Upgrading UX 10.20 -> 11.i on a serviceguard-cluster

[Hugo Wilhelmus]

I'm going to upgrade a cluster from UX10.20 to
11.i. Now 1 machine is currently no production, so i want to start upgrading this machine to 11.i. After that installing a new informix-version (the older version runs on the other machine) and switch users to the upgraded machine. Currently have Serviceguard 10.10. What is the best way to do this upgrade, esp. with a running cluster, 2 packages on 1 machine and 2 on the other.
The last 2 packages (on the machine i'm gonna upgrade first) wount be used in the new configuration.

[A. Clay Stephenson · Accepted solution]

Because the 10.20 to 11.x upgrade is almost certainly doomed to failure (I would rate your probability of success at 0.05 or less) a rolling upgrade is not feasible.

I did one of these several months ago and my approach was to save all the MCSG configurations, cold install the OS, applications including MCSG, and latest QPK's. Next, vgimport the VG's, and then finally build the new MCSG cluster and packages.

Make sure that you do an ioscan -fn | lp before and after to make sure that device nodes did not change. The actual time required to get the MCSG cluster configured for 4 packages in my case was less than 30 minutes.

[Ashwani Kashyap]

This is a rather cmplex upgrade and a rolling upgrade , I don't think would succeed .

I owuld rather take complete backup of the systems , save all the service guard configurations , then cold install 11i , apply all the patches including gold pack/QPK . Then install the latest service guard , apply all service guard patches , import all the old volume groups , build the service guard from the previously saved configs , bring up the cluster and then actually touch informix .

[Paula J Frazer-Campbell]

Hugo

Upgrade 10.2 to 11 is a problem and there are many threads on this forum stating DONT do it.

Do a fresh install.

Prior to install do two full backups and an ignite recover tape (again twice)

I have not added much to what has alredy been said but my voice agreeing to upgrade NO - fresh install YES.

Paula

[Geoff Wild]

1 - I would remove the node from the cluster.

2 - Upgrade cluster to at least MC/SG 11.14.

3 - Reload node you took out with 11i.
4 - Install MC/SG on it.
5 - Add it back to the cluster.

6 - Then, migrate your packages to the new node - if oracle, you will have to re-link oracle binaries (talk to your dba's).

7 - Then remove the 10.20 node, and repeat steps 3-5 for this box.

Check the SG matrix - make sure you can upgrade to 11.14 on 10.20....

You may even want to go to 11.15...

Rgds...Geoff

[Rita C Workman]

Like Clay, I went thru the same thing a few months ago, and did it basically the same way as him......following some old tips that JRF has hanging around the Forums on doing Cold Installs...

1.Document your systems....
2.Backup up your systems..backup..backup !
3.Either copy your special files to tape using tar or my personal favorite is to copy them over to another server for quick rcp restoring.

After doing my mapfiles and putting them off to safety, we removed two nodes and then cold installed the O/S; loaded the software & patches and built a new Cluster on the first two nodes, redid our packages, and from there did each other node seperately & then added them into the cluster.
Granted I'm stating things quite simplistically .. but you get the general idea..

Sounds hard...but taking it step by step...documenting & backing things up...it wasn't too bad.

Rgrds,
Rita

[Hugo Wilhelmus]

Thanks for all the answers, it's clear
to me to do a fresh install.
But i'm still wondering how to keep my cluster active (on 1 machine). The productionmachine is a 24h. machine so downtime should be low. I know that it sounds
stupid to have a cluster of 1 machine !

[G. Vrijhoeven]

Hi,

If you have the hardware for it, I think it is the best solution to delete one node, upgrade it, keeping your application running on the remaining node. Make sure you hava an ignite tape ready to reactivate the first node in case of emergancy. Your down time can be reduced to a few hours during the migration period.

Gideon

[melvyn burnard]

There is nothing wrong in having a cluster comprising of one node, some people do htis just to use the network failover portion, or testing.
In this case, if you prefer to go the route of re-installing, I would suggest:
make an ignite backup of both systems (playing it safe here)
modify the cluster to make it only a single node. (can be done without taking the clusyer offline)
install the first node, patch it, install SG, patch it.
Create a single node new cluster.
Install the newer version of your application(s) and then create your package(s) and test them.
Make another ignite image of this system.
Now allow your users to access the new packages, testing everything.
Once you are happy, switch all your users to hte new cluster/packages, and repeat the install/patch for the second node, and then add it in to hte running cluster.