Problems with service guard

[catdido]

I have 2 servers in cluster with HP-UX version 11i.
The problem is that the raw devices are not mounted.
The vg00 as the raw devices in the file fstab and in the control file of the cluster.
The raw devices are specified in also in control file of the cluster.
Raw devices are commented in fstab (#).
The question is who mount the raw devices fstab or control file of the cluster.
The other file systems are mounted only by fstab.
The cluster is not working.

HELP.............

[John Poff]

Hi,

I've never used raw devices before, but I would assume that you would want them mounted by MC/SG in your control script. Do you get any errors in the log file for your package when you try to start it?

JP

[Sridhar Bhaskarla]

Hi,

What do you mean by raw devices?. If you are talking about using raw logical volumes, you don't need to specify them in the control file. They are not mounted file systems and are accessed directly from the application. You will be specifying only the volume groups containing the raw logical volumes that get activated and deactivated. Your application shutdown scripts need to make sure that they free up the raw logical volumes before serviceguard deactivates them.

-Sri

[Frederic Sevestre]

Hello,

I am not sure to understand what you mean.
You can't share the vg00 of your server. Your raw devices sould be on a shared vg so each server can activate it.

Regards,
Fr??d??ric

[MANOJ SRIVASTAVA]

Hi Catido


I assume that you are using Oracle/Ingress with non cooked files systems , now these will never be mounted howver the control comes from the device files only unde /dev/vg00 , so you cannot mount in the control script , except you can jsut activate the volume group after the switchover takes place.
Also make sure the /dev/vg01/rlvol* is owned by the oracle or the databse so that it can write on it.

Manoj Srivastava

[Rajeev Shukla]

See filesystems local to the cluster node are mounted by fstab specific to the system and rest all filesystems and voulme groups are specified in the respective package scripts. The package script activates volume group in exclusive mode and that opens all Raw devices to be used also the filesystems specific to the packages which need to be mounted at package startup are specified in package script.

Rajeev

[Ashwani Kashyap]

I am not sure what you mean by raw devices . Are they logical volumes or filesystems .

VG00 should never be in the control script . It should be in fstab as it is local to the node and not shared . And also its is filesystems and not raw logical volumes .

Anything that is shared between the two nodes goes in the control script whether filesystems or raw volumes . If its filesystems you need to specify the mount points for each logical volume in the control script file and also the volume groups in the cluster ascii file as they need to be cluster aware .

If you use raw logical volumes you don't have to specify ay mount points in control script as they are not mounted . You just need to specify the cluster aware raw volume groups in the cluster ascii file . The application itself opens the raw logical volumes for read and write . In this case make sure that the logical volumes /dev/vgxx/rlvxx and /dev/vgxx/lvxx is owned by the application opening it for eg oracle:dba .

[Tim D Fulford]

I suggest you give a quick resume of
o What your fstab looks like
o What filesystems are mounted where.
o What raw vgs & lvs are used

I regularly use raw devices in SG for my database (Informix). IF you do not stop the database in the customer_halt_scripts (/etc/cmclustr//.sh) section the volume group will not deactivate (vgchange -a n vgxx).

1 - Check that the SG vg's are deactivated properly on the relavant node (vgdisplay)
If say vgdat is active then simply run fuser /dev/vg/r
This should say what process is locking/using it.
2 - You MUST NOT use vg00 as part of the cluster, It could contain the binaries or anything that is relatively static (like the OS!!!) but it will need to be kept in sync on all nodes. DO NOT USE IT FOR data, logical loga, redologs etc.

Regards

Tim