lvm on Linux is driving me nuts (vgexport/vgimport issue)

[Jean-Marc Beaudoin_1]

Good day,

Currently I am certifying Continuous Access scenario with LUNs from 2 sites on XP24000 raid arrays.

Node A is on site A and Node B on site B

LUNs from XP tour on site A is visible to Node A and B as for the LUNs from XP tour on site B is visible only for Node B.

+--------------+ +-----------------+
| Node A | | Node B |
| |<--CA--->| |
| A: 00:01 | | A: 00:01 |
| | | B: 01:01 |
+--------------+ +-----------------+

Where A represents Site A Tour and B, Site B Tour.

We use this type of configuration in case of disaster (loosing Site A) and all of our horcm instances runs on Node B (which controls the CA).

LUNs are mirrored.

I noticed on Linux that the vg is visible on node B even if it's local to Node A.

I noticed also that lvm on Linux is quite mininal with just a few options (HP-UX in that matters offers much more flexibility).

On Node A, the device is /dev/dm-8 and the mirrored LUN on Node B is /dev/dm-12
We are using multipath.

vgexport -p -m vgNAME.map vgNAME (Node A)
vgimport -m vgNAME.map /dev/dm-12 (Node B)

-m and -p (and many other nice options are not part of lvm under Linux).

vgchange --deltag NodeA
vgchange -a n NodeA
vgexport vgNAME

Just vgexport vgNAME on node A seems to work
So, of course, I cannot import the vg on node B

vgimport vgNAME on Node B fails.... of course

ltsap11:/hq/systprod_nfs/rds/scripts/data # vgimport vgTEST
WARNING: Ignoring duplicate config node: activation (seeking activation)
WARNING: Ignoring duplicate config node: activation (seeking activation/reserved_stack)
WARNING: Ignoring duplicate config node: activation (seeking activation/reserved_memory)
WARNING: Ignoring duplicate config node: activation (seeking activation/process_priority)
Found duplicate PV oJQv95YG5sGle9SnW1wSsaDiZ2QEsmpn: using /dev/dm-8 not /dev/dm-12
Volume group "vgTEST" is exported

What I am trying to do is it possible to do it under Linux ????

What would be the workarounds ?

Thanks in advance for your help

Regards,

[Ivan Ferreira]

There are some things that I don't understand in your description.

>>> LUNs are mirrored

Normally, every LUN mapped from an XP storage belogs to a raid group. ¿Is this what are you talking about? ¿That the LDEV is a mirror?


>>> LUNs from XP tour on site A is visible to Node A and B as for the LUNs from XP tour on site B is visible only for Node B.
>>> all of our horcm instances runs on Node B

¿How can your control your horcm instances if all instances runs on Node B and Node B only sees LUNs from site B.?


>>> I noticed on Linux that the vg is visible on node B even if it's local to Node A.

In a correct configuration, that should not be possible.

[Thomas Callahan]

I'm no expert on CA over 24k's...as we just got one 3 weeks ago...but I have been doing scenario's so here's my shot at it:

1. HORCM is not meant to run both instances on the same node. Several features appear to not work correctly when done like this, takeover's, pairsplits with -swaps, etc. You should be running an instance of horcm on each node to meet recommendations.

2. The error you are getting from LVM is because, even though you are changing the VGname, keep in mind that the underlying PV is still, block by block, the same (aka a mirror). There are commands to modify the PV label, last I remember.


My main suggestion is to revisit what you are doing, because to me, it seems like you are trying way to hard to accomplish site failure ability.

Why are you mirroring between sites if you are using CA?
Why are you making it so Host B can see both sets of CA LDEV's?

I'd love to lend my "expertise" in what I have learned, but need a bit more information.

[skt_skt]

"I noticed on Linux that the vg is visible on node B even if it's local to Node A"

it may show up in vgdisplay; But there wont be any "Open LV"..