[indiana-discuss] image-update from 91 to 100: new BE hangs on boot, old BE panicks on boot

[Haik Aftandilian]

> Haik Aftandilian wrote:
> > Upgraded from build 91 to build 100 last night by
> following the instructions at
> > 
> >
> http://mail.opensolaris.org/pipermail/opensolaris-anno
> unce/2008-July/001860.html
> > 
> > Afterwards, when trying to boot the new build 100
> BE, I see the "Use is subject to license terms" boot
> message and then a message from the NVIDIA driver
> telling me that my card will be ignored because it is
> not supported and that it will keep probing. I do
> have an older card that requires the legacy drivers.
> After that message, the boot hangs.
> > 
> > I tried to boot back into my build 91 BE and the
> system panicks on boot because it fails to mount the
> root pool. The panic message is along the lines of
> "this pool was last accessed by a machine with a
> different hostid".
> > 
> > I've burned a DVD of the 2008.11 build 100a ISO so
> I can boot from it and try and force mount the root
> pool (with the hope that that will fix the build 91
> boot panic).
> > 
> > Is this a known bug? My searches haven't turned up
> anything.
> 
> You need to boot to the b100 environment twice so
> that a hostid can be 
> generated.  I think that will resolve your issue.
> 
> However, you may need to boot into single-user mode.


Booting into single-user didn't help, but I was able to resolve my problem and boot into the build 100 BE. I also found out my root disk is failing.

Booting single-user still hung after the NVIDIA message so I booted from the 2008.11 b100a ISO live CD. The GUI came up correctly using the live CD. However, I used the live CD in text mode and mounted (via zpool import) my root pool. The zpool import caused some disk read errors to be printed to the console "Sense Key: Media Error" and caused the zpool import to take a few minutes. Two disk blocks were listed in the error messages.

A zpool scrub of the root pool didn't find any errors. I assume that the failing disk blocks are touched by the zpool import command, but are not being used by my root pool.

With the root pool mounted and the b100 BE mounted, I used pkgadd -R to install the legacy NVIDIA drivers SVR5 package and was then able to reboot into the build 100 BE. For now, I've set copies=2 on the root pool top level filesystem and will be replacing the root disk soon. It's unknown why the build 100 BE was originally hanging on boot. Since the b100a live CD boots up fine into graphical mode, I'll have to assume it was caused by the root disk failure and not the NVIDIA drivers.

I've also noticed that with 100, my hostid changes every time I reboot. This causes some zpools to fail to mount, but not the root pool. I think the hostid change was probably the cause of the older BE's failure to boot. I'm not going to spend much time debugging this because there are some bugs in this area already fixed in the later 2008.11 build.

Haik


Possibly related bugs:
http://bugs.opensolaris.org/bugdatabase/view_bug.do?bug_id=6748436
http://defect.opensolaris.org/bz/show_bug.cgi?id=4174
-- 
This message posted from opensolaris.org