We already do this in the common role but not in bootstrap or abbey roles.

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.

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.

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.

Use SSD backed EBS instead of standard.
Delete the volume upon abbey termination.

At 30 minutes we still hit it sometimes when building from a bare ubuntu box.

Defaulting it to hammer for now.

The instance gets tagged with the repo and hash that was deployed.

Instead, we take any extra_vars that have a suffix of '_version' and
make sure those exist as tags on the AMI. All other vars do not get
saved anywhere.

Turn on instance tagging with abbey.