to notice that the box didn't come
> > up anymore.
> >
> > After some dmesg examination, i noticed that the sata disks did come up only
> > after i-t tried to bring up the RAID and LVM stuff, which is really not nice.
>
> the trouble is that udevsellte exists to early that mean when
> the scsi/usb discs are not up yet.
> hitting raid/lvm2 roots on those devices and more general lilo boots
> as there you have no root dev to wait for.
> i notified udev upstream for it, but got no response i'll reask privately.
> http://marc.theaimsgroup.com/?l=linux-scsi&m=116189244404693&w=2
Yeah, please do, we really need to fix this before etch is out, i don't think
that having a server who cannot reboot sometimes is something we want to ship
in etch.
> > Furthermore, playing offline without any info with the box, i was not able to
> > convince i-t to mount the partition and investigate, but well, i guess i would
> > have been able to do so if i had the code available, or more time to
> > investigate.
>
> you should have rerun the early initramfs-tools stages, see init.
> than it works.
Ah, that was the missing info. I searched a bit, but didn't find easily what
to do, and then i had to leave. I will be at the box on thursday again.
Now, as a temporary workaround, maybe one solution, if RAID/LVM was not found
is to delay for a given time (1/2 minute ? a configurable time ?), and then
try again.
Another nice thing would be able to rerun the early initramfs-tools stages, or
maybe even have a stamp of each completed stage, and a single command which
allow to restart the detection from there ?
Friendly,
Sven Luther
Reply to: