Comment #1 on issue 150 by [email protected]: Implement auto-failover if the cluster suports OOB
http://code.google.com/p/ganeti/issues/detail?id=150
I haven't tried this yet, but if I'm ever convinced that failing node is really down (and, say, it is, I know by clear response from hardware) I would manage cluster state (failover nodes) with monitoring tools.
Of course its a neat idea to have it right inside ganeti and I would really vote for it.
