HP-UX update to 11i in the cluster environment.

[Insu Kim]

Hello,

I have been running a two-node cluster for years on HP-UX 10.20.
Now, I plan to update 10.20 to 11i in a limited time frame because HP announced supprt end for HP-UX 10.20.

I know that rolling upgrade shown in the MC/SG user's guide is possible but I wonder if I can update each node using cold-install procedure while cluster is running because I'd like to adjust the size of root file system on each node.

( As I know, rolling upgrade doesn't give us an opportunity to resize the root file system.)

I know that cold-install is going to wipe out all the existing configurations including MC/SG configuration and binary files but I have a plan to install every product using swinstall and restore every configuration file from the backup after upgrade is done.

Any ideas or opinions would be greatly appreciated.

[melvyn burnard]

Well the answer is you cannot do a cold install of one node while the cluster knows about this node.
This is unsupported, and usually results in a cluster that is no longer able to work.
If you wish to do this then the best practice would be to reconfigure the cluster so that it does not know about the node you are about to cold install, i.e. remove this node from the cluster. This leaves the cluster running as a single node cluster.
Make a make_recovery tape of hte removed node, JUST IN CASE.
You then do the work on the removed node, ensuring your application software is installed, and make sure you are happy with what you have done.
Once you are at this stage, you will need to decide how you proceed, as you are now talking about downtime.
You now need to create the updated node as another single node cluster, import your vg's etc,and check these are ok.
Now halt the old cluster, and mark the shared vg's as cluster unaware.
On the new clsuter, activate the vg's in non-exclusive mode, and ensure the file systems mount at the correct points, etc.
Now setup your package configurations, check them, and apply the config, and finally TEST.

At this stage, if all is ok, you now re-install hte other node, and when ready, you add it into the new exisiting cluster.

As you can see, a lot of work, and I recommend you document the steps, with checkpoints, and contingency plans.

HtH

[Stephen Doud]

Hello Insu,

The "rolling upgrade" procedure specified in the "Managing MC/ServiceGuard" manual is the "supported" strategy to keep the applications as available as possible. HP has not tested any other method that will preserve some measure of availability of the application.

Having said that, I have assisted customers who have done what you want to do, ie. install the update O/S, and join the server back into the cluster. As a minimum, like Melvyn said, create a make_recovery tape before the upgrade.

Per the manual: "You can roll forward from any previous revision to any higher revision. For example, it is possible to roll from MC/ServiceGuard version A.10.05 on HP-UX 10.10 to version A.11.12 on HP-UX 11.00.
"

Customers like you haved used this procedure on the node to be upgraded:
0) cmhaltnode, make_recovery, set AUTOSTART_CMCLD=0 in /etc/rc.config.d/cmcluster
1) install the new O/S and customize it
2) vgimport the "clustered" VGs (use the '-s' option that utilizes MAP files that contain the VGID)
3) copy /etc/cmcluster directory and files to the newly upgraded system
4) run the /usr/sbin/convert command on the cmclconfig file to make it compatible with the cluster:
# cd /etc/cmcluster
# convert -f cmclconfig

Step 4 makes the new node's cluster binary compatible with the older version node. The node may now be joined into the cluster. Whether it can run the applications at the new version of the O/S is another issue altogether.

We don't guarantee such a procedure will work. Just passing along how some customers have performed the upgrade if it helps you succeed.

-s.

[melvyn burnard]

Please note, the procedure that Stephen has outlined sometimes works, sometimes dos not, and is UNSUPPORTED.