third mirror and service guard

[Quin Hammes]

We have 2 L2000s in a cluster. They share access to two fc60 disk arrays. The service guard volume group is mirrored between the two arrays. I would like to create a 3rd mirror for backup purposes on the primary system. Do we have to establish this third mirror on the second system also? We will not be using this third mirror during a failover situation. If we have to establish this third mirror on the second system I assume that we would have to failover service guard to the second system, correct? btw, we are running hpux 11.0. Thanks in advance for any help and of course points will be given for any help.

Quin

[melvyn burnard · Accepted solution]

If you setup the VG to have lvs that have a 2nd mirror copy, i.e. -m 2 to the lvextend command, than you are probably going to add more PV's.
If so, then you have to extend the VG.
This updates the lvmtab on the system you do the command on, and this will result in the VG setup on th esecond node being inconsistent.
This could result in inconsistent behaviour/results in cluster operations.
I would strongly recommend that you do ensure tthat you have consistency across the cluster, and if you then need to split off the mirror for backup purposes, you can do that on the node the pkg is activated on.

[Quin Hammes]

So what would be the best way to add a pv to an existing service guarded volume group and establish the 3rd mirror? I am thinking...

1) stop the cluster
2) add the pv to the vg
3) create the 3rd mirror
4) vgexport on the first system
5) vgchange -a n on the first system
6) vgimport on the second system
7) vgchange -a y on the second to see if it worked
8) vgchange -a n on the second system
9) start the cluster

Or am I way off?

Quin

[melvyn burnard]

As far as I can remember, you would only need to stop the pkg to do this.
1) stop the pkg
2)vgchange -a e vgname
3) add the pv to the vg
4) create the 3rd mirror
5) vgexport -p -m on the first system
5) vgimport -m on the second system
7) vgchange -a n on the first system
8) vgchange -a e on the second system
9) vgchange -a n on the second system
9) start the pkg

I might suggest reading the Managing MC/ServiceGuard manual at:
http://docs.hp.com/hpux/ha
as this should also help

[Paul te Vaanholt]

Because of the possible inconsitencies between lvmtab and the VGDA on disk, I would not use a 3rd mirror. It will be difficult to create scripts that are fault proof, for example if the active node fails during execution of the script. As an alternative you can use the snapshots from Online JFS. This way you only create extra logical volumes once and you won't have to stop your package (although you may need to stop your application). For more information on snapshots see www.docs.hp.com.