Fencing in VMware virtualized Pacemaker nodes

Posted on Fri 18 May 2012 in hints-and-kinks • 2 min read

For users of VMware virtualization, it’s becoming increasingly common to deploy Pacemaker clusters within the virtual infrastructure. Doing this requires that you set up fencing via ESX Server or, more commonly, vCenter. Here’s how to do that.

The cluster-glue package contains node Pacemaker’s fencing (STONITH) plugins, one of which is the external/vcenter plugin. It enables Pacemaker to interface with an ESX Server host or vCenter server. When a Pacemaker node needs to be fenced, the fencing node contacts the vCenter host and instructs it to knock out the offending node.

For this to work, your configuration needs to satisfy a couple of prerequisites:

  • Your setup needs a reasonably recent cluster-glue package (the one that ships in Debian squeeze-backports and Ubuntu precise is fine).

  • You need to install the vSphere Web Services SDK on your nodes. This itself has a number of Perl prerequisites. On Debian/Ubuntu systems, you should be able to install them with:

aptitude install libarchive-zip-perl libcrypt-ssleay-perl \
  libclass-methodmaker-perl libuuid-perl \
  libsoap-lite-perl libxml-libxml-perl

Now, create a set of vCenter credentials with the credstore_admin.pl utility that comes bundled with the SDK:

/usr/lib/vmware-vcli/apps/general/credstore_admin.pl \
  -s <vCenter server IP or hostname> \
  -u <vCenter username> \
  -p <vCenter password>

This creates a credentials file in .vmware/credstore/vicredentials.xml relative to your home directory. Copy this file into a location where Pacemaker can find it, say /etc/vicredentials.xml, and make sure it gets 0600 permissions. Also, remember to copy it to all your cluster nodes. Once your credentials are properly set up, you can test the STONITH agent’s functionality by invoking it directly, like so:

  VI_SERVER=<vCenter server IP or hostname> \
  VI_CREDSTORE=/etc/vicredentials.xml \
  HOSTLIST="<pacemaker hostname>=<vCenter virtual machine name>" \
  RESETPOWERON=0 \
  /usr/lib/stonith/plugins/external/vcenter gethosts

is the name of one of your cluster nodes as per uname -n, and is the corresponding machine name in your vCenter inventory. If everything is working fine, the gethosts command should return the Pacemaker hostname again.

Now, on to adding this to the Pacemaker configuration. The example below is for two hosts named alice and bob, which in the inventory happen to be listed by their FQDN in the example.com domain:

primitive p_fence_alice stonith:external/vcenter \
  params VI_SERVER="vcenter.example.com" \
    VI_CREDSTORE="/etc/vicredentials.xml" \
    HOSTLIST="alice=alice.example.com" \
    RESETPOWERON="0" \
    pcmk_host_check="static-list" \
    pcmk_host_list="alice" \
  op monitor interval="60"
primitive p_fence_bob stonith:external/vcenter \
  params VI_SERVER="vcenter.example.com" \
    VI_CREDSTORE="/etc/vicredentials.xml" \
    HOSTLIST="bob=bob.example.com" \
    RESETPOWERON="0" \
    pcmk_host_check="static-list" \
    pcmk_host_list="bob" \
  op monitor interval="60"
location l_fence_alice p_fence_alice -inf: alice
location l_fence_bob p_fence_bob -inf: bob
property stonith-enabled="true"

At this point you should be able to test fencing with stonith_admin -F or crm node fence. Or simulate a node problem with killall -9 corosync.

Special thanks for this goes to Nhan Ngo Dinh both for writing the plugin in the first place, and for providing an excellent and straightforward README file for it.

This article originally appeared on the hastexo.com website (now defunct).

Pacemaker