MC/SG Cluster Removal

[Cliff Lim Kok Hwee]

Gd Day Forum,

Currently, I have a pair of MC/SG nodes (NodeA n NodeB) running on HP-UX11.11.

NodeA has a packageA and NodeB has a packageB.

Currently both packages packageA n packageB is running on NodeA.

PackageA will mount the required Oracle filesystems and start up Oracle instanceA.

PackageB will mount the required Oracle filesystems and start up Oracle instanceB.

Work to be done: I would like to remove the MC/SG cluster. Free up NodeB for other application while keeping NodeA running both the 2 Oracle instanceA n instanceB.

I have come out a procedure for the above work to be done:

1) Halt both packages packageA and packageB on NodeA

cmhaltpkg -v packageA
cmhaltpkg -v packageB

The Oracle filesystems are residing inside vg02 and vg03.

2) Unmark vg02 and vg03 from cluster and activate vg02 and vg03

vgchange –c n vg02
vgchange –a y vg02

vgchange –c n vg03
vgchange –a y vg03

3) Edit the /etc/fstab to add in the required Oracle Filesystems to be mounted manually rather then with the help of the package.

eg. /dev/vg02/lvol1 /mnt/KCSCCPRD/data01 vxfs rw,suid,largefiles,delaylog,datainlog 0 2

4) mountall and ensure required Oracle filesystems are mounted

5) Started Oracle manually and verify successful

If the above procedure is successful, can I conclude that NodeA is currently a standalone server w/o MC/SG?

6) If the above steps are successful, I will then proceed to remove the MC/SG cluster from NodeA and NodeB totally.

As I am quite new with MC/SG, I am seeking advice on how to proceed with removing MC/SG from the cluster node pair.

Your help is deeply appreciated. thanks/c

[Sridhar Bhaskarla]

Hi Cliff,

Your steps are looking pretty good. Only thing that you need to make sure is - if you are using the 'virtual IP' to connect to the database, you may want to setup a logical interface (For ex., lan0:1) for it so that the clients will continue to connect without any changes. To remove the cluster and it's configuration use 'cmdeleteconf' command.

Save the configuration files that you are using now so that in case if you need to expand the cluster later, it will be very easy.

You can leave the cluster runnign with only one node (single node cluster) as you can have local lan failover with it.

Since you are removing NodeB out of the cluster, ensure that the disks are 'unseen' from that node. Someone may accidentally use them causing issues to the database on the other node.

-Sri

[Rajeev Shukla]

Hi cliff,

To freeup NodeB and let NodeA be in the single cluster, i would do the following things.

On NodeA do -
Edit the cluster.ascii file for package A and Package B and remove the Node B from the configuration (package ascii configuration file).
and then
cmcheckconf -v -P packageA.ascii
cmcheckconf -v -P packageB.ascii

then

cmapplyconf -v -P packageA.ascii
cmapplyconf -v -P packageB.ascii

then edit the cluster configuration file on NodeA.
Remove the cluster lock (comment the lines containig cluster lock and cluster lock disk) and comment the configuration for NodeB in the cluster configuration file.
cmcheckconf -v -C cluster.ascii
then
cmapplyconf -v -C cluster.ascii

this will remove the NodeB from the cluster and the packageA and B will run on NodeA only

Cheers
Rajeev

[Cliff Lim Kok Hwee]

Guys, Thanks for the inputs.

Sri, as suggested by u

To remove the cluster and it's configuration use 'cmdeleteconf' command.

Save the configuration files that you are using now so that in case if you need to expand the cluster later, it will be very easy.

You can leave the cluster runnign with only one node (single node cluster) as you can have local lan failover with it.

Since you are removing NodeB out of the cluster, ensure that the disks are 'unseen' from that node. Someone may accidentally use them causing issues to the database on the other node.

Cliff: I will be changing NodeA to a standalone server and will not be running any cluster (single node cluster).

Will the cmdeleteconf be sufficient enough?

Cliff: I will be removing both the virtual IP belonging to packageA and packageB. As suggested by application team I will alias it to the physical IP of nodeA. Bcos the packageA and packageB name is actually the database names.

Correct me if I am wrong from OS system side, I don't see that will be an issue. But I am checking with the application team will there be any application issue instead.

Cliff: I understand too that I also need to modify the /etc/lvmrc AUTO_VG_ACTIVATE=0 to AUTO_VG_ACTIVATE=1 correct?

Guys any stuffs I might have overlooked?
thanks/cliff

[Sridhar Bhaskarla]

Cliff -

cmdeleteconf is sufficient enough to delete the cluster after you shutdown the cluster. But you will still have the disks and VGs present on the other server. Export all the ServiceGuard VGs on NodeB and configure/detach your disk array not to see the disks to prevent any future accidents.

For the virtual IP, you can configure a 'logical interface'. For ex., if the interface is lan0, then configure lan0:1 with the floating IP. That way you don't have to change anything. Just add entries in 'netconf' file as if it is another network interface. Even aliasing will work if they are using the 'names' instead of IPs.

Yes. If you had AUTO_VG_ACTIVATE set to 0, make it 1.

-Sri

[Cliff Lim Kok Hwee]

Hi Sri,

As your advice,

For the virtual IP, you can configure a 'logical interface'. For ex., if the interface is lan0, then configure lan0:1 with the floating IP. That way you don't have to change anything. Just add entries in 'netconf' file as if it is another network interface. Even aliasing will work if they are using the 'names' instead of IPs.

Cliff Reply: This virtual IP is only applicable if I will be using single node cluster correct me if I am wrong?

If correct, I can do away with the virtual IP since I will be configuring it as standalone db server running 2 oracle instances?

thanks/cliff

[Sridhar Bhaskarla]

HI Cliff,

You can configure the 'logical interfaces' outside serviceguard. Serviceguard uses 'cmmodnet' to add the IPs. Outside of it, you use 'ifconfig'. SAs find them very convenient to configure them with the left-off virtual IPs resulting out of dismantling serviceguard.

-Sri

[Cliff Lim Kok Hwee]

Gd Day Forum,

I need to disable MC/Service Guard cluster and packages from performing an auto restart when I perform a server reboot.

This is to prevent my updated /etc/fstab filesystems having conflicting interest with any auto cluster/packages restart during a server reboot.

Need advise on it.

Thanks/cliff

[Sridhar Bhaskarla]

Hi Cliff,

I believe you are complicating the things a bit. All you have to do is to 'cmdeleteconf' to delete the configuration so that serviceguard will not come into picture.

If you still want to keep the cluster configuration in tact, make CMCLD=0 in your /etc/rc.config.d/cmcluster file. This will disable cluster daemons from starting. You will have to manually run 'cmruncl' to start the cluster. You will need to remember that when you run 'cmruncl', the cluster will be reformed and the packages will automatically start if AUTO_RUN is enabled. However, if you already unmarked the VGs (vgchange -c n), then packages will fail as they will not be able to activate the VGs.

-Sri

[Cliff Lim Kok Hwee]

Hi Forum,

Purpose of disabling both cluster/packages from auto-restarting with a server reboot is to ensure /etc/fstab and oracle can be started clusterless with a server reboot.

I was thinking once the above is achieve then upon 1 day or 2 days monitoring then I will proceed with cmdeleteconf. This is to minimism any contingency recovery that might occur.....

Any views on my approach? cliff