clarkdave.net

Managing EBS volumes with Chef

Chef works great for creating and attaching Amazon EBS volumes to instances, and there’s an Opscode chef-aws cookbook which has providers for working with single EBS volumes and also for creating RAID arrays from multiple volumes.

The chef-aws cookbook is fairly straightforward on its own, but requires a little DIY to get it fully working. I’ll explain here how to use this cookbook to effectively create single or RAID EBS volumes and allow these to be customised using attributes in roles or environments (so you could have your staging DB use a single 20GB volume, and your production DB use a 200GB RAID).

Getting started

We’ll start by grabbing the aws cookbook. The version released in Opscode community is 0.100.6 and does not have the ebs_raid provider which is present in the master on GitHub. The cookbook doesn’t see a huge amount of development, so it should be fine to grab a copy from master. You can get it yourself, or use the nifty Knife Github plugin which will let you run:

knife cookbook github install opscode-cookbooks/aws

which will install the master from GitHub and track changes as a branch.

Recipes and roles

For the sake of these examples, I’ll assume you have a cookbook called app.

You’ll probably have a specific use-case in mind for your EBS volumes - for example, you may be using them for a database. So, let’s create a database recipe in the app cookbook and add this inside it:

# The database recipe should be included by any server running a DB. It creates
# a /data directory and, if on EC2, will mount an EBS volume here

directory '/data' do
  mode '0755'
end

recipe[app::database] should be added to any of your database server roles, e.g. PostgreSQL or MongoDB. These databases would be configured to use the /data directory provided by the app::database recipe, like so:

name 'mongodb_server'
description 'MongoDB server'

run_list 'recipe[app::database]', 'recipe[mongodb]'

default_attributes(
  :mongodb => { :dbpath => '/data/mongodb' }
)

Configuration and attributes

I suggest using attributes to determine our EBS configuration, which will allow you to change the EBS parameters in roles and environments. I use this to create smaller EBS volumes on staging instances, and larger, RAIDed volumes in production. If you run different DB servers, you can also override these attributes in their respective roles (perhaps you want smaller volumes for PostgreSQL, and larger for MongoDB).

Start by creating some default attributes in the app cookbook (so in cookbooks/app/attributes/default.rb):

default[:app][:ec2] = false
default[:app][:ebs] = {
  :raid => false,
  :size => 20 # size is in GB
}

As you can see, our roles/environments can now selectively dictate the size of their EBS volume, and if it should be a RAID.

We’ve also set one other default, [:app][:ec2], which is set to false. The idea is you set this to true in those environments which will be running on EC2, so you can avoid running any EC2-related stuff on nodes that aren’t on AWS (e.g. local machines, vagrant test runs, etc).

Alternatively, you could use Ohai’s cloud attributes to determine if the current node is running on EC2. If this works reliably for you, feel free to use that and eschew the [:app][:ec2] attribute entirely.

Creating the EBS volumes

Now, back to our recipes/database.rb file. Before we add the code for creating the volumes, make sure you’ve added your AWS credentials to a data bag, as explained on aws cookbook’s GitHub page.

if node[:app][:ec2] || node[:cloud][:provider] == 'ec2' # the latter check uses Ohai's cloud detection

  # get AWS credentials from the aws data_bag
  aws = data_bag_item('aws', 'main')

  include_recipe 'aws'

  if node[:app][:ebs][:raid]

    # use the aws_ebs_raid provider to create and mount a RAID volume. This provider
    # basically does everything for us, so there's nothing more to do!
    aws_ebs_raid 'data_volume_raid' do
      mount_point '/data'
      disk_count 2
      disk_size node[:app][:ebs][:size]
      level 10
      filesystem 'ext4'
      action :auto_attach
    end
  else
    # create a single EBS volume
    ## TODO
  end
end

There are several other options for a RAID disk: the number of disks, the level and filesystem. If you need these to change across roles/environments, you should add these as attributes in [:app][:ebs] too.

Now let’s fill in the ## TODO section and create single EBS volumes. The provider for this does not mount or create a filesystem for single volumes, so we’ll have to do that by hand:

# get an unused device ID for the EBS volume
devices = Dir.glob('/dev/xvd?')
devices = ['/dev/xvdf'] if devices.empty?
devid = devices.sort.last[-1,1].succ

# save the device used for data_volume on this node -- this volume will now always
# be attached to this device
node.set_unless[:aws][:ebs_volume][:data_volume][:device] = "/dev/xvd#{devid}"

device_id = node[:aws][:ebs_volume][:data_volume][:device]

# create and attach the volume to the device determined above
aws_ebs_volume 'data_volume' do
  aws_access_key aws['aws_access_key_id']
  aws_secret_access_key aws['aws_secret_access_key']
  size node[:app][:ebs][:size]
  device device_id.gsub('xvd', 'sd') # aws uses sdx instead of xvdx
  action [:create, :attach]
end

# wait for the drive to attach, before making a filesystem
ruby_block "sleeping_data_volume" do
  block do
    timeout = 0
    until File.blockdev?(device_id) || timeout == 1000
      Chef::Log.debug("device #{device_id} not ready - sleeping 10s")
      timeout += 10
      sleep 10
    end
  end
end

mount_point = '/data'

# create a filesystem
execute 'mkfs' do
  command "mkfs -t ext4 #{device_id}"
  # only if it's not mounted already
  not_if "grep -qs #{mount_point} /proc/mounts"
end

# now we can enable and mount it and we're done!
mount "#{mount_point}" do
  device device_id
  fstype 'ext4'
  options 'noatime,nobootwait'
  action [:enable, :mount]
end

And that’s pretty much it. The ebs providers will save attributes on the node (in [:aws][:ebs_volume] and [:aws][:raid]) so they remember which volumes are associated with it. This means that if the volumes are detached, the next chef-client run should reattach them correctly. It is also possible, using the same providers, to attach existing volumes instead of creating them on demand.

All you need to do now is adjust your database server configurations to use the /data directory, and configure your roles and environments with suitable EBS settings.

You can see the entire app::database recipe on GitHub.

Updated 2014-05-14: added sleeping_data_volume improvements by Joshua Timberman

blog comments powered by Disqus