I think it's crucial we get your failover tests integrated with the rest of the clustering tests which start/stop servers.
Killing a server should be pretty trivial.
Starting a new one could be done by running the script from java, take a look at ServerManagement::VMstarter - we should be able to get something like this working.
Next week, when everyone is back from holiday, I would like to spend the week working with Clebert on finishing off the rest of failover, with a target of getting it all done by the end of the work.
I think this is doable.