Currently, we have a few roles that don't apply cleanly on the first puppet run.
We should set up a pontoon project to check that it's the case and as a general sandbox for OS upgrades/etc for serviceops stuff.
I would say we need to support:
- role::configcluster (etcd is a prerequisite for the appservers)
- role::mediawiki::memcached
- role::mediawiki::appserver
- role::mediawiki::jobrunner
I expect the route to be quite bumpy with the appserver roles, so I would suggest we set up memcached first, to get acquainted with pontoon.