How to predict drive number change for 7.3->8.1 upgrade?

[Jeremy Chadwick]

On Thu, Sep 16, 2010 at 11:40:43AM +0200, Michael Sperber wrote:
> Stefan Bethke <stb at lassitu.de> writes:
> > Am 16.09.2010 um 11:05 schrieb Michael Sperber:
> >
> >> I just upgraded my desktop system from 7.3 to 8.1, and the main hard
> >> drive, which was /dev/ad6 before is now /dev/ad10.  Consequently, the
> >> initial boot failed when trying to mount the root file system from ad6.
> >> 
> >> The desktop system is now fixed, but I also have a rented server with
> >> only a serial console, and I worry that the upgrade is going to leave me
> >> with a dead machine.  Is there any way to predict how the drive number
> >> changes?  (Why does it change at all?)  If so, what's the proper way to
> >> tell the system the initial root device *before* rebooting?
> >
> > If you have a serial console, you can always enter the root device at
> > the prompt, so you can recover there.
> 
> I know.  But given the serial-console problems recently reported here, I
> was a bit reluctant to take the risk.

I assume you're referring to the issue reported by Oliver Fromme.  That
issue may -- not 100% certain at this point -- be related to the DCD
line on a serial port being used/honoured by uart(4).
 
We have numerous systems using RELENG_8 with reliable/working serial
console, but as I stated in the other thread, our wiring/equipment and
adapters differ from Oliver's.

I still have not tested the patch Ed provided due to my day (night) job
keeping me busy the past 24-48 hours.  I'll see if I can get to testing
it tonight.  The only reason I'm testing the patch, by the way, is to
see if *our* stuff suddenly breaks -- and if it does, I can still roll
it back remotely (via serial console).

Soapbox, for what it's worth:

Serial console unreliability and OS installs are both reasons why I rent
co-location space that's local to me (within driving distance).  I
cannot imagine having servers in another state or country which only
have serial console (e.g. PXE is not configured in the BIOS, BIOS lacks
serial redirection, no remote rebooter/power-cycle unit, no dial-in
modem, etc.).

Depending on how mission-critical your setup is, I would highly
recommend investing the time and money into a setup that does allow
access to the servers when serial console breaks -- an KVM-over-IP
device would be ideal, since it gives you VGA console via VNC or a Java
client.

In my case I'm just a single guy with a bunch of servers, and run what I
do as a (expensive) hobby.  KVM-over-IP devices are unreasonably
overpriced (like most "enterprise-grade" things), and I tend to shy away
from HP/Compaq ProLiant hardware (which have LOM/LOM2) since over the
years I've seen too many problems with them posted on the FreeBSD lists
(mainly relating to storage device driver problems), not to mention the
support contract costs...

-- 
| Jeremy Chadwick                                   jdc at parodius.com |
| Parodius Networking                       http://www.parodius.com/ |
| UNIX Systems Administrator                  Mountain View, CA, USA |
| Making life hard for others since 1977.              PGP: 4BD6C0CB |