Lower capacity mirrored disks to be upgraded to higher capacity disks

[Anil S Bhandarkar]

I have 2 x 36 GB mirrored data disks(non-os) say c1t0d0 and c2t0d0.They have to be replaced with higher capacity disks .The max PE per PV is set to accomodate the 73 GB disks.
I would like to retain the data on the old disks.

I intend to do the following
1. Pull out the secondary of the mirror ie 36Gb.
2. Install the higher capacity disk ie 73GB
3.vgcfgrestore -n /dev/vg01 /dev/rdsk/c2t0d0
4. Vgsync /dev/vg01
5.Remove the primary 36 Gb disk
6. Install the other higher capacity disk ie 73 gb
7. vgcfgrestore -n /dev/vg01 /dev/rdsk/c1t0d0
8.vgsync /dev/vg01

Is the above the right way of doing it?When i do the vgcfgrestore ,will it show total PE ( from vgdisplay -v) equal to the size of the old 36 Gb or the size of 73Gb?

[Anu Mathew]

Hi Anil,

Greetings..!

vgcfgrestore reads from the default configuration backup file (created with 36 GB drives).

Hence when you restore the same vg on 73 GB drives, you get the same number of PEs, effectively giving you not more than 36 GB.

I guess your PVs will be utilised only for 36 GB if you attempt a vgrestore.



Thanks

Anu Mathew

[Tracy_5]

I don't think that will work. Or rather, it will give an LE-to-PE mapping just like before (but no more PEs). Better to pvcreate / vgextend over the new big disks, mirror lvols, then reduce out the copies associated with the smaller disks, and vgreduce them out when they're unused.

Regards,
Tracy

[Brian M Rawlings]

Anil: I don't think this is the best way to go.

I would suggest, if possible, just add the larger drive/s, then do:
vgextend /dev/ /dev/dsk/ (add new 73GB to VG)
lvextend -m2 /dev// /dev/dsk/ (add 73GB as 3rd mirror)
lvreduce -m 1 /dev// /dev/dsk/ (remove one 36GB drive, either one)
vgreduce /dev/ /dev/dsk/ (remove it also from the VG)
[repeat steps to add 2nd 73GB as 3rd mirror]
vgextend /dev/ /dev/dsk/ (add 2nd 73GB to VG)
lvextend -m2 /dev// /dev/dsk/ (add 2nd 73GB as 3rd mirror)
[now you have both 73GB drives in the VG, with same extents written to each. So, you can repeat the 'lvreduce' and 'vgreduce' steps for the remaining 36GB drive]

This way, you DON'T keep the old 36GB info in the VG, it knows about the 73GB drives, and all extents are available to do 'lvextend' on your LVOL/s to make use of it (or add LVOLs, whatever), with extendfs (unmounted) or fsadm or fsadm_vxfs (mounted/live) to extend the file system/s to use the new space.

If you don't have free slots, the procedure is basically the same, but the order is different:
1> remove the mirror secondary
2> add one 73GB, and extend to it
3> Once all extents are in sync, remove the other 36GB
4> add 2nd 73GB, and extend to it.

Same basic result, but you encounter periods where you are unmirrored (a few hours to sync back up, is all, twice).

This is not very detailed, but it is an approach that I've used, even when moving from an old Nike array to a much larger Autoraid array, with good results. We could have been on-line the whole time, but elected to reboot after all was said and done, just to prove that all the old "PVs" (LUNs, in my case) were no longer expected, and that a reboot didn't show any wierd problems.

(It did, actually, but we were on an ancient rev of 10.20, no patches since 1997... you can imagine. The issues were not related to the procedure, but to the fact that it was a cluster, and we were activating volume groups in "exclusive" mode)

So, this is my suggestion.
NOTE: this would NOT work if the max PE per PV could not accodate the 73GB drives (you state that it can). If this value is stuck at the default size, you have to wipe the VG and recreate it with correct figures for max PE per PV.

Hope you find this useful.

Regards, --bmr

[Scott Van Kalken]

Just to add to the fray....

What I tend to do is the following, assuming that they're data disks and not root VG.


1. IgniteUX backup (why not?)
2. lvreduce filesystems.
3. vgreduce disk out of VG.
4. replace one disk.
5. pvcreate new disk.
6. vgcreate /dev/vgtmp
7. create new filesystems of new size on new disk named fs_tmp.
8.run the following command for each filesystem on old disk.

vxdump 0f - /filesystem | (cd /newfs; vxrestore xf -)

do this from the old disk to the new disk. This is documented in the restore man page.

do this until you've got everything across. Scripting it is best.

9. lvreduce, vgreduce original disk out.

10. vgremove vg01

11. Follow this process for renaming /dev/vg_tmp to /dev/vg01:

vgchange -a n /dev/vg_tmp
==> umount the FS and deactivate it
vgexport ???m /tmp/mapfile /dev/vg_tmp
mkdir /dev/vg01
mknod /dev/vg01/group c 64 0x010000
vgimport ???m /tmp/mapfile /dev/vg01 /dev/dsk/ /dev/dsk/ ..
vgchange -a y /dev/vg01
vgcfgbackup /dev/vg01

many thanks to SK Chan for this process. It's flawless.

Then you're all set.

It take a little longer, but doing it this way I've never had anything go wrong....


I also tend to write up a diagram to check off as I go.

For example:

c5t5d0 c5t6d0 (old in blue)
vg01 vg01
lvol1 lvol1
***** *****
***** *****
***** *****
vg_tmp vg_tmp (new in red)
lvol1 lvol1


something like that, so that I can double check what I'm doing as I go. It also means you don't start making things up on the fly.

When I'm done, I do an lvdisplay -v on every FS to ensure that there are no stale extents and that mirror copies = 1 (I ALWAYS forget to mirror at least one of them - usually toward the end).

Hope this helps a little

Scott.

[Anil S Bhandarkar]

Thanks to all for the solutions.

There are no slots available.Therefore it is imperative that i break the mirror and pull out one of the 36 GB disk.Insert 73 GB disk and re-mirror.Then break the mirror and pull out the other 36 GB disk and insert the other 73GB disk and re-mirror.

I want the data on both the 36 Gb disks to be intact so that i can revert to the original situtaion if required.Can i pull the 36 Gb disk (so that data remains.This disk with have stale extents)and do
1.lvreduce the logical volumes
2. vgreduce and remove the pv from vol. group
3Remove the 36GB secondary
4.Install 73 GB
5. Run pvcreate
6.vgextend and include the 73GB in the vol group
7.lvextend and mirror the logical volumes.
8. Now remove 36 GB primary
9.Repeat steps 1 to 7

Is the above ok to attain my dual objective to upgrade my mirrored disks to 73 GB and retain the data on the 36Gb disks.In case i wish to revert my original config. I ineret the two 36 GB disks and do a vgcfrestore on the volume group.I have taken a fresh vgcfgbackup.

[Volker Borowski]

Hi,

I would say, your last approach would give you the desired result on the 73 GB disks, but if you do the lvreduce an so on, the configuration on the 36 GB Disks will be lost, so no "plug back" is possible.

So the way to split and re-mirror to the new disks will not be possible with a re-plugable configuration in my opinion. Instead you might consider this:
- Do a clean shutdown
- pull out one mirror disk
- put in new disk with diffrent SCSI ID !
- Create new volume group on new disk
- Create new lvol Layout as desired
- Copy data to new disk (cpio / tar / dd, if raw data like database on it)
- Do a clean shutdown, and exchange second disk with new SCSI-ID
- complete mirror configuration

Now this setup will complain for missing disks and volume groups missing. But the entire Volume configuration is intact!
So to fall back, pull out your 73ers and plug back the 36er you pulled out last. Then remirror this 36er to the second one again.

This is more costy in terms of the copy process, but will leave your lvol information intact on both 36er disks and give you a chance to fallback. If you destroy the lvol configuration on the 36ers, you will need to re-create and restore for fallback.

...but to be honest, I would do a good tape backup and forget the 36er Disks, because if you need them anyway, my fallback would be to do a new configuration on the 73GB disks an do a restore.

Just my 0.02???
Volker

[Roberto Arias]

Hello Ali:
The PE is limited for the higher capacity disk when you made pvcreate. The solution is make pvcreate to alternate disk of 73Gb, recover from backup and syncronize disk.

best regards