This cookbook is now up for adoption! See CONTRIBUTING.md for details.
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, 2.3, 4.4.0, 5.0.1 (the default)
- 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 and kitchen tests for the full list.
This also serves as an example for developing and testing Chef cookbooks. It uses:
ChefDKGiven up on this for now; uses standard RVM and Gemfile to manage installation- chef-client 13.6
- Berkshelf for dependency resolution; 6.3
- Test Kitchen for
comprehensive testing across multiple platforms, with tests written in
serverspec; 1.19
- Docker, with kitchen-docker integration
- Foodcritic for style checking; 12.2
- RSpec (3.7)/Chefspec (7.1) for unit testing
In addition:
- Vagrant to provide an out-of-the-box working example; only tested with 2.0.
If you want to just have a working Anaconda VM, install:
- Vagrant
For the full experience (e.g. running the test suite), also install:
- vagrant-omnibus
- vagrant-berkshelf
- Docker
The sample 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 4.3.30
# or you add it to PATH manually
$> vagrant ssh
$vagrant> export PATH=/opt/anaconda/5.0.1/bin:${PATH}
$vagrant> conda --version
conda 4.3.30
It includes a Jupyter notebook server accessible at http://33.33.33.123:8888. Token authentication is disabled in the quickstart Vagrant setup.
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.
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 for more details.
anaconda
version
: the Anaconda version to install. Valid values are:- 2.2.0
- 2.3.0
- 4.4.0
- 5.0.1
- 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 stringyes
(there are no defaults); any other value will reject the license.owner
: the user who owns the installgroup
: the group who owns the installsystem_path
: adds the bin path to the system's profile.d directory
Include this to have the environment set for all users (login shells) via
/etc/profile.d
. Useful for development.
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 for additional options; in
general, all it does is present the same options as conda install
/conda remove
.
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/jupyter/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 jupyter_user
group jupyter_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
See TESTING.md for details.
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
- 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:: Matt Chu ([email protected])