This is awesome! We now have the ability to fully customize the trade off between cost and interruption while still guaranteeing a base level of capacity/reliability.
This should make it much easier to run properly containerized web apps using (a percentage of) Spot instances without worrying.
In some cases the two minute Spot interruption warning time would be sufficient to launch a new instance and avoid a drop in capacity, but I don't get the impression that AWS handles this for you. I need to look into this some more.
Of course you shouldn't ignore the additional complexity this brings, so be sure to understand, design and test the different scaling and failure scenarios!
