8.2.3. Configuring CMAN Fencing

We configure the fence_pcmk agent (supplied with Pacemaker) to redirect any fencing requests from CMAN components (such as dlm_controld) to Pacemaker. Pacemaker's fencing subsystem lets other parts of the stack know that a node has been successfully fenced, thus avoiding the need for it to be fenced again when other subsystems notice the node is gone.

Warning

Configuring real fencing devices in CMAN will result in nodes being fenced multiple times as different parts of the stack notice the node is missing or failed.

The definition should be placed in the fencedevices section and contain:

	
 <fencedevice name="pcmk" agent="fence_pcmk"/>

Each clusternode must be configured to use this “device” by adding a fence method block that lists the node's name as the port.

	
 <fence>
   <method name="pcmk-redirect">
     <device name="pcmk" port="node_name_here"/>
   </method>
 </fence>

Putting everything together, we have:

Example 8.2. cluster.conf for a two-node cluster with fencing

	
<?xml version="1.0"?>
<cluster config_version="1" name="mycluster">
  <logging debug="off"/>
  <clusternodes>
    <clusternode name="pcmk-1" nodeid="1">
      <fence>
        <method name="pcmk-redirect">
          <device name="pcmk" port="pcmk-1"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="pcmk-2" nodeid="2">
      <fence>
        <method name="pcmk-redirect">
          <device name="pcmk" port="pcmk-2"/>
        </method>
      </fence>
    </clusternode>
  </clusternodes>
  <fencedevices>
    <fencedevice name="pcmk" agent="fence_pcmk"/>
  </fencedevices>
</cluster>