SUMMARY: Live Upgrade ordeal SXDE -> S10 10/08

You may recall I was trying to Live Upgrade from SXDE (nevada build 64)
to Solaris 10 10/08 (on AMD64), and the resulting kernel panicked.

I got one reply that said what I was trying to do really was not
supported (I still wonder what SXDE adopters are supposed to do when
they want to go to the released S10? Punt?) and that some things were
not going to work anyway, such as SMF collisions, and particularly that
the ZFS pools could not be taken "back" to S10.

Well.

I was resigned to doing a fresh install while booted from the DVD into
the partition I was trying to LU into.  However, that failed because it
claimed that the disk partitioning was invalid (without saying why!),
even though fdisk reported the partitioning just fine and it was the way
I had set it ages ago, and the installer wanted to Blow It Away and use
the whole disk for Solaris, even though I had two other OSes on that
disk already.  That was unacceptable, and the installer wouldn't have it
any other way.  My theory is that it wanted sector 0 to be in the
Solaris partition (not that the Solaris parition had to be first in the
table--that would be trivial and harmless to arrange), and it wasn't. 
But grub was already working on this disk; maybe those other two OSes
aren't bootable right now.

So that was a non-starter.

Undaunted, I still wanted a fresh install in that partition.  I believe
I tried just creating a root filesystem and pkgadding all of S10 onto
it, plus an installgrub, but that didn't work for some reason. What I
ended up doing was so cumbersome and time-consuming that I may not ever
want to run a developer edition again.  First, get things to the point
where they failed the first time, i.e. lucreate and luupgrade. Then
lumount it and pkgrm everything in sight. Use sledgehammer if necessary
(remove the pre/post-remove scripts that complain). Then go to the
mounted S10 10/08 DVD Solaris_10/Product and do  for i in `cat .order`
do yes | pkgadd -d . -R /.alt.whatever $i done.  There are three or four
packages very early in .order that won't install that way, and
everything depends upon them; install them by hand first as they are .i
packages, then do that loop.   When this finishes, luactivate it and
boot it.

[There needs to be a Live Install that does more-or-less the above, plus
whatever additional configuration it needs. For fun I may just try
installing into VirtualBox (needs nv 80 or S10U4).  Maybe "Live Install"
is coming with the ZFS root stuff.  I may do ZFS root here later.  Much
later.]

While this instance is pretty-much unconfigured, the one thing I was
happy to see already taken care of was the ZFS pool (it wants an
upgrade--later).  It attached and mounted without incident (except for
sharing; I have a 3rd party gigE card needing a driver installed
first).  The pool is version 6, if that explains anything.  And there's
the smf stuff to deal with. (The gigE now is working.)


P.S. I cannot remember the last time a Solaris upgrade was painless, and
flexible enough.  Must have been around 2.6 -> 7 or so.

-- 
Jeff Woolsey {woolsey,jlw}@jlw.com,first.last@gmail.com
Nature abhors a straight antenna, a clean lens, and unused storage capacity.
"Delete! Delete! OK!" -Dr. Bronner on disk space management
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers