We deploy on our sandboxes on OpenStack machines with 4 GB of RAM and no swap
enabled by default, which results in memory pressure already during the
installation process, so we need to have the swapfile role first.

edxlocal is too generic and does a lot of things.  This is the beginning
of separating it into roles that are more single purpose and easier to
understand.

ECOM-2185

This fixes a startup error with the "certs" role: upon restart, the
"certs" supervisord service tries to contact the service through the
public port (which should be) served by NginX and fails.

The nginx configuration was set to pass X-Forwarded-For and related headers
unodified to backend services.
This works fine when nginx is deployed behind another load balancer
(for example ELB), which already properly sets the X-Forwarded-* for headers.

But in simpler deployment scenarios, where nginx is directly facing the end user,
nginx should discard any existing X-Forwarded-For headr and set it to remote IP of the request,
the X-Forwarded-Port header to the nginx server port, and the X-Forwarded-Proto
to the scheme of the current request.

The 'NGINX_SET_X_FORWARDED_HEADERS' is set to False by default, but set to True
in the edx_sandbox playbook, where an additional load balancer in front of nginx
is not usually present.

removing sync'd folder for /vagrant

change identity file default to none

removing check for git identity

typo

typo

adding python-yaml for dumping vars

certs settings changes for local disk generation

adding var dump var

adding template for certs

adding certs to default sites

adding 400 condition to the certs nginx template

updating update script for certs, removing discern

adding cert to help

giving www-data write permission to the web root

typo

adding download url

fixing web root

adding certs download url

adding more config for cert template dir, verify url

moving extra vars into the vagrantfile

adding verify url to vagrantfile

syntax

look for valid.html

adding ownertrust export

adding the trust import

removing skip-tags for reconfigure

upgrading to latest released ansible

adding carriage return to owner trust

setting cert urls in extra vars

moving extra vars to vagrantfile

adding new accept_hostkey parameter to all git calls

set key_file

changing dir for userknownhostsfile

adding more verbosity

changing knownhostsfile in environment

don't use userknownhostsfile

setting accept_hostkey back to yes

adding .ssh dirs for the 1.5 git module

adding certs nginx site

Because the CMS uses the git hash of the edx-platform for the
static path it can be problematic if you update the edx-platform
repo for the worker AFTER you update the edx-platform repo
for edxapp.  When running the edxapp role in worker mode
rake gather_assets is skipped.

This is a workaround but we may want to consider using separate
repos for worker and edxapp for situations where both are installed
on the same server.

Some playbooks are missing the new forum nginx reverse proxy configuration
file, which prevents the LMS to access the forum service.

Also renaming the deploy_* plays to the play name

services if it changes
deploy tasks no longer stop the service and will notify the restart
if any task changes

of rbenv and virtualenv roles

* vagrant-shortstack
  - lms / studio / forums
* vagrant-shortstack-xml
  - lms / studio / forums (xml primary)
* vagrant-fullstack
  - everything that we think is working

Swap lms and lms-xml nginx ports and remove edxapp_sandbox.yml from vars_files because it no longer exists.