Remove node & package from the cluster

[CFI-beheer]

Hi Folks,

We have a 7 node hp-ux cluster.
OS version 11.23.
Alle the servers are rx2620

The hp-ux cluster is running 7 packages.
All of the packages contain oracle software + oracle database file except for one, the hpuxdb04 package.
This package contains the hp-ux backup software dataprotector and its internal database.
The data of all the packages resides on virtual disk (luns) which are distributed through a EVA 3100.

Problem Situation:
If for some reason are EVA3100 is unavailable we will also lost our dataprotector software including the database of this software.

To prevent this from happening we are thinking of storing this the dataprotector software + database locally on physical disks on a server(rx2620).

The rx2620 servers have only 3internal slot. Two of the slot are already being used by the disk (36,4 GB )system. They are mirrored and have the hp-ux 11.23 OS installed on them.
I have only one free slot.
I was thinking of putting a 300 GB disk on this free slot en store the dataprotector software in there.
But what will happend when this disk get corrupted?

Could some help me with a best approach for
1 - take out a node from a cluster?
2 - secure the dataprotector software data + its internal database.
( make it independent of the EVA)
3 - get this dataprotector software mirrored?
Any other solution is appreciated.

Thanks,

Kinds Regards

Frank.

[melvyn burnard]

For the Serviceguard side, take a read of teh Managing Serviceguard manual for your version available at htpp://docs.hp.com/en/ha
But the basic steps are:
Removing the package:
1) Halt the package
2) remove the package from the cluster using cmdelete conf
3) delete the package subdirectory on each node in the cluster
4) free up any resources, remove mount points etc from the nodes that this package may have used.

Removing the node:
1) Halt the node using cmhaltnode
2) remove the node from any package configuration file that may have this node listed
3) reapply each package that you change in step 2
4) edit the cluster ascii file, deleting the referenced node
5) apply the change with cmapplyconf
6) if required, removve SG from the node, and the /etc/cmcluster subdirectory

[Rita C Workman]

You could....and Melvyn has given the cluster info. But disk can fail, be it internal or EVA - disk [or servers] can fail.

But here's a couple thoughts.

For your Data Protector, you could look it at as two needs:
1) Need to backup your DP database:
You could copy the database off to tape regularly; or ftp to some other safe disk. Simly by creating a tarfile that contains /var/opt/omni & /etc/opt/omni. One simple tarfile that you could use to restore your whole database in one simple command.
2. Need of DP software:
You could use your ignite tape if your server fails. Just rebuild it from your last ignite tape - and then use your tarfile to get your data to the most recent version of your database.

You can cluster a DP server, but I just keep mine outside the cluster. We use the above approach for Disaster Recovery, and it has worked. It's not fancy, but we found it quite quick to put it back up and start restoring.

Rgrds,
Rita

[CFI-beheer]

Hi Melvyn,Rita,

I have no test environment to test cmdeleteconf command so I want to be certain that this command works.

I want to only delete package pkg4 from thecluster.
The packages pkg1, pkg2,pkg3,pkg7,pkg12 should not be affected.
Also the cluster clusterA should not be affected.
Is this the right commando for doing this.

cmdeleteconf -f -c clusterA -p pkg1

I have read that the cmdeleteconf hangs sometimes. Could this created
problem for my cluster clusterA.

In point 3 of your answer you mention this:
3) reapply each package that you change in step 2

Can I do this without bringing the packages ( pkg1,pkg2,pkg3,pkg7,pkg12 ) down first?

Could a hotswappable 300 GB disk be detected without first bringing the server?
Do I have to do some configuration by EFI shell?

I was thinking of doing this:
Put the 300 GB disk in the free internal disk slot?
Then doing a
ioscan -fnC disk to see the disk.

Thanks,

Kind Regards,
Frank

[Rita C Workman]

Greetings,

I want to only delete package pkg4 from thecluster.
Also the cluster clusterA should not be affected.
Is this the right commando for doing this.

cmdeleteconf -f -c clusterA -p pkg1

>>>> The -f option is only used when you want to FORCE the command. I would run it withouth the -f option first and only force it if you have to.
The cluster will be affected - your removing a package. But you do this with everything up and running. It will update the nodes.
NOTE: You will also want to edit the cluster ascii file to remove reference to pkg associated VG_NAME

Can I do this without bringing the packages ( pkg1,pkg2,pkg3,pkg7,pkg12 ) down first?

>>> Yes you can reapply a package with it up generally. If your changing network items (like changing IP's, etc) you would stop - but for most everything else you can re-apply with things up and running.

Could a hotswappable 300 GB disk be detected without first bringing the server?
Do I have to do some configuration by EFI shell?... doing a
ioscan -fnC disk to see the disk.

>>> If it's hotswappable then try the ioscan and confirm that the hardware is CLAIMED. Run insf -e to build any special device files as needed.

/rcw