why list volume groups in cluster ascii?

[Dan Copeland]

Admins,

Please correct me if I'm wrong...the only situation where listing cluster vg's in the cluster ascii file is necessary -- is when you shutdown a cluster (w/ cmhaltcl) and then restart the cluster.

If I understand it correctly, the cluster unmarks the cluster vg's (vgchange -c n) upon cluster shutdown and re-marks those vg's listed in the cluster ascii (vgchange -c y) upon cluster startup.

Basically, I don't understand the need to list cluster vg's in the ascii file. Escpecially since it adds an hour or more to any cmapplyconf -C (even w/ the -k option).

Please clear up my misunderstanding.

tia,
Dan

[Sanjay_6]

Hi Dan

The vg don't get unmarked and marked at the cluster startup / shutdown.

The reason the vg's are listed in the cluster vg is because it tells the cluster daemon about the vg's that are part of the packages for the cluster. When a package is configured and it uses filesystems from a vg, unless the cluster is aware that the vg is part of the cluster, it won't be able to start the package on the adoptive node. The vg gets activated on the node on which the package that is using that vg is running.

Check out the managing SG manual on the hp doc site. http://docs.hp.com

Hope this helps.

Regds

[Dan Copeland]

our failover works fine w/o any vg's listed in the cluster ascii...

[melvyn burnard]

there is no NEED to list the VG's in the ascii file, unless you have a cluster lock disk. In this case the cluster lock disk VG MUST be declared in the ascii file.

However, having all of the SHARED VG's listed is useful, as it documents which VG's are used in the cluster packages, but it often also speeds certain operations up.
There is also the issue that if you had to reapply the cluster binary after making changes to the ascii file (say change the iP adress or node timeout values) then any VG's that are NOT listed in the ascii file will get unmarked as belonging to a cluster. Those that are listed do get the new cluster id rewritten to them

[Dan Copeland]

how can it unmark VG's that are active?

[melvyn burnard]

When a cmapplyconf -C Cluster.ascii is run, this causes a vgchange -c n to be run on each VG, and then a vgchange -c y for each VG listed in the cluster ascii file.
If you have a vg already active, it may or may not do the vgchange -c n succesfully, I have not tested this recently

[Stuart Abramson]

Dan:

I was always taught to include the Volume Groups in the cluster ascii file.

I forget why...

[Nguyen Anh Tien]

Hi Dan

1. Let discuss about package control file:
pkg control file will only mount each logical volume to assigned mount point (it do not active or deactive volume group)
2. Cluster ascii file.
Cluster ascii file need you list vg because it must enable each vg whenever it start (cmruncl). Expecically lock vg
You can get more information at.
http://docs.hp.com/en/B3936-90065/index.html