Upgrade from HP-UX 11.00 to HP-UX 11.i

[Sebastian Cross]

Hello,

we are looking to Migrate our 3 Oracle Databases from 9i a to 10g but this only support HP-UX 11i.
We have currently 3 oracle instances in the same server so I won't be able to perform a fresh install. I will have to upgrade the curren 11.00 to 11i. What is the best way to do this ?

Thanks in advance !

[Pete Randall]

http://www.docs.hp.com/en/5991-2805/index.html


Pete

[James R. Ferguson]

Hi Sebastian:

Cold-install. Upgrades often do not fare well. By doing a cold-installation you have the opportunity to resize vg00 to best accomodate the larger number of files and products in newer releases. Your patch base is clean from the start.

If you vg00 is pure operating system filesystems all you need do is 'vgimport' your non-vg00 volume groups.

Regards!

...JRF...

[Florian Heigl (new acc) · Accepted solution]

Sebastian, I'll do my next upgrade on saturday, so it's quite fresh in my head and I'll try to put down most of the steps.

- order a 2nd set of harddisks and put them in the system rack
- check firmware, cpu and general hardware support in 11.11
- run cfg2html to gather all config data, i.e. hardware paths
- check swlist for server-specific software
- check swlist and show_patches for server-specific patches
- aquire a downtime and set a bingo point, in my case this is 9am to 9pm, fallback is fixed at 7pm
- create a script that will save oracle startup scripts, passwd, services, group, nsswitch.conf, ssh keys and everything I can think of
- create a script to merge passwd, services, etc...
- vgexport -p -v -s -f -m /dev/ for every VG in the system.
- create a tarball containing all of the above, and store it offline
- schedule full backups
- make two make_tape_recovery images
- make one lan image
- verify on-line access to last night's backup, test by recovering something.
- at beginning of downtime check that all apps were shut down by the project admins
- shutdown system
- label and remove the original hard disks
- insert the new disks
- install 11.11 via ignite or any other method
- verify basic setup
- put initdefault to 2 in /etc/inittab
- reboot to rc2
- run pwconv to disable /etc/shadow for a moment
- merge old users in /etc/passwd
- run pwconv again
- check all hardware paths are as they were, otherwise fix them using the data I gathered in cfg2html using ioinit
- reboot and check everything looks nice
- verify kernel parameters are suitable and at least same as the old ones, maybe reboot again (ummm, I could've done that earlier on, but I don't think I'll do it tomorrow :)
- import volume groups and merge fstab
- mount oracle filesystems, check permissions et all
- relax
- give the DBA a ring to relink oracle instances, then to try bring them up to DBA mode
- in the meantime recover OvO and our batch scheduler and test if they work.
- if that's ok, do more checks and put initdefault 3 and reboot to production state
- build another ignite image and store away the old disks for a few weeks.

the hp-ux wise parts take 2-3 hours depending on the amount of preparation and reboot count. You can help Yourself by disabling extended hardware tests till the last reboot, You just have to be sure You re-enable them, usually it's not worth it, and it'll steal away Your coffee breaks.
Try to have everything prepared and possible contained in simple, step-by-step scripts, this will safe the most time.

The oracle relinks take about 5 minutes per instance on a very powerful and about 30 minutes on an average host.

If I find a problem unresolvable I'll try to fix it, restart and fix again till I reach 7pm, where I will cancel the upgrade and put the old disks back in; for the same reason it's important to hold off the non-DBA-mode database startup to the very last reboot.

Well, I hope that's a few pointers. :)

[Florian Heigl (new acc)]

I'll post an update tomorrow night containing all things that went wrong :)

my not being in bed already but facing a bottle of wine instead is probably one of them :p

[Florian Heigl (new acc)]

Well, it went bad, but for the sake of completeness I'll post a bit about the lessons learned :)

- All system backups of the prior night failed and were restarted in the morning (no big deal)
- The Ignite server configuration lost the part containing the hardcoded 100_FD duplex setting, bumping install time to somewhat estimated 20 hours (so we cancelled in the end)
- I shouldn't have gone past my self-set deadline hoping the install would eventually come to an end
- The old primary bootdisk failed (no big deal either)
- The root mirror was not ok (very big deal) and looking at my list - yes - I didn't check that.
- While I had sorted the ignite tapes on the rack, I didn't remove the outdated ones, and it turned out there is a high chance of picking an outdated one, even if the tape is labeled correctly. So better add a note about checking the ignite tape by interrupting the non-interactive restore.

In the middle of the night I calmed down and restored from a current tape and all was fine.

I'm now waiting to get an updated Ignite BootCD with correct duplex settings. :)

[Sebastian Cross]

Well, based on your experience now I'm thinking about to perform a cold-install.
I have two 18 GB mirrored boot disks (we won't buy a new one) and unfortunately we don't have a unit tape attached to the system (9000/800/L2000)....I know, i know it is very risky.
So, i thought about to:
1. Collect all the system conf. info.
2. Export all the Oracle VGs. (vg00 has only the OS)
3. Boot from the CD and perform the installation on the alternate boot disk (ex-mirror).
4. Boot from the new boot disk.
5. Import the oracle's VGs.
This is a very gral snap about the process but please let me know..what do you think ?

[Florian Heigl (new acc)]

Okay, if there really is no chance of getting another hard disk, splitting the mirror will have to do.

I think I'd shutdown the box to split the mirror by removing the primary disk (pull it out like two inches).

What is Your regular backup like?

No Ignite is quite a bit less of fallback than I had available, and it was the one I had to use, simply because the disk failed.

I'm not sure You should take responsibilty for going cheap and not buying a $150 harddisk.

Hmmmm. Can't You even borrow a tape drive from someone?
Maybe You can use a HPPA Linux or NetBSD boot cd and make a dd image of Your primary bootdisk via ssh to a linux host with like 20GB of diskspace. (Never tried it)

[Steven E. Protter]

Shalom,

After you begin accessing Oracle again, you will need to relink Oracle.

It is necessary before you start to make sure your Oracle version is supported on both the new and old OS.

If not, you will need to upgrade oracle prior to the OS upgrade.

Full up to date patching on the new OS should be done prior to attempting any oracle work.

In your situation, you will need to perform the Oracle software install after the OS upgrade and then follow the Oracle data migration steps. Some of those migration steps will need to be performed prior to the OS upgrade.

SEP

[Geoff Wild]

I was able to upgrade 11.0 to 11.i (quite some time ago) with no trouble at all.

Rgds...Geoff

[Deoncia Grayson_1]

I had the pleasure???? of upgrading from 11.0 to 11.11 because of Oracle and basically I went for a cold install off the bat because of my adventures with an upgrade prior to this. You are running a risk with having no way of taking a backup, but copying off vital system information to another server would be helpful and praying none of your disks are corrupted.

Anways, so I basically broke the mirror of the root disks and installed the operating system on one disk and ran all necessary patches and kernel changes before turning the system over to the database team so they could install Oracle 10G. this transistion went pretty smoothly. I didn't mirror the new operating system to the other disk until we had ran successfully for a week with the new upgrades.

[Sebastian Cross]

Today, I built an Ignite Image of the current vg00 in another server. On this way, and though I don't have a tape drive, I will be convered about the a DR backup and I will be able to boot thru the network for any probem. Even I will use this backup to resize the stand and the root file systems.