One node cluster

[Michael F. Dick]

Hi there,

here is a somewhat odd sounding question..I would like to run a one node MC/SG (yeah, I know, its not really a cluster). The reason behind that is, that I have to move a cluster to a differnt location. I have to move the 2 nodes at different times (1 - 2 mths appart). I am looking for some ideas of how to configure a one node cluster.

Any ideas or suggestions

[Victor BERRIDGE]

Hi,
I dont see how you could hope doing something of the sort since you cannot do any takeover...
but tell us more If one machine is beeing moved at a different location, will this box be used once there?, then 1-months after you wil reconfigure your cluster?
I would simulate a disaster=>force a takeover of node1 on node2 then take node1 on travel and use it without MC-serviguard services running till node2 joins the party again...

Just thoughts
All the best
Victor

[Peggy Fong]

Actually I've done exactly that - configured a 1 node cluster. My reason was to take advantage of the lan failover capability while I still had ServiceGuard. Unfortunately, I cannot write the specifice right now, so if no one else provides them, I'll post later tonight. You can email me at pfong@mwd.dst.ca.us - but I still won't be able to do anything until much later (8 hours or so....) sorry. But it is definitley doable.

[Michael F. Dick]

yup thats the plan. After awhile, we will be rconfiguring this cluster to a 2 node cluster again. I agree, I could run the apps without cluster, but then I'd have to change all volume groups to be not cluster aware, and change 'em back (and we are talking about tons of volume groups). My thoughts behind the 1 node cluster are to introduce as less change as possible to the machine and to the configuration...

Cheers,

Michael

[Peggy Fong · Accepted solution]

Hi Michael

I'm going to assume that each node has a package. You can let me know if that is incorrect.

To configure a single-node cluster (split 2-way into 2 single-node serviceguard systems).

On system 1:
Backup the cluster ascii file.
Edit the file and delete all references to the 2d node. You can use standby lans if you have them. If you are using an internal heartbeat lan that is going to be disconnected then comment it out in the file. You should not need a cluster lock device ( if you do, make sure it on this node).

Copy the cluster ascii file to your second node and edit the file. You will need a cluster name (something you didn't need before) with an IP address. Edit the file for the second system to be its own cluster and node with it's package(s). And again if it has its own standby lan it can be used but if there is an internal heartbeat lan that will be disconnected you need to delete it.

This command you can run online, but to implement you will need downtime:

On the primary node:
cmcheckconf -v -C /etc/cmcluster/{cluster.ascii} -P /etc/cmcluster/pkga/pkgaconf.ascii
Note: Move the binary file to another name prior to doing this command.

On the secondary node:
Not sure but you may need to remake the package. Remember, I'm assuming each system has a package.

Make sure you have the current control file then run:
cmmakepkg -s /etc/cmcluster/pkgb/pkgb_control.sh

Check syntax
cmcheckconf -v -C /etc/cmcluster/mycluster.ascii -P /etc/cmcluster/pkgb/pkgbconf.ascii

Now to convert from 2nodes to one.
If the above all worked then halt the packages and nodes and cluster.

On each node apply the new config file with 1 package each:

nodeA:
cmapplyconf -v -C /etc/cmcluster/mycluster.ascii -P /etc/cmcluster/pkga/pkgaconf.ascii

nodeB:
cmapplyconf -v C /etc/cmcluster/mycluster.ascii -P /etc/cmcluster/pkgb/pkgbconf.ascii

On each system start the cluster:
cmruncl -v
Should bring up the package for each system.

You can try cmrunnode nodename as well.

If you are going to remove the package as well then that involved moving the scripts into /sbin/init.d and a few other things. It is easier to make as few changes as possible, then it will be easier to put it back together.

Sorry for the long post and I probably missed a few things, but I hope you got gist of it and I really hope it helps you out.

Regards,
Peggy

[melvyn burnard]

Michael
If the plan is to move the one node with all packages then the simplest way is to make the required changes to the package files to remove the information regarding the second node, remove the info re the second node in the cluster ascii file and then simply re-apply the binary. This leaves you with a one node cluster. Obviously you back up both the nodes' /etc/cmcluster directories ;-}
Move the node, startit up and check all is ok.
When you move tyhe next node, check all connectivity etc, then add in the node to th ecluster ascci file, cmcheckconf, and then qcmapply, to recreate the 2 node cluster. Or simply shutdown the one node cluster, confirm atht all configs are ok, cabling ip etc, and rewcover the directories from before, and test.

HTH

[Michael F. Dick]

Thanks to you all for the suggestions and ideas.

Cheers,
Michael