why the cluster can't be formed until another nodes is up ?

[Hanry Zhou]

I have 2-nodes cluster. Now, I reboot these two nodes in the sequency of 10 mis apart.

Node A is back to service first, but why the cluster can't be formed, or started until the Node B is coming back to service?

I would expect, the node would form the cluster first by itself, and then when the Node B is back to service, it will join the cluster.

thanks!

[Devender Khatana]

Hi,

The purpose of sevice guard is to avoid single-point failure and in the event of one node missing you are allready have something failed. So there will be no point in providing the option to start it normally when one machine is missing.

Allthough you can form it manually even with single node by specifyinh the nodename. Try this

#cmruncl -n nodename

HTH,
Devender

[Sanjay_6 · Accepted solution]

Hi,

At the time of cluster formation after a whole cluster reboot, you need to have more than 50% of the node available to form the cluster. Otherwise it will timeout and you'll have to manually form the cluster with the available number of nodes.

the command should be,

cmruncl -f -n node1

where node1 is the node that is available for cluster formation after the reboot of a two node cluster.

In a 2 node cluster, you must have both the nodes available to start the cluster automatically after reboot. The requirement is (n/2)+1 where n is the number of nodes in the cluster.

Hope this helps.

Regds

[melvyn burnard]

The basic design behind Serviceguard REQUIRES 100% of the cluster nodes be available to form a cluster at initial cluster start time.
The default time allowed for al nodes to become available is 10 minutes.
There is a configurable parameter in the cluster ascii file that allows you to modify this.

[Steven E. Protter]

Shalom Hanry,

It is possible to configure SG to a single node cluster.

I've seen it done at work.

It is a lot of work and defeats the purpose of SG.

The explanations above need no duplication.

Your cluster is working correctly.

SEP

[Patrick Wallek]

If it is a requirement that you reboot your machines automatically, you would be better off increasing the time difference between the reboots.

We have an MC/SG cluster where we reboot the machines weekly. The reboots are schedules to start 30 minutes apart. So far (knock on wood) we have not had any trouble with the cluster coming up.