[geeks] 139556-08

[velociraptor]

On Wed, Dec 30, 2009 at 4:07 PM, Phil Stracchino <alaric at metrocast.net> wrote:
> velociraptor wrote:
>> On Tue, Dec 29, 2009 at 6:20 PM, Phil Stracchino <alaric at metrocast.net>
>> wrote:
>>> I got it in the end, but don't remember the steps that were necessary.
>>> At least, I THINK I did ...  now you're giving me doubts.
>>
>> ls -al on /usr/lib/libc.so.1 and check the date :-)
>
> Actually, for the actual file, /lib/libc.so.1; usr/lib/libc.so.1 is a
> symlink.  Mine's dated September 14, which seems about right.

Yeah, sorry, switched them around when I typed them.

After about 8 hours of fiddling around to try to fix it--including a
LU "upgrade" of Sol10u8 on an alternate BE which *still* had the
broken Java, I tossed in the towel and built a from scratch new
instance of Solaris 10 u8 in a new root pool.  The kicker was that if
I booted into failsafe and looked at the U8 boot environment I'd used
live upgrade on, the correct libc.so.1 was there!  Boot normally,
/lib/libc.so.1.  This leads me to believe there was some kind of
shenanigans going on with the overlays, and I just didn't feel like
frakking around with it anymore.  Not that I've seen much detail on
the whole overlay system in my readings over the last 2 years working
with Solaris 10.  I may run it by my guru ex-co-worker, since I
haven't nuked the old root pool yet.

Install to a new root pool, zfs send/receive the other data sets on
the old root pool, copy a few config files over, reinstall Virtual
Box, 98% done in about 6 hours including waiting for the OS to load
from DVD and datasets to copy.  I still need to rebuild netatalk.  I
chucked out Blastwave, trying to see if I can get a couple of more
packages, including netatalk, going under my Gentoo alt-prefix.
Biggest hassle was me missing two parent mounts on the old root pool
when I did the zfs import and having to reboot a couple of times
because I was too slow to realize it and do a zfs set mountpoint on
them (did the children when I imported but missed the parents).  Oh,
and I still need to install the patch cluster.

Lessons learned:

Install patches to an alternate boot environment using live upgrade OR
boot to failsafe and run the cluster against the alternate root.
Maybe this later is too conservative, not sure.  We'll see how
applying patches to a BE goes (though I know there's no libc patch in
there) in U8.

Patches are now package installs (maybe they always were, but I never
noticed that before).

They have improved "installcluster" dramatically in the last 18
months.  It's a lot better--it gives a summary of successes and
failures at the end and filters out the "this package is not
installed" false failures.  It also allows you to apply it to an
alternate root path, and requires a passcode to start--so you at least
have to give a head nod to reading the README.  This seems kind of
problematic for Jumpstart, though.

Check *all* kernel patch updates for "this could fsck up your system"
type files and make sure that they are updated after you patch.  Seems
obvious, but with the lack of problems I've had for the previous 3-4
years with the pca script, I guess I got lax.

Rebuilding Solaris, even manually, is pretty painless when you have
ZFS datasets.  Decided to move /opt to a separate dataset to make
keeping Gentoo alt-prefix snapshots lower overhead for easy roll-back
if the world breaks (it's Gentoo, it's inevitable :-).

=Nadine=