ServiceGuard 11.14 forced up.

[Lars Mousten_1]

Hi,

I have a five node campus cluster on 2 sites, which I want to extend to a seven node cluster.

Now, because of some other work, I want to run the cluster on the secondary site with only two servers, which means that I have to force the cluster up, otherwise there is not quorum in the cluster. At a later stage I will add the three exsisting servers to the running cluster. After some hours again, I will online add my two new servers (one at each site), to the five node running cluster. I will add one at a time in order to keep the binary files uptodate.
Now.. my question. Will the cluster know that its forced up, and will it complain when I try to add two completely new servers to the cluster, when it was forced up ???

Thanks in advance....
Please comment

/Lars

[melvyn burnard]

So if I understand correctly, you will halt the cluster, then do cmruncl -n node1 -n node2 to get the cluster up just as a two node cluster?
Why not just cmhaltnode the three nodes on the site that you want to not be in the cluster, leaving the two other nodes running as the cluster?


Whether you do start the cluster up as a 2 node cluster,or just halt the other three nodes, once you restart the other three exisiting nodes, they will just rejoin an already running cluster.
When you come to add in the new nodes, there should be no problems.

[Lars Mousten_1]

Hi,

Thanks.
well, the three servers I want to take out of the cluster are running all the cluster packages (6 in total). And the six packages are going to be moved to the remaining 2 servers.
I just read that all servers needs to be up and running and member of the cluster before a new server could be online added to the cluster... .... Originall I thought about adding my two new servers when the cluster was only running on two of the servers, but apprently thats not possible ...

(My job is to recable all the servers .... with as little outage as possible.... and add two new servers to the cluster)

[Steve Lewis]

This situation can even be generalised.

When extending a cluster of any size it often makes sense to pre-configure the new nodes as far as possible outside of the cluster and have them join the running cluster, then move the packages onto the new nodes instead of re-building the whole thing from scratch. Its a slightly different approach which can be used even in initial cluster configuration.

[melvyn burnard]

So it sounds like you should do (for each package)
cmhaltpkg
cmrunpkg -n

Then cmhaltnode each of the three nodes, and do
cmmodpkg -e so that it is able to switch between the two remaining nodes.
Do your work at the site where the three nodes reside, and then restart the three nodes back into the cluster either with a reboot, or by doing cmrunnode

Check all isok using cmviewcl -v.
If necessary, move the packages back to their primary nodes.
Then you can add in the two new nodes.

One caveat here: As you have a 5 node cluster, you obviously do not have a cluster lock disk. Unless you are using Quorum Server for this cluster, the two node cluster runs a risk of possibly having one of the two nodes crashing (for what ever reason) and the surviving node will TOC as there is no cluster locking mechanism.