We currently always use c3.large instances in our acceptance tests, which is the older Amazon platform. Beaker as it sits now can't launch the c4.large instances, so the first step here is making the necessary updates to support launching that kind of instance. While making that update we should also make sure we can launch t2.medium instances. Once we have proven we can do that, we should switch our acceptance test configs to use c4.large on the PuppetDB instances and t2.medium on the agent instances. The switch to medium for the agent side could save us money on our testing as the agents don't need all of that capacity.
If we build the new images via packer, note that our current cleanup scripts don't appear to call the PE uninstaller, which has caused trouble (PDB-1431).
Ryan Senior there is currently a bug in the upstream image. We can either wait until that is resolved, or use c4.large. Given that we need to get Xenial in for couch, I decided the most prudent solution was to test on c4.large. Once the upstream bug is resolved, I have no issue switching back to c3.large