Bug 1319070

> Description of problem:

Fencing related names in cluster.conf like those in <method>, <device> and <fencedevice> tags can cause pcs to crash on adding a node to cluster if the cluster configuration has been created or modified outside pcs.

The reason is that pcs creates the config file with "pcmk-method" and "pcmk-redirect" used as a name identifier and presumes it is always there when adding additional nodes. If the configuration however has been created or been changed manually to include custom identifiers, the node add procedure will fail.

Example cluster.conf:

<cluster config_version="4" name="STSRHTS30477">
  <cman/>
  <totem token="3000"/>
  <fence_daemon clean_start="0" post_join_delay="20"/>
  <clusternodes>
    <clusternode name="virt-006" nodeid="1" votes="1">
      <fence>
        <method name="mymethod">
          <device name="mypcmk" port="virt-006"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="virt-007" nodeid="2" votes="1">
      <fence>
        <method name="mymethod">
          <device name="mypcmk" port="virt-007"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="virt-008" nodeid="3" votes="1">
      <fence>
        <method name="mymethod">
          <device name="mypcmk" port="virt-008"/>
        </method>
      </fence>
    </clusternode>
  </clusternodes>
  <fencedevices>
    <fencedevice agent="fence_pcmk" name="mypcmk"/>
  </fencedevices>
</cluster>

Now when I try to add a node:

[root@virt-007 ententyky]# pcs cluster node add virt-009
Error: unable to add virt-009 on virt-006 - Error connecting to virt-006 - (HTTP error: 400)
Error: unable to add virt-009 on virt-007 - Error connecting to virt-007 - (HTTP error: 400)
Error: unable to add virt-009 on virt-008 - Error connecting to virt-008 - (HTTP error: 400)
Error: Unable to update any nodes

An incomplete intersection with missing device details and # of votes is added into cluster.conf:

    ...
    <clusternode name="virt-009" nodeid="4">
      <fence>
        <method name="pcmk-method"/>
      </fence>
    </clusternode>
    ...

This is probably because pcs runs ccs internally to do the job, but ccs fails, having pcs silently ignoring the error and failing later with HTTP 400. This is what pcs runs in the background:

[root@virt-007 ~]# ccs -i -f /etc/cluster/cluster.conf --addfenceinst "pcmk-redirect" virt-009 "pcmk-method"  "port=virt-009"
Fence device 'pcmk-redirect' not found.
[root@virt-017 ententyky]# echo $?
1



> Version-Release number of selected component (if applicable):

pcs-0.9.148-5.el6.x86_64



> Additional info:

We could either:
 - read cluster.conf beforehand to figure out what name has been used for the pcmk fencing device and use that one automatically
 - read cluster.conf beforehand and add our own secondary device if not present under the expected name
 - read cluster.conf beforehand and error out properly