topic Re: Volumegroup minor number in ServiceGuard cluster in Operating System - HP-UX

Volumegroup minor number in ServiceGuard cluster

Bruno Bossier_1 — Wed, 11 Aug 2004 04:27:52 GMT

All,

When you create a LVM volumegroup in a 2-node ServiceGuard cluster that will be shared, is it required that the minor number for the vg is exactly the same on both nodes ?

I have a vg with minornr. 0x040000 on node1, but with minornr. 0x060000 on node1. I had no problem de-activating the vg on node1 and activating it on node2 and mounting all filesystems. What will happen if this is done via a package ?

Cheers,
Bruno

Re: Volumegroup minor number in ServiceGuard cluster

Bharat Katkar — Wed, 11 Aug 2004 04:32:48 GMT

Hi Bruno,
Definitely yes. VG minor no. should be same on both the servers.
Infact you have to create Map file of that VG on Node1 and then import the same on Node2 to maintain the similarity of VG structure as well. Hope you have done that already.
Regards,

Re: Volumegroup minor number in ServiceGuard cluster

Bruno Bossier_1 — Wed, 11 Aug 2004 04:39:06 GMT

Well, I got a different answer from HP support in the meantime. A shared VG can have different minornr. between the 2 nodes.

What makes you say that they SHOULD have the same minornr. ?

Anyway, thanks for the reply !

Cheers,
Bruno

Re: Volumegroup minor number in ServiceGuard cluster

Ravi_8 — Wed, 11 Aug 2004 04:43:31 GMT

Hi,

yes minor should be the same on both nodes.

Re: Volumegroup minor number in ServiceGuard cluster

Bharat Katkar — Wed, 11 Aug 2004 04:52:11 GMT

Bruno,
Well, it is not compulsary but nodes being a part of one cluster it is recommended that they should not differ by Hardware/Software configurations. Exceptions works sometimes.
Regards,

Re: Volumegroup minor number in ServiceGuard cluster

Bruno Bossier_1 — Wed, 11 Aug 2004 04:57:14 GMT

Strange .... HP says it is not required ....

Node1:
------
crw-rw-rw- 1 root sys 64 0x040000 Aug 9 15:40 /dev/vg_PE1_exe1/group

Node2:
------
crw-rw-rw- 1 root sys 64 0x060000 Aug 10 09:56 /dev/vg_PE1_exe1/group

I had no problem on node1 manually :

umount all filesystems
vgchange -a n vg_PE1_exe1

and on node2 manually :

vgchange -a y vg_PE1_exe1
mount all filesystems

FYI : HP-UX 11.11 and ServiceGuard A.11.16.00

Cheers,
Bruno

Re: Volumegroup minor number in ServiceGuard cluster

Carsten Krege — Wed, 11 Aug 2004 05:01:11 GMT

The VG minor numbers of a shared VG should be the same on both nodes. In many cases you would not notice if they are different, but if your package is an NFS package, ie. the NFS server is configured to fail over from one machine to the other, the NFS clients will report "stale NFS file handle" after the package came up on the adoptive node.

The reason for this is that the NFS file handle also includes the VG minor number.

Therefore having consistent VG minor numbers is a must for users of the NFS toolkit and for SGeSAP (SG Extension for SAP). I don't know of other cases where this is relevant.

However, for the above reason, we recommend to have it consistent.

BTW, the SG manual, June 2004, p 193, states in a tiny note in the chapter "Distributing the Volume Group with LVM Commands": "Use the same minor number as on [the primary node]".

Carsten

Re: Volumegroup minor number in ServiceGuard cluster

John Palmer — Wed, 11 Aug 2004 05:04:23 GMT

Hi Bruno,

The VG minor number is a pointer to a table within the kernel which contains information about each volume group and as such is only relevant to the node on which it's configured.

Consequently you'll have no trouble at all manipulating your VG from within a package. The package control scripts issue the same commands that you've done manually.

However it's always a good idea in clustered environments to keep things simple and configure things in a consistent manner.

Regards,
John

Re: Volumegroup minor number in ServiceGuard cluster

Bruno Bossier_1 — Wed, 11 Aug 2004 05:36:34 GMT

So, if I understood this correctly :

As for the matter of NFS, the minornr. of the vg which holds the NFS filesystems SHOULD be the same.

As for any other vgs, it's not strictly required, although it is recommended for consistency reasons.

Correct ?

Bruno

Re: Volumegroup minor number in ServiceGuard cluster

Carsten Krege — Wed, 11 Aug 2004 07:49:59 GMT

> As for the matter of NFS, the minornr. of
> the vg which holds the NFS filesystems
> SHOULD be the same.

No, it MUST be the same. :)

> As for any other vgs, it's not strictly
> required, although it is recommended for
> consistency reasons.

Yes, exactly.

Carsten

Re: Volumegroup minor number in ServiceGuard cluster

Rita C Workman — Wed, 11 Aug 2004 07:55:53 GMT

I didn't read every response...so I may be repeating someone..

The minor DOES have to be the same on both nodes. If they are not, quoting the book :

...."then a cross link error can occur when the cluster comes up.
All major and minor numbers for both volume groups and logical volumes must be consistent across all machines in the cluster."

Rgrds,
Rita

Re: Volumegroup minor number in ServiceGuard cluster

Carsten Krege — Wed, 11 Aug 2004 08:13:39 GMT

Rita,

I'm curious what book you quoted. In my eyes you'll never get a cross device link error when your VG have different minor numbers. Cross device link errors are mainly caused by inconsistencies in /etc/lvmtab.

If it's an HP manual, I'll ask them to correct this.

Carsten

Re: Volumegroup minor number in ServiceGuard cluster

Rita C Workman — Wed, 11 Aug 2004 08:44:59 GMT

Carsten,

My MC/SG training manual that HP supplied from class.

Rgrds,
Rita