Adding space in layered volumes..

[SaGu]

We have a volume which has two plexes each plex has LUN's from  storages in different datacentre(DC). This was done to avoid a situation that even if storage in one DC fails the data will be available from storage in other DC. Typical scenario of layered volume.


A command somewhat similar to the below  one was used to build the volume. As you must be aware that ordered was used so that the first four disks get added in plex 1 and last four from other DC get added in plex2.

#vxassist -g datadg -b -o ordered make new_volume <size> layout=stripe-mirror disk06_P_s1 disk06_P_s2 disk06_P_s3 disk06_P_s4 disk06_S_s1 disk06_S_s2 disk06_S_s3 disk06_S_s4

Now I want to add space in this volume.  The problem is how do i ensure that the disks from a specific DC get added to the plex which already has SD from that DC. This DG is configured in a box running Solaris(sparc).

[Marianne]

Which version of SF?
If you have 5.0 or above, you can enable site tagging. This feature is described in Chapter 14 of 5.0 VxVM Admin Guide.

[SaGu]

It's 4.1 . any ideas how to go ahead in that?

[565613584 · Accepted solution]

SaGu,

yes you have hit a classic problem and the approach in the manuals is too theoretical. In the real world you do not use a volume extension command or conversion or even use vxassist with an ordered option.  This is because the command line is atomic and too complicated and you are operating on live data.  In all probability you will make an error and the conversion or extension will commence putting the data on the wrong luns meaning you will have to clean up these records  before you can re-commence.

First what you need is enough luns on each data store to build a temporary  provisional volume. I assume you are using HDS, EMC or IBM shark as it sounds like you have a fairly high end enterprise set up. The key is use vxmake not vxassist.     vxmake will enable you to build the sub-disks and plexes from the ground up in ordered discrete steps, that you can verify as you go, to make a provisional mirror.

You can then attach the new  provisional mirror plex which has the exact config to the original volume. You can then divorce the primary and secondary original plexes  - maybe keep the secondary as a temporary backup.  Destroy the original plex to free your storage, Then re-create it at the larger size you require, once again using vxmake.  If you are feeling lucky you can use vxassist, but i would not advise it

When the new original plex has synced up you can divorce the provisional mirror and destroy it to reclaim the storage.  Finally wait say 24 hrs then destroy the original secondary mirror and re-create at the extended size and then reattach it.

The critical issue is you need enough temporary storage to build the provisional mirror and i need to see "vxprint -g data_dg -qht" [ assuming your disk group name is data_dg ] of your set up
and "vxdisk -o alldgs list" and i can step you through the vxmake commands or Symantec support can do this.  If you ring pm US East Coast time it will switch to Sydney and most of the chaps there are long term and very experienced.  If you have no contract then email me the outputs
and your contact details and i will see if i can step you through them.

-gregoire

[Marianne]

I see all of this as motivation to upgrade to 5.x...
Site tagging just makes so much sense!

[SaGu]

gregoire,


Thanks for the detailed solution.  Even I was suspecting that i might have to go with this long method. But now since you have confirmed so, I have started making a plan for it.

I think i'll be ok with the vxmake and should be able to come up with a decent plan. If I have doubts at any point then i'll trouble you again. :)

Thanks again for the advise.

[SaGu]

Marianne van den Berg 2,

Yes it motivates to upgrade!!! :)