A library for creating machines and infrastructures idempotently in Chef.
Latest commit 6d9cd90 Jan 12, 2017 @lamont-granquist lamont-granquist update CHANGELOG
Signed-off-by: Lamont Granquist <lamont@scriptkiddie.org>


Chef Provisioning

Gitter Stories in Ready Status Gem Version

Driver build status:

AWS Azure Docker Fog ssh Vagrant
Status Status Status Status Status Status
Gem Version Gem Version Gem Version Gem Version Gem  Version Gem Version

This library solves the problem of repeatably creating machines and infrastructures in Chef. It has a plugin model that lets you write bootstrappers for your favorite infrastructures, including VirtualBox, EC2, LXC, bare metal, and many more!


These are the primary documents to help learn about using Provisioning and creating Provisioning drivers:


This video explains the basics of chef-provisioning (though provisioners are now called drivers). Slides (more up to date) are here.

Date Blog
2014-12-15 Using Chef Provisioning to Build Chef Server
2014-11-12 Chef Launches Policy-Based Provisioning
2014-11-12 Chef Provisioning: Infrastructure As Code
2014-06-03 machine_batch and parallelization
2014-06-03 Chef Provisioning, Configuration and Drivers
2014-03-04 Chef Metal 0.2: Overview - this is a pretty good overview (though dated).
2013-12-20 Chef Metal Alpha

Try It Out

You can try out Chef Provisioning in many different flavors.


To give it a spin, install Vagrant and VirtualBox and try this from the chef-provisioning/docs/examples directory:

gem install chef-provisioning chef-provisioning-vagrant
export CHEF_DRIVER=vagrant
chef-client -z vagrant_linux.rb simple.rb

This will create two vagrant precise64 linux boxes, "mario" and "luigi1", in ~/machinetest, bootstrapped to an empty runlist. For Windows, you can replace myapp::linux with myapp::windows, but you'll need your own Windows vagrant box to do that (licensing!).


If you have an AWS account, you can spin up a machine there like this:

gem install chef-provisioning chef-provisioning-aws
export CHEF_DRIVER=aws
chef-client -z simple.rb

This will create two linux boxes in the AWS account referenced by your default profile in ~/.aws/config (or your environment variables).


If you are on DigitalOcean and using the tugboat gem, you can do this:

gem install chef-provisioning chef-provisioning-fog
export CHEF_DRIVER=fog:DigitalOcean
chef-client -z simple.rb

If you aren't using the tugboat gem, you can put driver and driver_options into your .chef/knife.rb file.

This will use your tugboat settings to create whatever sort of instance you normally create.

Cleaning up

When you are done with the examples, run this to clean up:

chef-client -z destroy_all.rb

What Is Chef Provisioning?

Chef Provisioning has two major abstractions: the machine resource, and drivers.

The machine resource

You declare what your machines do (recipes, tags, etc.) with the machine resource, the fundamental unit of Chef Provisioning. You will typically declare machine resources in a separate, OS/provisioning-independent file that declares the topology of your app--your machines and the recipes that will run on them.

The machine resources from the cluster.rb example are pretty straightforward. Here's a copy/paste:

# Database!
machine 'mario' do
  recipe 'postgresql'
  recipe 'mydb'
  tag 'mydb_master'

num_webservers = 1

# Web servers!
1.upto(num_webservers) do |i|
  machine "luigi#{i}" do
    recipe 'apache'
    recipe 'mywebapp'

You will notice the dynamic nature of the number of web servers. It's all code, your imagination is the limit :)


Drivers handle the real work of getting those abstract definitions into real, physical form. They handle the following tasks, idempotently (you can run the resource again and again and it will only create the machine once--though it may notice things are wrong and fix them!):

  • Acquiring machines from the cloud, creating containers or VMs, or grabbing bare metal
  • Connecting to those machines via ssh, winrm, or other transports
  • Bootstrapping chef onto the machines and converging the recipes you suggested

The driver API is separated out so that new drivers can be made with minimal effort (without having to rewrite ssh, tunneling, bootstrapping, and OS support). But to the user, they appear as a single thing, so that the machine acquisition can use its smarts to autodetect the other bits (transports, OS's, etc.).

Drivers save their data in the Chef node itself, so that they will be accessible to everyone who is using the Chef server to manage the nodes.

Drivers each have their own repository. Current drivers:




Bare Metal:

Seeking Maintainers: These repositories are not actively maintained and are seeking maintainers.

Machine options

You can pass machine options that will be used by machine, machine_batch and machine_image to configure the machine:

  convergence_options: {
    chef_version: "12.4.1",
    prerelease: "false",
    chef_client_timeout: 120*60, # Default: 2 hours
    chef_config: "log_level :debug\\n", # String containing additional text to inject into client.rb
    chef_server: "http://my.chef.server/", # TODO could conflict with https://github.com/chef/chef-provisioning#pointing-boxes-at-chef-servers
    bootstrap_proxy: "http://localhost:1234",
    bootstrap_no_proxy: "localhost, *.example.com, my.chef.server",
    ssl_verify_mode: :verify_peer,
    client_rb_path: "/etc/chef/client.rb", # <- DEFAULT, overwrite if necessary
    client_pem_path: "/etc/chef/client.pem", # <- DEFAULT, overwrite if necessary
    allow_overwrite_keys: false, # If there is an existing client.pem this needs to be true to overwrite it
    private_key_options: {}, # TODO ????? Something to do with creating node object
    source_key: "", # ?????
    source_key_pass_phrase: "", # ?????
    source_key_path: "", # ?????
    public_key_path: "", # ?????
    public_key_format: "", # ?????
    admin: "", # ?????
    validator: "", # ?????
    ohai_hints: { :ec2 => { :key => :value } }, # Map from hint file name to file contents, this would create /etc/chef/ohai/hints/ec2.json,
    ignore_failure: [1, 5..10, SomeSpecificError], # If true don't let a convergence failure on provisioned machine stop the provisioning workstation converge.  Can also provide a single exit code to ignore (no array) or `true` to ignore all RuntimeErrors
    # The following are only available for Linux machines
    install_sh_url: "https://www.chef.io/chef/install.sh", # <- DEFAULT, overwrite if necessary
    install_sh_path: "/tmp/chef-install.sh", # <- DEFAULT, overwrite if necessary
    install_sh_arguments: "-P chef-dk", # Additional commands to pass to install.sh
    # The following are only available for Windows machines
    install_msi_url: "foo://bar.com"
  ssh_username: "ubuntu", # Username to use for ssh and WinRM
  ssh_gateway: "user@gateway", # SSH gateway configuration
  ssh_options: { # a list of options to Net::SSH.start
    :auth_methods => [ 'publickey' ], # DEFAULT
    :keys_only => true, # DEFAULT
    :host_key_alias => "#{instance.id}.AWS", # DEFAULT
    :key_data => nil, # use key from ssh-agent instead of a local file; remember to ssh-add your keys!
    :forward_agent => true, # you may want your ssh-agent to be available on your provisioned machines
    :remote_forwards => [
        # Give remote host access to squid proxy on provisioning node
        {:remote_port => 3128, :local_host => 'localhost', :local_port => 3128,},
        # Give remote host access to private git server
        {:remote_port => 2222, :local_host => 'git.example.com', :local_port => 22,},
    # You can send net-ssh log info to the Chef::Log if you are having
    # trouble with ssh.
    :logger => Chef::Log,
    # If you use :logger => Chef::Log and :verbose then your :verbose setting
    # will override the global Chef::Config. Probably don't want to do this:
    #:verbose => :warn,

This options hash can be supplied to either with_machine_options or directly into the machine_options attribute.

Individual drivers will often add their own driver specific config. For example, AWS expects a :bootstrap_options hash at the same level as :convergence_options.

Anatomy of a Recipe

chef-zero comes with a provisioner for Vagrant, an abstraction that covers VirtualBox, VMware and other Virtual Machine drivers. In docs/examples, you can run this to try it:

export CHEF_DRIVER=vagrant
chef-client -z vagrant_linux.rb simple.rb

To use with VMware, simply update the prior example to read export VAGRANT_DEFAULT_PROVIDER=vmware_fusion

This is a chef-client run, which runs multiple recipes. Chef Provisioning is nothing but resources you put in recipes.

The driver is specified on the command line. Drivers are URLs. You could use vagrant:~/vms or `fog:AWS:default:us-east-1' as driver URLs. More information here.

The vagrant_linux.rb recipe handles the physical specification of the machines and Vagrant box:

require 'chef/provisioning/vagrant_driver'

vagrant_box 'precise64' do
  url 'http://files.vagrantup.com/precise64.box'

with_machine_options :vagrant_options => {
  'vm.box' => 'precise64'

require 'chef/provisioning/vagrant_driver' is how we bring in the vagrant_box resource.

vagrant_box makes sure a particular vagrant box exists, and lets you specify machine_options for things like port forwarding, OS definitions, and any other vagrant-isms.

Typically, you declare these in separate files from your machine resources. Chef Provisioning picks up the drivers and machine_options you have declared, and uses them to instantiate the machines you request. The actual machine definitions, in this case, are in simple.rb, and are generic--you could use them against Azure or EC2 as well:

machine 'mario' do
  tag 'itsame'

Other directives, like recipe 'apache', help you set run lists and other information about the machine.

Pointing Boxes at Chef Servers

By default, Chef Provisioning will put your boxes on the same Chef server you started chef-client with (in the case of -z, that's a local chef-zero server). Sometimes you want to put your boxes on different servers. There are a couple of ways to do that:

with_chef_local_server :chef_repo_path => '~/repo'

with_chef_local_server is a generic directive that creates a chef-zero server pointed at the given repository. nodes, clients, data bags, and all data will be stored here on your provisioner machine if you do this.

You can use with_chef_server instead if you want to point at OSS, Hosted or Enterprise Chef, and if you don't specify a Chef server at all, it will use the one you are running chef-client against. Keep in mind when using with_chef_server and running chef-client -z on your workstation that you will also need to set the client name and signing key for the chef server. If you've already got knife.rb set up, then something like this in the provisioning recipe will correctly create a client for the chef server on instance using your knife.rb configuration:

with_chef_server "https://chef-server.example.org",
  :client_name => Chef::Config[:node_name],
  :signing_key_filename => Chef::Config[:client_key]

Finally, you can specify a Chef Server for an individual machine by using the chef_server attribute. This attribute takes the same additional options (:client_name, :signing_key_filename) that with_chef_server does.

machine 'mario' do
  chef_server :chef_server_url => "https://chef-server.example.org"

Configuring ACLs

If you want to run your provisionning recipes from an actual chef node, rather than from your workstation, you need to give that node's client enough rights to create the node and client he'll be provisioning on the chef server. Without those additional rights, the provisioning will fail with a error along the lines of chef_client[mymachine] had an error: Net::HTTPServerException: 403 "Forbidden".

A clean solution to this problem is to use knife-acl to define a provisioners group with the required rights, and add your client to it:

$> chef gem install knife-acl
$> knife group create provisioners
$> knife acl add group provisioners containers clients read,create,update,delete,grant
$> knife acl add group provisioners containers nodes read,create,update,delete,grant
$> knife group add client my_provisioning_client_name provisioners


Chef Provisioning also works with Test Kitchen, allowing you to test entire clusters, not just machines! The repository for the kitchen-metal gem is https://github.com/doubt72/kitchen-metal.

Fixing conflict with chef-zero 3.2.1 and ~> 4.0

If you run into the error Unable to activate cheffish-1.0.0, because chef-zero-3.2.1 conflicts with chef-zero (~> 4.0) you'll need to update the version of the chef gem included in the ChefDK. Follow the instructions @ https://github.com/fnichol/chefdk-update-app and update chef to ~>12.2.1

Bugs and The Plan

Please submit bugs, gripes and feature requests at https://github.com/chef/chef-provisioning/issues, contact John Keiser on Twitter at @jkeiser2, email at jkeiser@chef.io

To contribute, just make a PR in the appropriate repo--also, make sure you've signed the Chef Contributor License Agreement (through your Chef Supermarket profile), since this is going into core Chef eventually. If you already signed this for a Chef contribution, you don't need to do so again--if you're not sure, you can check for your name here or if you signed up long ago check the old list!