Making a secondary site the cluster primary site (370 -> PS/2)

[Jack F. Vogel]

In article <1990Jan12.184310.25876 at cunixf.cc.columbia.edu> ben at cunixf.cc.columbia.edu (Ben Fried) writes:
>I have a small TCF cluster composed of a 9370 and several PS/2s, all
>running 1.2 F50.  The 9370 is the primary site, and the PS/2s are all
>secondary sites.  There is no backbone site.
 
>I need to make one of the PS/2s, now a secondary site, become the
>primary site for the cluster.  It would be nice to make the 9370 into a
>backbone site while I'm at it, but that's not essential.  It is
>absolutely essential that I make the PS/2 into the primary site.
 
My my, the times they be achangin', actually seeing this kind of TCF stuff
on the net :-} :-}! OK Ben, here is what you need to do, and up front
remember that no warranty is implied, do this at your own risk, and may the
force be with you :-}!!

First off, BACK UP YOUR SYSTEM COMPLETELY!! Maybe a DDR of all the system
minidisks would be a good idea, but whatever because if you screw up in
this procedure you could thoroughly zonk the system.

Second, you cannot convert a secondary into a primary because it does not
store all files of the replicated root and it needs to. So, what you need
to do first is to make the PS/2's root into a backbone copy instead of a
secondary copy. I hope you have a couple of large disks on it :-}! You will
need to see how large the root filesystem of the primary is and make one of
equal size on the PS/2. Depending on how you installed this PS/2 in the
first place this may require you completely reinstall it making a root
minidisk with enough space. If you actually have enough space in the mini-
disk there is a simpler way, you simply come up dependent (generic) on the
cluster, then remake the root filesystem with the correct number of blocks
and inodes to match the primary, use the command:

		mkfs -r /dev/root blocks:inodes

Be sure the blocks are sufficient, and that the inodes are exactly equal
to the primary. Also make sure you include the -r flag, which says to make
this a replicated filesystem. It will ask you a number of questions, be
sure to specify that this is a backbone filesystem. After the filesystem
is made mount it on / with the command:

		mount /dev/root /

Once it is mounted you will want to repropagate the contents of the
replicated root, assuming the primary is site 1 and you are up as site
31, give the command:

		/etc/primrec 1 31 1

This will take a while, but when it finishes the PS/2 will have a backbone
copy of the root. Reboot it and come up as an independent member of the
cluster once just to sync things up. 

Third, now comes the tricky part :-}! The only difference between a primary
and backbone copy of the root is a bit setting in the flags of the superblock.
The tricky part is keeping the integrity of your cluster through the transition
between the primary sites. Here is the sequence I would follow: First, make
sure all copies of the replicated root across the cluster are in sync. You
can tell this by issuing the command 'rdf /' and observe that all the low
and high water marks are identical. Then starting with the primary take all
sites down to single user mode. Next go to the PS/2 that you want to make
the primary, as root logged in on the console in single user mode do the
following:
		fsdb /dev/root

It will display a number of lines of information, then type in 'S', this
will display superblock parameters, the one you care about is the flags,
it should say:
		flag:	0xa0

The bits in this unsigned short are defined in /usr/include/sys/filsys.h
but this value indicates a replicated backbone filesystem, we want to change
it to a primary replicated filesystem. Do this with the command:

		P=0x30 <enter>	(notice the lack of spaces, this is important

Then type in S again to redisplay the parameters and verify that the change
has been made. You should also have an fstore value of all. Now type q to
quit. Next you will need to edit the stanza for the root in /etc/filesystems.
You need to change the line 'type = repl' to read 'type = repl,primary'.
Finally, sync a couple of times and reboot. Now hold your breath, keep your
fingers crossed and hope like hell it comes up as a primary site :-}! You 
will be able to tell if when you come up your root is read/write and not
read-only. Come up multiuser, and now you will want to bring all other sites
back up EXCEPT THE OLD PRIMARY, they should come back up and join the cluster
with your new primary.

The last step is to modify your 9370 to be a backbone instead of a primary.
To do this you basically do the opposite of what you did for the PS/2. Run
fsdb on /dev/root and alter the flag from 0x30 to 0xa0, edit filesystems to
remove the primary flag from the root stanza, then reboot and bring it
back up on the cluster, it should now be a backbone site. Notice one side
effect of this is that the primary is no longer site 1, this shouldn't matter
but it does alter the parameters of some commands. Keep in mind, this is all
described from theory, I personally have not done this, although I believe
it has been done here at LCC on occassion for some special development.

So, there you have it in all its gory details :-}. As you can see, this
procedure is not for the faint of heart nor for those of fumbling fingers
:-} :-}! Proceed at your own risk and with great caution for, as I said,
this is not a supported practice. But if you want we might set up a conference
call between you, IBM Level 2 support, and myself (they know how to contact
me at LCC).

Disclaimer: This TCF trivia is my responsibility, not IBM's or LCC's

-- 
Jack F. Vogel			jackv at seas.ucla.edu
AIX Technical Support	              - or -
Locus Computing Corp.		jackv at ifs.umich.edu