On Fri, Oct 4, 2013 at 10:52 AM, René Koch (ovido) wrote:
Hosts only get fenced when they aren't reachable, not during tests and also not if you confirm changes to power management settings - or at least I've never seen an issue with this.
Ok. Thanks for clarifying this
You can test fence_ilo2 from command line (if you try a reboot from command line I suggest to set the host into maintenance mode in oVirt as otherwise oVirt tries to fence the host too and ILO/IPMI interfaces sometimes show strange behavior when resetting it twice): # /usr/sbin/fence_ilo2 --help
These same blades was configured in the past with RHCS and RHEL 5.x I've already used fence_ilo2, fence_drac5 and fence_ipmilan in both RHEL/CentOS 5.x and 6.y I remember in the past the logic changed regarding fencing So sometimes there was this workflow depending on agent and its actions' support: poweroff status poweron And sometimes there were problems in timing responses so that the node remained in power off state. I also opened a bug for this with an improved suggestion, I don't remeber right now some years ago for the python script So the question is about how gui settings translates into fence_ilo2 generated command so that I can reproduce it Thanks for the suggestion about putting host into maintenance
If ilo fencing isn't working as expected (e.g. host gets powered off instead of rebooted even if action is set to reboot you could try if fence_ipmilan does the trick. You could also check if you're using the latest BIOS version...
Eventually I'll try it. Blades bios is the latest available as they don't receive updates any more. For iLO2 fw I think they are on 2.x version even if the better release probabbly was around 1.80 I'll check also this Gianluca