Disk Mirror break for backup purposes?

[Tim Lines]

We're about to do an oracle upgrade for one of our production oracle servers. As part of the upgrade, existing oracle tables will be modified. We're looking for a method to roll back to the old version that does not require the 12 hours a tape restore would take.

I know this has to have been done before, so I'd like some input from someone with real world experience. What I'd like to do is:

1. Create LUNs on the SAN that are sufficeint to mirror my existing oracle disk groups.

2. Create Disk group mapfiles using "vxexport -m -p ..."

3. Break the mirrors prior to the upgrade.

4a. The oracle upgrade goes as planned, and I return the unused mirrors to the SAN for use elsewhere. OR:

4b. After the upgrade goes awry, I vgexport the upgraded oracle vgs. I them use the mapfiles created in step 2 to reimport the disk groups using data on the preupgrade mirrors.

I don't know what I'll have to do to the mapfiles to make them work, I haven't even looked a a mapfile yet to see what's in there.

SANITY CHECK please! Advice also gratefully accepted.

[Tim Nelson · Accepted solution]

Another option. Mirror the current drives then execute lvsplit when the app is quiet.

You will get new lvols that you can just substitute in place of the old.

If you do not need them then lvsplit back togeter.

The vgexport with the map file will have issue as the vgimport will be looking for the same number of disks as the origional.

Depending on your array you could do replication/snap/clone there as well, then use the vgimport if needed on then snapped or cloned device(s).

[Aussan]

use lvsplit

you'll get lvol1 and lvol1b the lvol1b is the mirror

once done if you want to bring them back together use lvmerge

[Marcel Burggraeve]

Not much special going on in a mapfile.
It just 'translates' the logical volumes on the disk to the names used in the volumegroup.
Expample of a mapfile of vg00 :
1 lvol1
2 lvol2
3 lvol3
4 lvol4
5 lvol5
6 lvol6
7 lvol7
8 lvol8

[David Dilly]

be careful with the lvmerge command the syntax is lvmerge lvol1b lvol1 it means the lvol1b is synchronyse by the lvol1.

If you execute this : lvmerge lvol1 lvol1b it means the lvol1 is synchronyse by the lvol1b and you will lost all the transaction between the lvsplit and the lvmerge

Bye.
David