Can I Disable Cluster on Superdome?

[kash_1]

Hi,
we have Two 32 way SuperDom servers with Manufacturer related HA/Cluster SW.
We are running Oracle Database on the cluster..
My Quoestion is:
Can we Disable the Cluster while the Oracle is running so when we shutdown Oracle DB it will not failover to the other node?
we would like to keep the File Systems mounted on the same node, if we shutdown cluster it will unmount all shared files systems.
----
the only solution we found is to shutdown the cluster then shutdown oracle database... then manually mount back all files systems.
Please help.

[Steven E. Protter]

Shalom

cmhaltnode on the failover node

SEP

[Sundar_7]

Kash,

The answer "depends" on how the Oracle monitoring is configured for your package.

Unlike VCS, MC-Serviceguard doesnt have a "freeze" option.

You will need to find out the how Oracle is monitored and disable the monitoring. Then you can shutdown Oracle and the package will not failover to the adoptive node.

Just cmhalting the failover node will not prevent the package from halting on the primary node if you do not stop the monitoring.

Thanks,

Sundar.

[melvyn burnard]

if you are using the Oracle Toolkit from the ECMT, then this has an option to tell it to use the maintenance flag.
By using this, yuou can simply use the touch command to create a "flag" file, then halt Oracle to do any maintenance work an dthe monitoring is put into "ignore" mode, i.e. it does not halt the package because Oracle processses are no longer running.

[Stephen Doud]

When an application and it's resources are put under control of Serviceguard, a package control script (and application control scripts) are called into play to shut down the package/application when the cluster is halted.
Most customers do not want this to happen if they merely want to back up the Oracle database.

To expand on what Melvyn said, if the systems are using the more recent forms of the Enterprise Cluster Master toolkit, the Oracle templates permit the use of a maintenance flag to disable monitoring of the Oracle processes. This permits the admin to shutdown Oracle temporarily, to perform a backup on the database.
The maintenance flag is called "/oracle.debug".

If, however, you really do want to shut down the cluster and leave the database running, you must modify the package control script such that it does nothing more than shutdown the package Service (the monitor).
The easiest way to do this is to ...
1) create a new package template file (cmmakepkg -s

[kash_1]

Steven, Sundar, melvyn and Stephen,
Thanks very much for you kind response..

if the application running makes a difference, then i would like to add that:
we are running SAP ERP and Oracle Database ..

what other information that can help you to identify the solution so i can reply.

Thanks very much for support.

[Steven E. Protter]

Shalom,

The application makes no difference.

If you want to keep the system running but stop failover you have two routes, the easy way and the hard way.

The easy way is cmhaltnode.

It takes the node out of the cluster thereby perventing failover to it. Assumming all package configuration was correct everything should run on the remaining node, freeing the now halted node for any kind of system maintenance required.

The hard way would be to reconfigure the cluster to have all packages run on one node. That really isn't much fun.

Welcome to ITRC.

Please read this at your leisure.
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=140137

The thank you is must appreciated.

SEP

[kash_1]

will cmhaltnode affect the mounted file systems (esp. the shared between the two nodes) ?
because i need to shutdown SAP & Oracle Database and still keep all environment and file systems mounted.

[melvyn burnard]

yes. issuing the cmhaltnode command causes that node to halt ALL packages it is running, thereby removing any IP addresses, unmounting any file systems and deactivating any VG's or DG's that may be used for these packages.

[Stephen Doud]

In my previous entry, I described a method that would allow you to keep the package resources available even after performing a cmhaltpkg.
If you attempt to perform a cmhaltnode or cmhaltcl with the shared volume groups active, Serviceguard will not terminate cmlvmd because it knows these specially activated VGs are still open and is responsible for their status. Therefore cmcld (the main Serviceguard daemon) will also not shutdown. Serviceguard cannot relinquish it's responsibility over volume groups activated in shared or exclusive mode (as seen in vgdisplay output). Therefore, it is not viable to attempt to halt Serviceguard when it's volume group resources are still active on the node.