RHEL5 Cluster help

epoh · 02-17-2011, 10:25 AM

I've finally managed to get my 2-node cluster to start up (horaay!). But then it wouldn't see the storage properly on both nodes. It looked like one node was maybe not seeing the SAN properly, so I rebooted it. After it came up I restart cman on both boxes and the other node got powered off! I feel like something must still not be configured properly. What I want is a 2-node cluster where one box is the active box and the other box is for fail-over. Here's my cluster.conf (hostnames/IPs marked out):

Quote:

<?xml version="1.0"?>
<cluster alias="netweaver" config_version="14" name="netweaver">
<fence_daemon clean_start="0" post_fail_delay="0" post_join_delay="20"/>
<clusternodes>
<clusternode name="NODE1" nodeid="1" votes="1">
<fence>
<method name="1">
<device name="NODE1"/>
</method>
</fence>
</clusternode>
<clusternode name="NODE2" nodeid="2" votes="1">
<fence>
<method name="1">
<device name="NODE2"/>
</method>
</fence>
</clusternode>
</clusternodes>
<cman expected_votes="3">
<multicast addr="239.192.6.227"/>
</cman>
<fencedevices>
<fencedevice agent="fence_drac5" ipaddr="10.##.##.##" login="root" name="NODE1" passwd="calvin" power_wait="180" secure="1"/>
<fencedevice agent="fence_drac5" ipaddr="10.##.##.##" login="root" name="NODE2" passwd="calvin" power_wait="180" secure="1"/>
</fencedevices>
<rm>
<failoverdomains/>
<resources/>
</rm>
<quorumd interval="1" label="appl" min_score="1" tko="5" votes="1">
<heuristic interval="2" program="ping -c3 -t2 10.##.##.1" score="1"/>
</quorumd>
</cluster>

What am I missing here?

epoh · 02-17-2011, 10:54 AM

So, now the cluster won't start. Arg! Now I'm seeing this in /var/log/messages:

Quote:

Feb 17 10:56:41 usxsl227 luci[4955]: Unable to retrieve batch 1917718398 status from usxsl228:11111: module scheduled for execution
Feb 17 10:56:41 usxsl227 luci[4955]: Unable to retrieve batch 1647498095 status from usxsl227:11111: module scheduled for execution
Feb 17 10:56:41 usxsl227 openais[5946]: [TOTEM] Retransmit List: 2
Feb 17 10:56:44 usxsl227 last message repeated 7 times
Feb 17 10:56:44 usxsl227 ccsd[5940]: Cluster is not quorate. Refusing connection.
Feb 17 10:56:44 usxsl227 ccsd[5940]: Error while processing connect: Connection refused

trickykid · 02-18-2011, 10:11 AM

What is your output of clustat?

When I've encountered this error, it was either issues with the fencing or multicast, with the fencing I have better luck creating a fence_xvm.key file for the communication between the two and fencing.

Here's a working cluster.conf file I recently created for a two node cluster for a GFS setup on two xen virtual machines:

Code:

<?xml version="1.0"?>
<cluster name="app_cluster" config_version="19">
  <cman two_node="0" expected_votes="1">
        <multicast addr="239.0.0.1" interface="eth0"/>
  </cman>
  <clusternodes>
        <clusternode name="nfs1.pinnacle.local" votes="1" nodeid="1">
          <multicast addr="239.0.0.1" interface="eth0"/>
                <fence>
                <method name="1">
                <device name="xvm" domain="host1.pinnacle.local"/>
                </method>
                </fence>
        </clusternode>
        <clusternode name="nfs2.pinnacle.local" votes="1" nodeid="2">
          <multicast addr="239.0.0.1" interface="eth0"/>
                <fence>
                <method name="1">
                <device name="xvm" domain="host2.pinnacle.local"/>
                </method>
                </fence>
        </clusternode>
  </clusternodes>
  <quorumd interval="3" tko="23" votes="1" label="appquorum"/>
  <fencedevices>
        <fencedevice name="xvm" agent="fence_xvm"/>
  </fencedevices>
  <rm/>
</cluster>

One thing that does stand out to me is within your cman tags, you might try adding two_node="0" to it.