Serviceguard
cancel
Showing results for 
Search instead for 
Did you mean: 

Rolling Upgrade...what if it fails over before I'm done..

 
SOLVED
Go to solution
Highlighted
Rita C Workman
Honored Contributor

Rolling Upgrade...what if it fails over before I'm done..

Greetings Forumers -

I am, fairly soon, planning to begin an O/S Cold Install & then a Rolling Upgrade on a 3 node cluster. Running on 3 K boxes (current O/S 10.2 and MC/SG 10.10).
I plan to begin with our development server, obvious reasons. I will first do a cold install of O/S up to 11i, then do load MC/SG 11.14, at that point I would have to re-config MC/SG on this dev_box.
Now I am aware of the rules for a rolling upgrade...but going from a 10.10 to an 11.14 I got to wonderin...

Soooo.... My question is:
If one of the other 2 remaining (still at old versions) servers has a failover event, will it be able to failover to the dev_box, now at 11i and MC/SG 11.14???
-or-
Will the failover only work if 2 of nodes are of the same versions O/S & MC/SG versions?

Thanks,
Rita


9 REPLIES 9
John Poff
Honored Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Hi Rita,

I think your last statement is correct. You'll have to kick your dev box out of the cluster before you start the OS and MC/SG upgrade, so it won't be available for the other nodes in case of failover.

If you have a failure during the upgrade, just do what I do. I put on a really nasty face (kind of natural anyway!) and tell them, "Yes it is broken, I am fixing it. Go away!" ;)

JP
BFA6
Respected Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Hi Rita,

I've just had a look in the ServiceGuard course notes about rolling upgrades, and it says
If a cluster fails before all nodes are upgraded remember the only nodes which can restart the cluster (via cmruncl) are those with the latest version of MC/ServiceGuard.

Not sure if this is of any help.


Regards,

Hilary
Christopher McCray_1
Honored Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Hello,

Check this link:

http://www2.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000062922323

http://www2.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000062686657

The answer is yes, you should be able to recover from a failover event. You can't have mixed versions of SG when initially creating a cluster, nor can you make modifications to the cluster w/ mixed versions.

Hope this helps

Chris
It wasn't me!!!!
Jean-Louis Phelix
Honored Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Hi,

I check in http://www.docs.hp.com/hpux/pdf/B3935-90050.pdf that this transition is supported for Rolling Upgrade. John is right, your 11i node will be out during first upgrade so there should be no problem. My question would rather concern a problem during a second upgrade (while having a running 10.20 node and a running 11i node). But it should also failover correctly. You are only limited with configuration changes during this period of time. The only problem I can see is this paragraph in the release notes :

NOTE A cold install of the operating system during a rolling upgrade is not supported. This is because the ServiceGuard conversion tools cannot be guaranteed to match the same network PPA numbers that were in use before the upgrade. Also, disk device file names can be changed by a cold install, causing mismatches between the ServiceGuard configuration and the system configuration.

Regards.
It works for me (© Bill McNAMARA ...)
melvyn burnard
Honored Contributor
Solution

Re: Rolling Upgrade...what if it fails over before I'm done..

doing a cold install of a SG node is NOT supported in a rolling upgrade. This is known to cause issues, and should be avoided.
I would suggest you change ths cluster config to remove any reference to the first node you ar erebuilding, remove it from ht cluster, and then do what you wish on htat system.
Then fire it up as a single node cluster, set it upr ready for your application, and then test what you can, remembering your application will ned to be upgraded normally. f you are happy with htis, hten start planning to move the app onto the new cluster as a single node cluster, move it and ensure all is ok, hten re-install hte other two nodes, bring them up to the same patche levels etc, and then add them in to the already running 11.x cluster
My house is the bank's, my money the wife's, But my opinions belong to me, not HP!
linuxfan
Honored Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Hi Rita,

I have never done a rolling upgrade from MC/SG 10.10 to 11.14 but more importantly what about the application that is running in the package.

Would it run without any changes on 11i? If you are using oracle then it would mean migrating the database. I would be more concerned about that.

In the past when i did rolling upgrades, the node (which was upgraded/cold installed) joined the cluster without a problem but i had to make sure all the middleware and application related environment was set up correctly.

I would also check to see if the 11.14 can convert the binary config(cmclconfig) from 10.10 version of MC/SG.

Also you stated you want to start with the development server. Is the application already ported/tested to run on 11i?

-Ramesh
They think they know but don't. At least I know I don't know - Socrates
Rita C Workman
Honored Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

SUMMARY - - - -

Melvyn,
Good suggestions, since I have to do both an O/S upgrade (not to mention we have to upgrade Oracle big time..) and MC/SG. Basically recreating the entire cluster new...and bringing each node, in turn, into the new cluster, looks to be the best way for me here.

Failover in the interim, I am simply going to tell them, won't occur. That will keep them moving to get this done asap. Otherwise, they will drag their feet, which is what got this place so far behind with everything in the first place.

Thanks for everyone's input !

Consider this post closed - points a'coming
A. Clay Stephenson
Acclaimed Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Rita,

Unfortunately you are in an almost impossible situation for a rolling upgrade because the 10.20 to 11x is almost ceratinly going to fail and unfortunately no other method would insure that the device paths remain constant especially if cards have been installed, moved, or removed since the original OS was installed. What is now controller c2 may become c1, for example.

Your only real option, as has been indicated, is starting a single-node cluster from scratch. A near-zero downtime ain't gonna happen in your case.
If it ain't broke, I can fix that.
Rita C Workman
Honored Contributor

Re: Rolling Upgrade...what if it fails over before I'm done..

Clay,

Don't I know it.

Remember when I mentioned the 'politics' of this place. It just keeps getting stickier.
I have no doubt that once we start, something will happen to the box on the old version, and I'm gonna hear about it.
..(I didn't put them in this mess, I just have to clean it up..Ah-A woman's work is never done..ha ha)

Ramesh,

Yeah, I know. The DBA's are on it. And I'm trying to scrape enough diskspace up for them. The fact they have to go from Oracle 7.3.4.0 (yes I meant ..0..not one patch ever loaded on this thing) is not going to be pleasant.

Final Thanks to All !

POST NOW CLOSED !!


...no more points will be awarded.