![Gitter](https://badges.gitter.im/Join Chat.svg)
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!
This video explains the basics of chef-provisioning (though provisioners are now called drivers). Slides (more up to date) are here.
Date | Blog |
---|---|
6/3/2014 | machine_batch and parallelization |
6/3/2014 | Chef Provisioning, Configuration and Drivers |
3/4/2014 | Chef Metal 0.2: Overview - this is a pretty good overview (though dated). |
12/20/2013 | Chef Metal Alpha |
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-fog
export CHEF_DRIVER=fog: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.
When you are done with the examples, run this to clean up:
chef-client -z destroy_all.rb
Chef Provisioning has two major abstractions: the machine resource, and drivers.
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 example myapp::small are pretty straightforward. Here's a copy/paste:
machine 'mario' do
recipe 'postgresql'
recipe 'mydb'
tag 'mydb_master'
end
num_webservers = 1
1.upto(num_webservers) do |i|
machine "luigi#{i}" do
recipe 'apache'
recipe 'mywebapp'
end
end
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:
Cloud:
Virtualization:
- Vagrant: VirtualBox, VMWare Fusion, etc.
- VSphere (not yet up to date with 0.11)
Containers:
Bare Metal:
- Hanlon
- OpenCrowbar OpenCrowbar controls your real metal. It discovers, inventories, configs RAID & BIOS and networks, and installs your OS. OpenCrowbar website OpenCrowbar github
- SSH (no PXE) (not yet up to date with 0.11)
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
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'
end
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'
end
Other directives, like recipe 'apache'
, help you set run lists and other information about the machine.
chef-provisioning also comes with a Fog provisioner that handles provisioning to Openstack, Rackspace, Amazon's EC2 and other cloud drivers. Before you begin, you will need to put your AWS credentials in ~/.aws/config in the format mentioned in Option 1 here. There are other ways to specify your credentials, but this is the standard one for the Amazon CLI.
Once your credentials are in, basic usage looks like this:
export CHEF_DRIVER=fog:AWS
chef-client -z simple.rb
Other valid URLs include fog:AWS:myprofilename
and fog:AWS:profilename:us-west-2
.
Most Chef Provisioning drivers try hard to provide reasonable defaults so you can get started easily. Once you have specified your credentials, AMIs and other things are chosen for you.
You will usually want to create or input a custom key pair for bootstrap. To customize, specify keys and AMI and other options, you can make recipes like this:
require 'chef/provisioning/fog_driver'
fog_key_pair 'my_bootstrap_key'
with_machine_options :bootstrap_options => {
:key_name => 'my_bootstrap_key',
:image_id => 'ami-59a4a230',
:flavor_id => 't1.micro'
}
fog_key_pair
creates a new key pair (if the files do not already exist) and uploads it to AWS (it will toss an error if the key pair already exists and does not match). By default, fog_key_pair
will look for matching key files in .chef/keys, ~/.chef/keys and ~/.ssh. If it does not find one, it will place the key in .chef/keys
. You can override this path in fog_key_pair, but if you do, you will want to modify private_key_paths
in your configuration to match.
with_machine_options
specifies machine_options that will be applied to any machine
resources chef-client encounters.
You will notice that we are still using simple.rb
here. Machine definitions are generally driver-independent. This is an important feature that allows you to spin up your clusters in different places to create staging, test or miniature dev environments.
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 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]
Note for Hosted/Enterprise Chef Servers
Currently, you will need to add the 'clients' group to the 'admin' group in order for machine provisioning to work:
knife edit /groups/admin.json -e <editor>
Then add:
{
"users": [
"pivotal" # This is an internal superuser for Hosted/Enterprise Chef
],
"groups": [
"clients" # This is what you need to add
]
}
This can also be done through the Chef Server web UI (Administration tab > Groups > select admins Group > Add 'clients'
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.
Please submit bugs, gripes and feature requests at https://github.com/opscode/chef-provisioning/issues, contact jkeiser on Twitter at @jkeiser2, email at jkeiser@getchef.com
To contribute, just make a PR in the appropriate repo--also, make sure you've signed the Chef Contributor License Agreement (quick couple of minutes online), since this is going into core Chef eventually. It takes some time to process, so if you've just done it, let me know in the PR :) 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!