On 11/02/12 01:07, Nathan Stratton wrote: Hi Nathan, Currently OE (oVirt engine) supports only one power management device per server. Livnat

On 02/12/2012 11:18 AM, Nathan Stratton wrote: It is possible to put both power supplies on a single external APC device, however that sort of negates the redundancy of having dual power supplies (to some degree). Ideally you'd want each power supply on a separate circuit and putting them on the same circuit means that both power supplies will fail if the circuit fails. I think that oVirt Engine needs the ability to configure more than one fencing device per server, with logic to determine if either fence device is sufficient to power off the machine, or if both are required. The cluster stack (rgmanager or pacemaker based) in Fedora and other distributions contains this sort of logic. So in those software systems you could create two (or more) fence devices per host with boolean logic like: fence_apc APC1 && fence_apc APC2 This means that in order for fencing to be successful, both devices must be successfully powered off. This would handle a dual power supply situation (dual primary) vs. fence_ipmi || fence_apc This would be used in a situation where the ipmi device is the primary fencing device and if that succeeds, there is no need to issue the fence_apc command. (single primary with backup) So similar logic is probably needed in oVirt. Probably would be appropriate to file an RFE here. IPMI is a protocol/standard that is implemented by many different hardware devices. Which specific hardware device were you using? What made you think it was unreliable? Also note that there are many BMC style fencing devices out there so you're not necessarily restricted to just using the IPMI protocol. Many people use IPMI based fence devices in production HA clusters, and by and large I haven't heard a lot of complaints about them. Perry