Lately, EC2 spot prices have been fluctuating quite a bit which has caused the lustre and ZFS build farms to become unstable. Currently the solution is to manually choose a new zone and restart the buildbot master. The alternate solution is to stop using spot instances.
We can potentially make EC2 slaves a bit smarter and have them get running averages of the price per zone, then try each zone that is below our max bid price. We can even have the EC2 slaves fall back to using on demand instances if our max bid price can't be met or when no suitable zone exists.