- Just add PyYaml as required dependency - doesn't seem problematic
under Windows.
- Update setup.py to conditionally install nix bash script or Windows
bat script depending on platform.
- Outline some work on bootstrapping Pulsar install via PyPI (should be
merged into scripts/config.py at some point).
- Version update to PyPI publication tests.

 - Add pulsar-config script that sets up a directory with server.ini
   file describing web process and app.yml describing Pulsar
   configuration..
 - Move run.sh into scripts and rename to just 'pulsar', modify behavior
   to target configuration files in current working directory or directory
   specified by -c/--config or PULSAR_CONFIG_DIR.
 - Add a new mode `--mode webless` to pulsar/run.sh that acts as an
   entry point into pulsar.main and starts pulsar without a webserver
   for MQ operation.
 - Move test/check.py into pulsar/clint/test/check.py so it is installed
   with pip. Add entry point so 'pulsar-check' is available after installing
   pulsar via pip/setup.py.

Allows run as real user stuff to work in an unmodified Pulsar root if these values are specified.

One possible way to test.

- Setup Mesos cluster (for testing see https://github.com/mesosphere/playa-mesos).
- Install LWR on Mesos cluster.
- Install Mesos client into LWR virtualenv (e.g. easy_install http://downloads.mesosphere.io/master/ubuntu/14.04/mesos-0.19.0_rc2-py2.7-linux-x86_64.egg)
- Configure LWR to talk to a message queue (set message_queue_url in server.ini and ensure pycurl is available).
- Configure Galaxy to submit to this LWR instance via message queue (see notes in job_conf.xml.sample_advanced).
- Run new LWR Mesos framework process passing it host and port Mesos master.

  ./script/mesos_framework --master=10.141.141.10:5050

  Can also pass this command --manager, --ini_path, --app to target a different manager or a "paste" app other than default of main in server.ini.

- Run a Galaxy job.

Limitations: Probably several of these - most obvious to me is these processes are not dying cleanly (because of non-daemon message queue threads I guess) and cancelling jobs in Galaxy has no effect. Need to reconsider how to do this (cli-based submission has same limitation and can probably be fixed at same time). Then of course there is the ever present lack of documentation.

Create script infrastructure for launching LWR jobs from a command-line (either using a base64 encoded json object or a simple file). Fixup client stuff to deal with this properly.

TODO: Documentation, on this and much, much more.