Comment 4 for bug 1484866

I think this is caused by some timing issues in the setup of the cluster; specifically:

    if True in [ra.startswith('ocf:openstack')
                for ra in resources.itervalues()]:
          apt_install('openstack-resource-agents')

does not happen until the hacluster charm parses relation data from the principle, at which point it might be installed and part of a corosync cluster with its peers; the cluster configuration is propagated from the lead unit, which already has things installed, and we end up with an ocf failure of this type.

I'd actually suggest we move any resource agent install to much earlier in the charm lifecycle, thus avoiding an inconsistent on-disk state with the actual cluster source configuration.