Shared Volume Group Activation

[Matt Goodfellow_1]

I am having an issue with activating a shared volume group on both nodes of a 2 node cluster. When I activate it on node 1 with a vgchange -a s vg# it activates as the server. When I go and activate it on node 2 I get "quorum not present or some physical devices are missing". However, if I deactivate the vg on node 1 and activate it on node 2 it will become the server and node 1 has the "quorum .... missing" issue. I have exported and imported the VG several times thinking I may have fat fingered something but I keep getting the same issue. Any answers would be great!!!

[Patrick Wallek]

You don't typically mount VGs on 2 servers at once. If this is part of a MC/SG cluster then you typically only mount the VG on the node that the package is running on. Having the VG activated and potentially filesystems mounted on 2 machines at the same could result in some major data corruption.

[Matt Goodfellow_1]

We are using OPS in which the vg's are active in a shared state on both nodes.

[Florian Heigl (new acc)]

Matt,
could this be some scsi reservation issue?

maybe You could check which volumes are missing and cross-check the pvid's against the volumes active on server #1

[Uday_S_Ankolekar]

Try this ..
vgchange -c y -S y

If this does not resolve problem try first convert vg in standard activation by

vgchange -c n -S n followed by vgchange -a y. Then activate VG in shared mode.

-USA..

[Kent Ostby]

I would check the VGID as mentioned and see if you can isolate which disk is giving you the problem.

I'm still not sure why it would only be the second activation that would give us a problem though.

If its the same disk then perhaps we need to restore the LVM headers on it.

[Matt Goodfellow_1]

Folk's it has to do with the time stamps on the VGSA (Volume Group Status Area). The server and the client see different time stamps. There is a primary and an alternate time stamp and if they are different you get this behavior. HP has a couple of unofficial work arounds and a patch that fixes the problem. PHKL_30553 for 11.0 and PHKL_30697 for 11.11.

[José Enrique González]

I got the same in a two-node cluster too. I solved it by applying vgcfgrestore from the first node:

vgcfgrestore -n /dev/vgname /dev/rdsk/disk1 /dev/rdsk/disk2 ...

and then vgexporting and vgimporting in the other node, being carefull you have well identified disk paths for that VG in the other node. It will sure work.