Service Guard Upgrade for 6 node

[Tan Tian Ho]

Wonder if someone can help? How can I upgrade
an 6 nodes cluster with mirror disk in UX10.20 to UX11.00 without interupting the services from the Servers?
Can I create a new cluster and transfer one node at a time over during upgrading ?

[generic_1]

I would imagine you are going to have to take some outage. It is generally best to build a fresh install then to upgrade to OS. But that is my personel preferance. Depending how your six node cluster is set up and how utilized the servers are. If you could drop say two systems out at a time upgrade them. There are obviously some issues to consider though such as resources, time windows, and testing. Also if you are going this route why not just upgrade to 11i an be up to date? Are there some software support issues for you to do that? Also do you have a mock/testing environment where you could test your upgrade to help you prepare as well as make sure things will go smoothly rather than experimenting on production :)? If you could tell us a little more about how many packages you have, what kind of servers, and utilization it might help.
I wish you the best with your upgrade.

[Steven E. Protter]

You probably want some outage.

You can do this though:

cmhaltnode

upgrade SG or whatever you want to upgrade.

put the node back in the cluster

cmhaltnode the next node.

You can possibly do this one at a time and avoid downtime if your packages are all properly configured to failover.

At some point, your cluster may go down because SG will not run mixed versions perfectly.

Best you plan some downtime.

I would not upgrade to 11.00, its scheduled to go away December 2006. Make the big jump to 11.11

SEP

[melvyn burnard]

Well the steps for a rolling upgrade are in Appendix E of the Managing Serviceguard manual available at
http://docs.hp.com/hpux/ha

If you do NOT want to do a rolling upgrade, and by that I mean you wish to cold install each node, then do the following:
1) Ensure you are running at LEAST SG 11.09 with a patch to support Multi-OS cluster, but preferably 11.13 or 11.14 of SG.
2) Create a n Ignite make_recovery tape to be able to recover in the event of problems.
3) Remove the target node from the cluster by editing your config files, and doing a cmapplycinf on line.
4) Install the node, install SG, and Patch it!.
5) Ensure your disk/file system configs are set up for the cluster on this node.
6) re-edit the SG config files and add in the node
7) test.

If all it works, repeat steps 2-7 for every other node.

[Geoff Wild]

I agree with Steven - go to 11.11.

You should be able to do this 1 node at a time - whether you upgrade or fresh install is up to you. Things to look out for - if running Oracle - talk to your DBA's - as they will have to re-link the Oracle binaries on each node you upgrade - if you move a package there - in order to fail back - you will have to re-link the Oracle binaries down to 10.20.....

As Melvyn said - follow the rolling upgrade steps in "Managing MC/SG"...


Rgds...Geoff