They fail with

  Waiting for system status:               An error occurred building
  AMI for loadtest edx edxapp.  The Exception was IndexError('list index
  out of range',)

I'm guessing that it is this list index lookup, but seems likely.  Hard
to test as it requires AWS slowness.

We only run the previous stanza if git and pip are missing.
So if you have an existing 14.04 image and cut a new one, you'll
still run into problems where ansible dies missing httplib2 and we
never install it without a scratch build.

Our user play uses uri which requires httplib2.

Pass down the region.

The new code in the pipeline looks for the 'cluster' tag not the 'play' tag.  Add both for now,
we are working on a full replacement for abbey.py so this is just a shim until we can get to that.

It was causing 500 internal server errors because after
sanitization it would be set to null but a valid value would be
needed for a couple of actions on the website, like assigning
students to programs and assigning students to courses.

Truncating a hash function is kind of terrible, but this gives us a
unique 30 character string and should avoid conflicts.

! should be understood by django to mean "no password set".  It's handled
better in newer versions, but even in 1.4 it doesn't cause the "login
hangs" errors on stage like '' does.

This setting gets propogated to the launch configuration and is currently
causing unexpected behaviour.

Sometimes we get an instance id back from the creation call but then
get an exception saying the ID was not found when we do a get_all_instances.

Set params on BlockDeviceType instead of BlockDeviceMapping.