MC Serviceguard cluster COM version

[doris grady]

Hello,
I note that all Serviceguard nodes in cluster must be running the same version of Cluster Object Mgr (COM) that is installed with HP-UX, in order for it to failover successfully. I have one K580 that has HP-UX 11.11, COM version B3935DA, and Service Guard version A.11.09. But we need to cluster in a N4000 and it is going to have another version of COM with a new HP-UX build. Question is, although 1 version of COM comes with a specific version of HP-UX, does HP offer a patch to get to the same version...or how does that work? We are not at liberty to change anything on the K580.
Thanks!

[Geoff Wild]

If you install the same version of ServiceGuard - you should have the same version of COM...

BTW - you should really upgrade your version - 11.09 is quite out of date...current version is SG 11.17. (11.16 on HPUX 11.11)...

Rgds...Geoff

[doris grady]

According to the HP documentation for HP-UX 11i Mission Critical OE, Cluster Object Manager is automatically installed as a part of HP-UX -- not as a part of Service Guard...let me know if I am interpreting this wrongly.

[Geoff Wild]

That does sound right...

One thing that is bothering me - you say you are running 11.09 of serviceguard on a 11.11 server? I was sure the the earliest version supported on HPUX 11.11 was 11.14.

Why not just upgrade the K580 to the same HPUX version you are installing on the N class?

Rgds...Geoff

[doris grady]

Hi Jeff, glad you asked that. It is not straight forward. We are unable to touch the K580 because of the following environment:We are planning to replace 2 appl servers in the K580 production SG cluster environment -- and the replacement servers will be 2 N4000s. We want to test performance on the new N4000 servers by breaking the cluster of the 2 K580s, replacing the inactive K580 with the new (already built) N4000, re-instating the cluster (now the active K580 and inactive N4000 combo) and then failing over from still existing active K580 to the new N4000 so that this new server becomes the active box. This way, we can ensure that performance measures up on the new N4000 boxes before fully committing to them. So far we've confirmed that both the K580 and N4000s currently have HP-UX 11.11, COM version B3935DA, and Service Guard version A.11.09.(don't know how we got this version but it is what it is)
The potential prob is that we must rebuild the replacement N4000 boxes with HP-UX 11.11 and the application piece before testing performance as mentioned about, and have not fully determined what it is but we believe our current HP-UX IT build will have a newer version of COM (as u have mentioned) and if that is the case, we wanted to know if there is a patch we can apply to get to the older version of COM to match...so that we can get a successfull failover. So bottom line, we can muck with the N4000 but not the K580. Or after reading this scenario, if you have alternative remedys, I'd be quite interested. Thanks.

[Geoff Wild]

Well...I have done ServiceGuard migrations in the past - there are 2 ways - bring new servers into the existing cluster or build a new cluster and migrate the data over.

I went from a 4 node 11.0 V-class/Kclass SG to 2 node 11.11 RP7410.

I chose to bring the new ones into the existing cluster (6 node) - then when happy - remove the old servers.

In order to doi this - you have to have the same version of ServiceGuard running on all nodes...(well, you could have different versions - but not recomended - only for when one does a rolling upgrade).

That said, you may be able to install 11.09 on the "new" servers - then bring them into the cluster - but my gut is saying to upgrade your existing to at least 11.14 (11.16 would be even better).

Then bring the new server into the existing cluster (no reason to not have a 3 node - unless you can't share the disks - IE - your cluster isn't on a SAN).

Rgds...Geoff

[Carsten Krege · Accepted solution]

Doris,

COM is automatically installed as a dependency when you install SG, it does *not* come with the OS. Each version of SG installs its own version of COM. See the following page what SG version uses which COM version:
http://docs.hp.com/en/5971/SG-SGeRAC-EMSSupport.pdf

This said: if different nodes have the same SG version, they should have the same COM version.

COM is basically the API of SG and is used by applications like SG Manager, Continentalclusters and Veritas Cluster Volume Manager 3.x. COM does not have an influence if SG failover works or not and your SG cluster would (in most cases) work fine even if COM was not available at all.

Note that support for SG A.11.09 ended on Dec 31, 2003, see http://www.hp.com/softwarereleases/releases-media2/notices/SGSupportLetterdocv4.pdf .

Also note that you cannot run a cluster using different versions of SG, ie. if you add nodes to an existing cluster that runs SG A.11.09, the new nodes also need to have SG A.11.09. Mixed versions clusters are unsupported and don't work.

From one of your comments I'm under the impression that the old nodes and new nodes run different OS versions. It is allowed to mix HPUX 11.00 and 11.11 nodes in a cluster (still using the same version of SG!!), but not to mix HPUX 11.11 and 11.23 nodes (well SG A.11.09 is not supported on HPUX 11.23 anyway).

Hope this helps.

Carsten