HPSG logical volume configuration

[gourv]

Hi ALL,

I am using HPSG cluster on Redhat 6.5 x86. I have put all lvm disk information on cmcluster configuration file which resides from one volume group ( vg1) 

I want to create another two logical volmue from that volume group (vg1) and want to use localy, that means want to mount localy  like /sapmnt/PX3 differenet mount point whcih is not mentoned in cluster configuration.

Is it possible to use that local file system , is there any discripency?  Please sugget me.

Gourv

[Patrick Wallek]

I would not recommend doing that.

Having a file system from a cluster-aware volume group mounted outside of a package will cause problems when you try to stop the package and likely prevent it from failing over to the other node.

When your package stops via its stop script, it will attempt to umount all filesystems and deactivate the volume group on node1.  Then, when you start the package on node2 it has to activate the VG and then mount the file systems.

If you have logical volumes mounted from the cluster VG on node1 that you package is not aware of it will NOT be able to umount everything and deactivate the VG, because there are still file systems mounted., then preventing the package from starting on node2.

If you need some local mount points you would be better off finding some additional disk and creating a completely new VG that is outside of SG control.

[gourv]

Hi Patrick,

I understood your point . But we did not failover the cluster or down the server also. 

I have craeted two lvm and mounted on particuler mount point and application team hosted new application on that partition.

But after someday , i have seen no data found on one mount point i.e /sapmnt/PX3. And it was showing mounted and size was showing 1GB instaed of 6GB. and application was down.  Then i have remounted the lvm on same mount point and it was showing orginal size .i.e 6 GB. That mount point data was accessible via samba . 

Now my question is why it was happend like that ? Is there any reson ?  

Note: no server reboot and no cluster failove happned.

[Patrick Wallek]

When you say the mounted size was showing 1 GB instead of 6 GB, are you absolutely sure it was showing the correct mountpoint and not a different one?

If you try to mount a file system that is already mounted you will typically get an error that says the filesystem is already mounted.  Since you didn't get that error, I would guess that your file system was unmounted for some reason.  Also, since it came back to the 6 GB after you remounted it, that also indicates that your file system was somehow unmounted.

[gourv]

Hi Patrick,

Yes , it was showing same mount point , only size was incorrect. That was strange.

When I remonted the lvm on same mount point , there was no error occured. After that I was seeing the mount point size was proper ( 6GB).

That was the stragne , how it was possible . I need that RCA why it was happned. I could not find any logs also to find out RCA.

I talked with Redhat vendor and they are also unable to find out anything wrong. But I have to submit RCA to the client as buisness was down.

Please help me or please share any document where clearly mention that local file system will not support in HPSG environment which file system or volumegroup is under cluster configuration. 

Actuall, 8 file system / lvm on that VG. 6 lvm is using by HPSG cluster and another two used for local moutn point but volumegroup is same.

Regards,

Gourv

[Mike_Chisholm]

What you are trying to do will not work. At package start/stop time, Serviceguard will attempt to activate and deactivate the VG. If you have lvols in the VG that are mounted outside of the SG package, when the package halts, SG will not attempt to umount those filesystems. Subsequently, the vgchange -a n <VG> will fail because there are lvols still mounted on filesystems and Serviceguard does not know how to unmount these. It could also cause problem with LVM activation protection and SCSI persistent reservation as these operations are tied to packages and affect the VG's associated with the package at the VG and PV layers.

There might be a little more inforamtion in the package log for the package assocated with the VG in question but certainly nothing about mount points SG knows nothing about. I am not sure you will find anything explictly stating what you are trying to do is unsupported but it is. If you review how SG works carefully I suspect you would arrive at the same conclusion. I would start here http://h20628.www2.hp.com/km-ext/kmcsdirect/emr_na-a00039044en_us-1.pdf

Specifically I would review the Volume Manager Planning section on page 103.