stack-kicker is a simple 'application stack' deployment tool, it's purpose in life is to spin up a set of instances in a repeatable, controlled fashion, and optionally run post-install scripts after each instance has been started.
stack-kicker has hooks to allow default & custom cloud-init templates to be built & passed to your compute provider (we currently use ruby-openstack, so are limited to OpenStack providers, however, a sister project, aws-kicker, uses fog.io, the interaction with the compute provider is minimal, so it's on the roadmap to merge aws-kicker & stack-kicker, and use either an internal abstraction layer or just fog.io for all compute provisioning requests)
Normally, stack configurations are stored in a Stackfile, which is a ruby hash of configuration options. Typically, the Stackfile contains the following sets of information
- Credentials for your compute provider
- Defaults for your Stack (compute flavor, image id, region/az, ssh-keys, instance name template)
- roles
stack-kicker sequentially iterates over defined roles, creating the required number of nodes (nodes are the compute instances). If an instance already exists with the proposed name, it's skipped, assuming that this node has already been built.
Hostnames are generated from a customizable template, which is effectively:
config[:name_template] = '%s-%s-%s%04d'
config[:global_service_name] = 'myapp'
site = <derived from region/az, via config[:site_template]>
hostname = sprintf(config[:name_template], config[:global_service_name], site, role, position)
So hostnames will be myapp-az1-chef0001, myapp-az1-web0001, myapp-az1-web0002 etc.
post-install scripts are executed from the same host as stack-kicker is being used, using the same credentials as the current user. They are can be used to retrieve information from a freshly built node (like certificates from a chef server), so block progress until the chef-client run has completed (we use this to block percona/galera & rabbitmq cluster builds so that the first node is up & running correctly before we try and add another node to the cluster)
TODO
An optional array that extends were the find_file fucntion searches for files (find file defaults to ., Stackhome & the gem lib (for shipped templates), the elements of find_file_paths are appended to Stackhome
Roles have several attributes, which control how & how many nodes are created, and how they are created. Below shows the default values for these attributes, which can all be overridden.
:role_name = {
:count => 1,
:azs[] => config['REGION']
:chef_server => false,
:skip_chef_prereg => false,
:security_group => :role_name.to_s,
:cloud_config_yaml => 'cloud-config.yaml',
:bootstrap => 'chef-client-bootstrap-excl-validation-pem.sh',
:data_dir => '/dummy',
:floating_ips => nil
:post_install_script => nil,
:post_install_args => '',
:post_install_cwd => '/.'
}
The number of nodes or instances of this role that will be created.
This can be an array of strings, so that nodes will be placed in specific availability zones. If no array is set, the REGION set in the global section will be used.
this is flag used to denote if the node created by this role should be used as a chef server for nodes created by other roles. If this flag is set, we extract the public & private IP address for use later, as well downloading validation.pem & creating a user account in chef & downloading the pen (this is done via chef-post-install.sh, or your own alternative methods)
This is usually only used when :chef_server = true, it stops stack-kicker from attempting to pre-create the chef client & node and applying roles to the node.
security group to be assigned to this node. (set to default is you don't want to manage security groups for every role)
defaults to a file which contains a simple template (lib/cloud-config.yaml in the github repo) that installs the http://apt.opscode.com repo & gpg key, as well as installing the opscode-keyring. Can be replaced with any filename that complies with cloud-init. If the filename supplied ends in '.erb', it will be processed with ERB. See ERB Templates for details on available data.
Optional filename, the contents of which will get combined with :cloud_config_yaml to form the cloud-init payload (using mime encoding, supported types are #include, ) with some variable substation (chef server ip, environment, validation.pem, roles) See lib/chef-client-bootstrap-excl-validation-pem.sh as an example.
:cloud_config_yaml & :bootstrap files can be of the following type:
mime-type | first line |
---|---|
text/x-include-url | #include |
text/x-shellscript | #! |
text/cloud-config | #cloud-config |
text/upstart-job | #upstart-job |
text/part-handler | #part-handler |
text/cloud-boothook | #cloud-boothook |
Can be replaced with any filename that complies with cloud-init. If the filename supplied ends in '.erb', it will be processed with ERB. See ERB Templates for details on available data.
data_dir is a hook into the optional :cloud_config_yaml template (lib/cloud-config-w-ephemeral.yaml), which formats & mounts ephemeral0 early in the boot process, allowing it to be used during the rest of the cloud-init. ephemeral0 is mounted as /mnt & then bind mounted to #{data_dir}
This can be an array of strings, such that node X will be assigned :floating_ips[X-1] via "nova add-floating-ip". If :floating_ips[X-1].nil?, then no floating ip will be attached. (the floating ip must already be assigned to a pool in your account)
These are used to construct a command to execute, which is executed locally where you executed stack-kicker. :post_install_args can contain %PUBLIC_IP%, which will be replaced by the public IP of the just created node. :post_install_script scripts are executed as soon the the instance returns a status='ACTIVE'. They can be used delay the creation of further nodes of the same role (for example, when creating a rabbitmq cluster, you need to wait for the rabbitmq process to be running before creating the next member of the cluster, or when you are creating a chef-server, you need to wait for the packages to install & daemons to start before attempting to create Chef users & retrieve keys)
Both the :cloud_config_yaml & :bootstrap role attributes can point to plain files, files with simple tokens (%HOSTNAME%, %CHEF_SERVER%, %CHEF_SERVER_HOSTNAME%, %CHEF_ENVIRONMENT%, %CHEF_VALIDATION_PEM%, %SERVER_NAME%, %ROLE% and %DATA_DIR%) or Ruby ERB templates.
%HOSTNAME% is the short name of the instance being built %CHEF_SERVER% is the IP address of the Chef Server, private address if the chef server & current instance are in the same AZ, public IP if they are in different AZs %CHEF_SERVER_HOSTNAME% name of the chef server, Fully Qualified if config[:domain] is set
There are 3 key data sets exposed to the ERB templates:
- instances - subset all_instances, just the nodes referenced/managed by this Stackfile
- all_instances - hash of all instances running in this account
- config - config contains all the config data from the Stackfile, as well as info about running instances in the account used by the Stackfile
instances & all_instances hashes look like this:
{
"webci-az1-web0001" => {
:region => "az-1.region-a.geo-1",
:id => 1395635,
:private_ips => ["10.5.170.11"],
:public_ips => ["15.185.114.181"],
:az => "az-1.region-a.geo-1",
:role => :web},
"webci-az1-web0002" => {
:region => "az-1.region-a.geo-1",
:id => 1396181,
:private_ips => ["10.5.172.145"],
:public_ips => ["15.185.110.42"],
:az => "az-1.region-a.geo-1",
:role=>:web
}
}
So say you wanted to drop the private IP address of webci-az1-web0001 into the hosts file of webci-az1-web0002 via cloud-init, this fragment in your :cloud_config_yaml could be used:
bootcmd:
- echo <%=instances['webci-az1-web0001'][:private_ips][0]%> webci-az1-web0001 >> /etc/hosts
As we're using ERB for these templates, you can use the standard Ruby iterators to work through the hash:
bootcmd:
<% instances.each do |node_name, node_details| %>
- echo <%=node_details[:private_ips][0]%> <%=node_name%> >> /etc/hosts
<% end %>
stack-kicker was built with the following workflows in mind:
This was the original requirement, a multi-role application stack build that started with building a chef-server, uploading roles, environments, cookbooks & databags to it, and then building the rest of the application-stack instances, using the freshly built chef-server to drop the application on to the instances. In this setup we used vanilla images (Ubuntu 12.04 LTS, but you could use any image, either vanilla or pre-populated with your software).
Here's an example Stackfile for this:
module StackConfig
Stacks = {
'web-w-chef-server' => {
# (we can access environment variable via ENV['foo'] instead of hard coding u/p here)
'REGION' => ENV['OS_REGION_NAME'],
'USERNAME' => ENV['OS_USERNAME'],
'PASSWORD' => ENV['OS_PASSWORD'],
'AUTH_URL' => ENV['OS_AUTH_URL'],
'TENANT_NAME' => ENV['OS_TENANT_NAME'],
# generic instance info
'flavor_id' => 103,
'image_id' => 75845,
:key_pair => 'ssh-keypair-name',
:key_public => '/path/to/id_rsa.pub',
:global_service_name => 'perconaconf',
# role details
:roles => {
# override the default cloud-init script & default bootstrap (which is a chef-client bootstrap)
:chef => {
# we are the chef server, so skip the chef-client steps
:chef_server => true, :skip_chef_prereg => true,
# override the default cloud-config with a chef-server template
:cloud_config_yaml => 'chef-cloud-config.yaml',
# skip the default chef-client bootstrap
:bootstrap => '',
# wait for the chef server to come up & download pem files & generate client account
:post_install_script => 'bootstrap/chef-post-install.sh',
# our post install script dumps out .pem files in the CWD
:post_install_cwd => '.chef',
# The post-install script needs to know the public IP of the just built instance so that this station can access it
:post_install_args => '%PUBLIC_IP%'
},
# much simpler role, just build 3 of these, chef-client will do the rest on boot
:web => { :count => 3 }
}
}
}
end
There is no requirement that stack-kicker do anything other than spin up your instances, your requirements may be such that you just need a number of instances started with certain images, region & flavor requirements.
aws-kicker (a sister project) had an original requirement of starting a simple 2-tier web application in multiple locations/environments (prod, stage, dev etc), to do this we configured the instances by bootrapping puppet, git clonig /etc/puppet and running "puppet apply", a simple pattern used in many places, this was all achievd with a carefully crafted cloud-init template (incidentally, this also allowed for simple prototyping using vagrant to provide local instances using the exact same '/etc/puppet' git repo.
These are only the workflows I've used, there is no reason a puppet master couldn't be built & used, or hosted/external puppet & chef servers. (pull requests accepted etc, including salt, ansible, cfengine etc..)
$ gem install stack-kicker
In addition to the the ruby dependencies which gem will install for you, access to python-novaclient is currently required to attach floating-ips to instances.
Usage: stack-kicker [options] task
Options:
-h, --help Show command line help
--stackfile Stackfile Specify an alternative Stackfile
(default: Stackfile)
--stack mystack Specify the stack in Stackfile that you want to work with
--ssh-user USER User to be used for SSH access
--version Show help/version info
--log-level LEVEL Set the logging level
(debug|info|warn|error|fatal)
(Default: info)
Arguments:
task
task to be performed validate|configure-knife|show-stacks|show-stack|show-running|build|replace|delete|secgroup-sync|ssh
- Clean/Add up provider logic
- Remove dependency on python-novaclient for floating-ip attach
- Remove dependency on a full chef gem install
- Better docs & examples
- Support for AWS EC2 (from aws-kicker)
- Support for DNS Updates on instance creation (from aws-kicker)
- gem bump
- gem release
- Fork it
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create new Pull Request