anaconda cookbook

Chef cookbook for installing Continuum Analytic's Anaconda: "completely free Python distribution for large-scale data processing, predictive analytics, and scientific computing". Specifically:

• Anaconda 2.2 or 2.3
  • python2 or python3
  • x86 or x86_64
• Miniconda
  • python2 or python3
  • x86 or x86_64
• Usage tested on Ubuntu, unittested on Debian, CentOS, and RedHat. See [rspec tests](spec/default_spec.rb#L100) and [kitchen tests](.kitchen.yml#L16) for the full list.

This also serves as an example for developing and testing Chef cookbooks. It uses:

• ChefDK; 0.8.1
  • chef-client 12.4.4
  • Berkshelf for dependency resolution; 3.3.0
  • Test Kitchen for comprehensive testing across multiple platforms, with tests written in serverspec; 1.4.2
  • Docker, with kitchen-docker integration
  • Foodcritic for style checking; 5.0.0
• RSpec/Chefspec for rapid testing; 3.3.2

In addition:

• Vagrant to provide an out-of-the-box working example; 1.7.4

Requirements

If you want to just have a working Anaconda VM, install:

• Vagrant

For the full experience (e.g. running the test suite), also install:

• ChefDK
  • vagrant-omnibus
  • vagrant-berkshelf
• Docker
  • Don't forget Docker Machine if you're on OSX; installing this via homebrew is highly recommended.

Quickstart

The sample [Vagrantfile](Vagrantfile) will build you an Anaconda VM with no changes necessary; note it might take a few minutes to download the Anaconda installer itself.

  $> vagrant up --provision
  ...

  # the sample image includes `recipe[anaconda::shell_conveniences]`, which
  # means conda is already in PATH via /etc/profile.d
  $> vagrant ssh
  $vagrant> conda --version
  conda 3.14.1

  # or you add it to PATH manually
  $> vagrant ssh
  $vagrant> export PATH=/opt/anaconda/2.3.0/bin:${PATH}
  $vagrant> conda --version
  conda 3.14.1

It includes a Jupyter (IPython) notebook server accessible at http://33.33.33.123:8888

Lastly, to use it in a cookbook:

  include_recipe 'anaconda::default'

Warning! If you're also using the python cookbook...

You MUST include recipe[anaconda::python_workaround], otherwise subsequent chef runs will fail. See the issue for details.

Usage, recipes, attributes, and resources

The main recipe is anaconda::default. Include it in your runlist, and it will install the package as well as any necessary dependencies.

The following are user-configurable attributes. Check [attributes/default.rb](attributes/default.rb) for more details.

• anaconda
  • version: the Anaconda version to install. Valid values are:
  • 2.2.0
  • 2.3.0
  • latest (for miniconda only)
  • python: which version of Python to install for. Valid values are:
  • python2
  • python3
  • flavor: what architecture the instance is. Valid values are:
  • nil (will autodetect)
  • x86 (32-bit)
  • x86_64 (64-bit)
  • install_type: which Anaconda distribution to install. Valid values are:
  • anaconda
  • miniconda
  • install_root: the parent directory of all anaconda installs. note that individual installs go into #{install_root}/#{version}
  • accept_license: must be explicitly set to the string yes (there are no defaults); any other value will reject the license.
  • owner: the user who owns the install
  • group: the group who owns the install

recipe[anaconda::shell_conveniences]

Include this to have the environment set for all users (login shells) via /etc/profile.d. Useful for development.

resource anaconda_package

You can use the anaconda_package resource to install new packages into the Anaconda environment:

  # I don't know what 'astroid' is, just using it as a sample package
  anaconda_package 'astroid' do
    # the other supported action is `:remove`
    action :install
  end

See the [resource definition](resources/package.rb) for additional options; in general, all it does is present the same options as conda install/conda remove.

resource anaconda_nbservice

This only works with a full Anaconda installation! I.e. the notebook service will not work out-of-the-box if installed with miniconda.

The anaconda_nbservice will run a Jupyter notebook server as a runit service:

  anaconda_nbservice 'notebook-server' do
    # listen on all interfaces; there will be a warning since security is
    # disabled
    ip '*'
    port '8888'

    install_dir '/opt/ipython/server'

    service_action [ :enable, :start ]
  end

The standard configuration should be good enough, but you might need to write your own run service template:

  anaconda_nbservice 'server-with-custom-template' do
    user ipython_user
    group ipython_group

    install_dir install_dir

    template_cookbook 'your_cookbook'
    # note that if your template name is TEMPLATE, then this value should be
    # 'TEMPLATE", but the file should be 'sv-TEMPLATE-run.erb'
    run_template_name 'your_template_name'
    run_template_opts({
      ...
    })

    service_action [ :enable, :start ]
  end

Tests

To run the full test suite:

  # this will take a while, especially the first time
  $> script/cibuild
  ...

  # check the final result; bash return codes: 0 is good, anything else is not
  $> echo $?

• to run just the [chefspecs](spec/default_spec.rb):

  $> rspec

• to run just the test kitchen serverspec [integration tests](test/integration/default/serverspec/default_spec.rb):

  # this is done via docker/kitchen-docker
  # the list of OSes is defined in .kitchen.yml
  $> kitchen verify

  # test a specific OS; `kitchen list`
  $> kitchen verify default-ubuntu-1204

• check for style issues with Foodcritic

  $> foodcritic

Releases and issues

Standard stuff:

• master is the active version in development
• releases are made with Github, and git tag'ed
• issues should be opened in the Github issue tracker

TODO

• add a pre-provision for kitchen tests to avoid redownloading the installer on every test (really slows down the tests)
• https://github.com/poise/python is now deprecated, in favor of https://github.com/poise/poise-python; see if the python workaround is still necessary
  • the supermarket version (https://supermarket.chef.io/cookbooks/python) looks like it also points to the deprecated one

Author

Author:: Matt Chu (matt.chu@gmail.com)