Faults can be gathered by enabling SNMP and installing some opensource tool to catch and poll SNMP. When using for example Zabbix one can also put agent running on host to catch any other fault. Here is some initial list of high level faults and how they can be caught. List assumes that one enables usage of SNMP and then would use tool like Zabbix. There is also Pacemaker mentioned if used. Usage of that is limited to number of nodes, so it works better only for controller nodes.

Many of the faults needs to be configurable while others not. Hardware faults especially might need different triggers in different HW while some Openstack internal fault will always be caught the same way.