awsborn

This Gem lets you define and launch a server cluster on Amazon EC2. The launch operation is idempotent, i.e., it only launches instances that are not running.

Installation

gem install awsborn

Synopsis

#! /usr/bin/env ruby

require 'rubygems'
require 'awsborn'

# If not set, will be picked up from ENV['AMAZON_ACCESS_KEY_ID']
Awsborn.access_key_id = 'AAAAAAAAAAAA'
# Can be picked up from ENV['AMAZON_SECRET_ACCESS_KEY'] or Keychain (OS X)
Awsborn.secret_access_key = 'KKKKKKKKKKKK'

# Logs to `awsborn.log` in current dir with level INFO by default.
# To suppress logging, do `Awsborn.logger = Logger.new('/dev/null')`
Awsborn.logger.level = Logger::DEBUG

class LogServer < Awsborn::Server
  instance_type    :m1_small
  image_id         'ami-2fc2e95b', :sudo_user => 'ubuntu'
  security_group   'Basic web'
  keys             'keys/*.pub'
  bootstrap_script 'chef-bootstrap.sh'
  monitor          true
end

log_servers = LogServer.cluster do
  domain 'releware.net'
  server :log_a, :zone => :eu_west_1a, :disk => {:sdf => "vol-a857b8c1"}, :ip => 'log-a'
  server :log_b, :zone => :eu_west_1b, :disk => {:sdf => "vol-aa57b8c3"}, :ip => 'log-b'
end

log_servers.launch

log_servers.each do |server|
  system "rake cook server=#{server.host_name}"
end

How it works

It's really simple:

Define a server type (LogServer in the example above).
Define a cluster of server instances.
Launch the cluster.

See Launching a cluster below for details.

Defining a server type

Server types can be named anything but must inherit from Awsborn::Server. Servers take five different directives:

instance_type -- a symbol. One of :m1_small, :m1_large, :m1_xlarge, :m2_2xlarge, :m2_4xlarge, :c1_medium, :c1_xlarge, cc1.4xlarge, and t1.micro.
image_id -- a valid EC2 AMI or a hash {:i386 => 'ami-3232', :x64 => 'ami-6464'} Specify :sudo_user as an option if the AMI does not allow you to log in as root.
security_group -- a security group that you own.
keys -- one or more globs to public ssh key files. When the servers are running, root will be able to log in using any one of these keys.
bootstrap_script -- path to a script which will be run on each instance as soon as it is started. Use it to bootstrap chef and let chef take it from there. A sample bootstrap script is included towards the end of this document.

A server cannot be started without instance_type, image_id and security_group. The don't have to be defined for the server type though, but can equally well be added as options to the specific servers.

Defining a cluster

The cluster method accepts two commands, domain and server. domain is optional and is just a way to avoid repetition in the server commands. server takes a name (which can be used as a key in the cluster, e.g. log_servers[:log_a]) and a hash:

Mandatory keys:

:zone -- the availability zone for the server. One of :us_east_1a, :us_east_1b, :us_east_1c, :us_east_1d, :us_west_1a, :us_west_1b, :eu_west_1a, :eu_west_1b, :ap-southeast-1a, :ap-southeast-1b.
:disk -- a hash of device => volume-id. Awsborn uses the disks to tell if a server is running or not (see Launching a cluster). volume-id can also be an array [volume-id, :format], in which case the_server.format_disk_on_device?(device) will return true. See contrib/cookbooks/ec2-ebs for example usage.

Optional keys:

:ip -- a domain name which translates to an elastic ip. If the domain name does not contain a full stop (dot) and domain has been specified above, the domain is added.

Keys that override server type settings:

:instance_type
:sudo_user
:security_group
:keys
:bootstrap_script
:image_id

Launching a cluster

The launch method on the cluster checks to see if each server is running by checking if the server's disks are attached to an EC2 instance, i.e., the server is defined by its content, not by its AMI or ip address.

Servers that are not running are started by calling the start method on the server, which does the following:

def start (key_pair)
  launch_instance(key_pair)

  update_known_hosts
  install_ssh_keys(key_pair) if keys

  if elastic_ip
    associate_address
    update_known_hosts
  end

  bootstrap if bootstrap_script
  attach_volumes
end

The key_pair is a temporary key pair that is used only for launching this cluster. In case of a failed launch, the private key is available as /tmp/temp_key_*.

A bootstrapping script

This is what we use with the AMI above:

#!/bin/bash

echo '------------------'
echo 'Bootstrapping Chef'
echo

aptitude -y update
aptitude -y install gcc g++ curl build-essential \
  libxml-ruby libxml2-dev \
  ruby irb ri rdoc ruby1.8-dev libzlib-ruby libyaml-ruby libreadline-ruby \
  libruby libruby-extras libopenssl-ruby \
  libdbm-ruby libdbi-ruby libdbd-sqlite3-ruby \
  sqlite3 libsqlite3-dev libsqlite3-ruby

curl -L 'http://rubyforge.org/frs/download.php/69365/rubygems-1.3.6.tgz' | tar zxf -
cd rubygems* && ruby setup.rb --no-ri --no-rdoc

ln -sfv /usr/bin/gem1.8 /usr/bin/gem

gem install chef ohai --no-ri --no-rdoc --source http://gems.opscode.com --source http://gems.rubyforge.org

echo
echo 'Bootstrapping Chef - done'
echo '-------------------------'

Running with Chef Solo

Awsborn integrates with Chef Solo. An example from reality:

Awsborn.access_key_id = 'AKIAJHL53MPX7LPCKIFQ'

class LogServer < Awsborn::Server
  instance_type    :m1_small
  image_id         'ami-2fc2e95b', :sudo_user => 'ubuntu'
  security_group   'Basic web'
  keys             '../keys/*'
  bootstrap_script 'chef-bootstrap.sh'
  monitor          true

  cluster do
    domain 'releware.net'
    server :log_a, :zone => :eu_west_1a, :disk => {:sdf => "vol-a857b8c1"}, :ip => 'log-a'
    server :log_b, :zone => :eu_west_1b, :disk => {:sdf => "vol-aa57b8c3"}, :ip => 'log-b'
  end

  def chef_dna
    {
      :user => "lumberjack",
      :host => host_name,
      :users =>  [
        {
          :username => "lumberjack",
          :authorized_keys => key_data,
          :gid => 1001,
          :uid => 1001,
          :sudo => true,
        }
      ],
      :packages => %w[
        vim
        heirloom-mailx
      ],
      :gems => [
        "rake"
      ],
      :ebs_volumes => [
        {:device => "sdf", :path => "/apps"}
      ],
      :recipes => [
        "packages",
        "users",
        "sudo",
        "openssh",
        "ec2-ebs",
        "git",
        "gems",
        "lumberjack"
      ]
    }
  end
end

Just add a Rakefile in the same directory:

require 'awsborn'
include Awsborn::Chef::Rake
require './servers'

Put your cookbooks directory in the same directory as the Rakefile or in its parent directory. Run rake to start all servers and run Chef on each of them. Other rake tasks include:

rake chef - Run chef on all servers, or the ones specified with host=name1,name2.
rake chef:debug - Ditto, but with chef's log level set to debug.
rake start - Start all servers (or host=name1,name2) but don't run chef.

Bugs and surprising features

There is a race condition in launch: Immediately after the launch request (launch_instance) we call instance_running?, but occasionally, AWS will not recognise the instance id yet (InvalidInstanceID.NotFound: The instance ID 'i-e9005283' does not exist)
Running with rake suppresses stack traces which makes it hard to debug rare errors. The rake tasks or entry points should catch exceptions and print the stack trace.
t1_micro instances will select an i386 image when image_id is specified as a hash. To run them with x64 images, specify the image name on the server directive.

Untested features

Launching without ssh keys. (I suppose certain AMIs would support that.)

To do / Could do

Tests.
Dynamic discovery of instance types and availability zones.
Hack RightAws to verify certificates.
Elastic Load Balancing.
Launch servers in parallel.

Note on Patches/Pull Requests

Fork the project.
Make your feature addition or bug fix.
Add tests for it. This is important so I don't break it in a future version unintentionally.
At the moment, the gem does not have much tests. I intend to fix that, and I won't accept patches without tests.
Commit, do not mess with rakefile, version, or history. (if you want to have your own version, that is fine but bump version in a commit by itself I can ignore when I pull)
Send me a pull request. Bonus points for topic branches.

Changes

New in 0.5.1

When awsborn uploads cookbooks and configuration, it looks for a cookbooks directory in the parent directory if none is found in the Rakefile's directory.

New in 0.5.0

Support for cc1.4xlarge and t1.micro instance types.

History

This gem is inspired by Awsymandias which was sort of what I needed but was too complicated to fix, partly because the documentation was out of date. Big thanks for the inspiration and random code snippets.