You never want node failover to be transparent or automagic. Doing that is begging to end up in a "split-brain" scenario, and have fun cleaning up from that.

EDIT: Yes I'm simplifying; saying "never" overstates somewhat. If your cluster management tool can account for STONITH, you're probably safe — or at least safer. If you don't know what STONITH is without googling, you shouldn't be playing with multi-master, highly available databases and failover betwixt them.