[3183] in SIPB-AFS-requests
Thoughts on upgrading the SIPB cell
daemon@ATHENA.MIT.EDU (Garry Zacheiss)
Tue Nov 24 03:10:50 1998
Date: Tue, 24 Nov 1998 03:10:43 -0500
From: Garry Zacheiss <zacheiss@MIT.EDU>
To: sipb-afsreq@MIT.EDU
I spoke with jhawk via zephyr about some of the planned changes
to the sipb cell, and he suggested that I send some of my thoughts here.
I've already spoken to some of the sipb afs maintainers about the
planned upgrade and expressed interest in helping out.
It would seem that the first thing that should happen to the
sipb cell is to upgrade it to afs 3.4a. From what I understand, the
planned replacement for rosebud is an SS5 running Solaris 2.6/Athena
8.2. Since rosebud is a DB server, this makes the upgrade especially
interesting, since ultrix does not currently run afs 3.4a, and the
solaris machine cannot run 3.3a. 3.3a and 3.4a database servers are
incompatible, and so upgrading the cell to 3.4a would need to be done
somewhere in the middle of the upgrade of rosebud to an SS5, which seems
very complicated. I believe the exact change was the addition of new
fields to the VLDB, but the exact technical details probably aren't
excruciatingly relevant.
I may have a way to make this upgrade easier. As some of you
may have seen from my message to zone-cell a day or so ago, I built most
of AFS 3.4a for Ultrix 4.2a a couple of weeks ago and have been testing
it in the zone.mit.edu cell since that time, on the machine
my-nefarious-purposes.mit.edu, which is currently both a DB server and a
fileserver; It's been running without problems for about a week and a
half now, and hasn't caused any problems in the cell. I would be very
happy to see the binaries I built used in the sipb cell. The build tree
and source is currently located in:
/mit/ops/src/afs34a-patches5/pmax_ul4
which should be readable by all of the people on this list. If it's
not, please let me know and I'll fix that. This source tree is the same
one the ASO is currently running on the servers in the athena and dev
cells. Anyone who would be interested in looking at my test machine
running these binaries is also welcome to inspect it, and can be added
to the klogin if they aren't already present in the standard zone cell
klogin file.
Assuming there is any interest in these binaries, I'd like to
propose that they be used to upgrade the sipb cell to afs 3.4a in
preperation for exchanging rosebud the decstation with rosebud the SS5.
A suggested plan for the upgrade would be to update all of the binaries
on the two SunOS machines to 3.4a, and update the ptserver, vlserver,
and (possibly) fileserver binaries on rosebud. This could be done at
the same time as the normal Sun. morning restart, and, if all went well,
no real outage time should be incurred. Assuming this went well, there
would be no problem in bringing up a solaris DB server in the cell
shortly thereafter. You would, however, want to back up the VLDB and
the PRDB before the upgrade.
Assuming the plan is still to use Solaris 2.6 as the OS for rb's
replacement, I have Solaris 2.6 afs 3.4a server binaries built from
essentially the same source base that I used to upgrade timex, a zone
cell db and fileserver, to solaris 2.6, which are currently
interoperating with SunOS and Ultrix DB and fileservers in the zone
cell.
If there is interest in using this plan for upgrading the sipb
cell, please let me know. Since it's currently after drop date, doing
the rosebud upgrade from DECstation to SS5 is probably a task best left
to IAP, since there might be some outage window for data during that
upgrade, but updating the cell to 3.4a could still possibly happen
before then, since there should be little to no data inaccessibility for
that upgrade, beyond what is normally noticed with the weekly
restart. Again, feedback on what I've outlined here would be greatly
appreciated.
Garry
