[Linux-cluster] Re: Linux-cluster Digest, Vol 47, Issue 4

[Lon Hohberger]

On Tue, 2008-03-04 at 15:09 +0530, Balaji wrote:
> Dear All,

>         <clusternodes>
>                 <clusternode name="corviewprimary" votes="1">
>                         <fence/>
>                 </clusternode>
>                 <clusternode name="corviewsecondary" votes="1">
>                         <fence/>
>                 </clusternode>
>         </clusternodes>

Ok, so there's no fencing - this might be more related than the bonding
configuration.

In order for the node to be quorate (and start services) with
two_node="1", the nodes must generally be able to fence each other.

I don't know how this could have happened in your cluster - so there are
two problems:

(a) The nodes don't see each other
(b) Fencing completes somehow; but we don't know how (yet)

So, the next thing to do would be to produce the problem and capture the
output of:

  cman_tool nodes 
  cman_tool services

Also, if you're using RHEL4 (which I assume based on a RHEL4
documentation manual you noted in an earlier email), you can try to
force multicast on a particular address + interface:

        ...
        <clusternodes>
                <clusternode name="corviewprimary"
                             votes="1" nodeid="1">
                        <multicast addr="225.0.0.12" interface="bond0"/>
                        <fence/>
                </clusternode>
                <clusternode name="corviewsecondary"
                             votes="1" nodeid="2">
                        <multicast addr="225.0.0.12" interface="bond0"/>
                        <fence/>
                </clusternode>
        </clusternodes>
        ...


-- Lon