Start a two node cluster with only the primary node available

[mogiro]

Hi,
We have a two node cluster and we are going to do a HW move from this servers to brand new SD2. The question is that we need to switch off the secondary node, but last time we tried it, we were not able to start the cluster with only one node. Is there any way to start the cluster only with the primary node ignoring and without checking the secondary node? This is a temporary situation, so we don´t want to waste time changing the cluster configuration if it is not absolutely necessary.
Thanks!!

[Jorge Pons]

Hi

cmhaltnode "node to shutdown"
in /etc/rc.config.d/cmcluster (node to shutdown)

AUTOSTART_CMCLD=0

Regards, Jorge

[mogiro]

I think this is not wat I need. I explain you exactly the situation:

- cluster up with two node
- cluster down and machined shutted down due to a electrical intervention.
- primary node machine up, but secondary node not switched on (so unreachable).
- start cluster with only the primary node available.
Is it possible? because last time we tried it, it says that the cluster were not able to start up with only one node.

Hope this clarifies the question, sorry for my first explanation, maybe it was not really clear.
Thanks!!

[Jorge Pons]

Hi

Ok. I must see if its possible.
In this case, I will do (If its possible)
- shutdown cluster and machines
- electrical intervention
- start machines and cluster
- cmhaltnode node_secondary
- edit /etc/rc.config.d/cmcluster and change AUTOSTART_CMCLD to 0 to prevent start cluster service.

There is a documentation:
http://bizsupport2.austin.hp.com/bc/docs/support/SupportManual/c02056214/c02056214.pdf

page 359 (Single operation cluster) to see more information

I try to investigate.

Regards, Jorge

[Jorge Pons]

Hi

Im not sure if this works:
- edit conf file to prevent start cluster (both nodes)
- after electrical jobs
- start only primary node
- start manually cluster (cmruncl -n node_primary)

Regards, Jorge

[Manix]

I guess this is pretty simple .-)

let the cluster be up & running

1) Stop all the packages on node one ( where you need the maintenance )
cmhaltpkg packagex

2) Stop the services on node 1
cmhaltnode node1

3) Start package one node2
cmrunpkg -n node2 packageX

4) Node 2 is out of cluster now.

5 ) modify the file
vi /etc/rc.config.d/cmcluster
AUTOSTART_CMCLD=0

so that it may not rejoin automatically after reboot.

6 ) do the maintenance & run "cmrunnode noede2"

Hope this helps.

Thanks & Regards

Manix
Now every thing is one

[Duncan Edmonstone · Accepted solution]

Wow, some confusing posts above!, some are just typos, but some stuff is just plain wrong!

I think one guy hit on the answer for you though... The reason a cluster won't start if one node is not present, is beacuase at intial cluster formation all nodes have to be present (otherwise there might already be a cluster running on node(s) that are unreachable. You can override this with the "-n" option on cmruncl

cmruncl -n

will start the cluster on a subset of nodes that would normally be required to form a cluster.

THis will prompt you to confirm you really want to do this... if you don't want the prompt then:

cmruncl -f -n

will do it without a prompt.

HTH

Duncan

[mogiro]

I thinks this last solution is the one I need. Let me let the post opened till I try this solution in orden to let you know if it works or not. Maybe it can be usefull for another people.
Thanks!!

[Manix]

If we want to bring down the entire cluster then "cmruncl -n " is needed to do explicit seeding of a particular cluster node.

But if cluster is already running as we needs
things up on at least one, we may switch every thing on other and do the needed stuff & then
let it join the cluster back.

It`s as per requirement.