[Linux-cluster] Two-Node Cluster Problem

[Marco Nietz]

Hi,

i have a Problem running a Two-Node Cluster (without qudisk) on RHEL 5.3
with all the latest Patches installed. For testing i use a very simple
setup with only one ip-address as the service.

Both nodes have two network-interfaces, both are configured in cluster.conf

<clusternodes>
   <clusternode name="10.102.10.51" nodeid="1" votes="1">
      <altname name="10.209.170.51"/>
         <fence>
            <method name="1">
               <device name="ipsdb01.drac"/>
            </method>
         </fence>
   </clusternode>
   <clusternode name="10.102.10.28" nodeid="2" votes="1">
      <altname name="10.209.170.28"/>
         <fence>
            <method name="1">
               <device name="ips08.drac"/>
            </method>
         </fence>
</clusternode>

When the service runs on node b and i take down the first interface
(10.209.170.28) with: 'ip link set eth1 down' the service is taken over
by node a within a few seconds. When i then disable the second interface
on node b i would expect that node a recognize the failure and fence
down node b but this does not happen. This only occurs in this one
direction, the other way - node a holds the service and need to be
fenced by node b works fine.

What could cause such a behaviour ?

Here are some more sniplets which could be interessting from my cluster.conf

<fence_daemon clean_start="1" post_fail_delay="10" post_join_delay="30"/>

<cman expected_votes="1" two_node="1">

Most of the configuration was done using conga (luci,ricci)


Regards
Marco