chef_client_updater

This cookbook updates the chef-client

Requirements

Platforms

• All platforms with a chef-client package on downloads.chef.io

Chef

• Chef 11.6.2+

Usage

This cookbook provides both a custom resource and a default recipe. The default recipe simply uses the custom resource with a set of attributes. You can add chef_client_updater::default to your run list or use the custom resource in a wrapper cookbook.

Init System Caveats

When Chef runs as a service under a system init daemon such as Sys-V or systemd each chef run forks off from the main chef-client process being managed by the init system. For a chef-client upgrade to occur the running chef-client as well as the parent process must be killed, and a new chef-client must start using the updated binaries. This cookbook handles killing the chef-client, but your init system must properly handle starting the service back up. For systemd and upstart this can be handled via configuration, and chef-client cookbook 8.1.1 or later handles this by default. This functionality is not available in sys-v (RHEL 6, Debian 7, and others), so you will need to employ a secondary process such as a monitoring system to start the chef-client service.

Updating Windows Nodes

There are a couple of considerations on Windows that have to be dealt with. The Chef Client installer uses a custom component to speed up the installation. This component does not gracefully handle open file handles the way the MSI installer does. To work around this, the resource moves the currently installed Chef Client to a staging directory and that clears the way for the newer installer to run. At the end of that installation process though, that Chef Client run must exit or it will fail trying to find files that do not exist in their expected locations. The next run of the Chef Client will use the newly installed version.

Running Chef Client as a Scheduled Task

If you run as a scheduled task, then this will work smoothly. The path to the newly installed Chef Client will be the same and the scheduled task will launch it. Part of this resource's job on the next run is to make sure the staging directory with the older client is removed.

Running Chef Client As A Windows Service

If you run Chef Client as a service, things get a tiny bit more complicated. When the new installer runs, the service is removed. This isn't a big deal if you've got the chef-client cookbook set to configure the Windows service. If that is the case, we can register a scheduled task to run shortly after the chef_client_updater terminates the current chef run. An example recipe might look like:

if Chef::VERSION < node['my_update_cookbook']['desired_version']
  run_chef_task_in_ten_minutes = Time.now + 600

  windows_task 'chef-client-upgrade' do
    cwd 'C:\\opscode\\chef\\bin'
    command 'chef-client'
    run_level :highest
    frequency :once
    start_time "#{run_chef_task_in_ten_minutes.strftime('%H:%M')}"
    action :create
  end
else
  windows_task 'chef-client-upgrade' do
    cwd 'C:\\opscode\\chef\\bin'
    command 'chef-client'
    run_level :highest
    frequency :once
    start_time "#{run_chef_task_in_ten_minutes.strftime('%H:%M')}"
    action :delete
  end
end

chef_client_updater 'Install latest Chef' do
  post_install_action 'kill'
end

Upgrading from Chef 11

Moving from Chef 11 has a few challenges when we are dealing with public update sources. Chef 11 ships with a very old cacert.pem. To work through this, we need to get a more current cacert.pem file and point OpenSSL to it. Unfortunately, for this to work consistently on Windows, we'll need to reboot. Chef 11 does not have the reboot resource, so this isn't a graceful process. However, on the next Chef run after the reboot, things will be back on track and the upgrade will perform as on other platforms.

Below is an example of a recipe that can set up Chef 11 to work using public update sources.

if platform_family?('windows') && (Chef::VERSION < '12')
  new_cert_file = File.join(ENV['USERPROFILE'], 'cacert.pem')

  remote_file new_cert_file do
    source 'https://curl.haxx.se/ca/cacert.pem'
    action :create
  end

  powershell_script 'restart' do
    code <<-EOH
    restart-computer -force
    EOH
    action :nothing
  end

  env 'SSL_CERT_FILE' do
    value new_cert_file
    notifies :run, 'powershell_script[restart]', :immediately
  end
end

chef_client_updater 'Install latest Chef' do
  post_install_action 'kill'
end

Resources

chef_client_updater

Installs the mixlib-install/mixlib-install gems and upgrades the chef-client.

properties

• channel - The chef channel you fetch the chef client from. stable contains all officially released chef-client builds where as current contains unreleased builds. Default: stable
• prevent_downgrade - Don't allow this cookbook to downgrade the chef-client version. Default: false
• version - The version of the chef-client to install. Default :latest
• post_install_action - After installing the chef-client what should we do. exec to exec the new client or kill to kill the client and rely on the init system to start up the new version. Default: kill
• exec_command - The chef-client command. default: $PROGRAM_NAME.split(' ').first
• exec_args - An array of arguments to exec the chef-client with. default: ARGV
• download_url_override - The direct URL for the chef-client package.
• checksum - The SHA-256 checksum of the chef-client package from the direct URL.
• upgrade_delay - The delay in seconds before the scheduled task to upgrade chef-client runs on windows. default: 61. Lowering this limit is not recommended.
• product_name - The name of the product to upgrade. This can be chef or chefdk default: chef

examples

chef_client_updater 'Install latest'

chef_client_updater 'Install latest Chef 13.x' do
  version '13'
end

chef_client_updater 'Install 12.13.36 and kill' do
  version '12.13.36'
  post_install_action 'kill'
end

Test Kitchen Testing

In order to test this cookbook it will be necessary to change the post_install_action to exec from kill. While kill is better in most actual production use cases as it terminates the chef-client run along with cleaning up the parent process, the use of kill under test kitchen will fail the chef-client run and fail the test-kitchen run. The use of exec allows test-kitchen to complete and then re-runs the recipe to validate that the cookbook does not attempt to re-update the chef-client and will succeed with the new chef-client. This, however, means that it is not possible to exactly test the config which will be running in production. The best practice advice for this cookbook will be to ignore common best practices and not worry about that. If you change your production config to use exec in order to run what you test in test-kitchen, then you will find sharp edge cases where your production upgrades will hang and/or fail, which testing will not replicate. In order to test you should most likely test upgrades on your full-scale integration environment (not under test-kitchen) before rolling out to production and not use test-kitchen at all. If you think that there's a rule that you must test absolutely everything you run under test-kitchen, you should probably read this or this.

In order to test that your recipes work under the new chef-client codebase, you should simply test your cookbooks against the new version of chef-client that you wish to deploy in "isolation" from the upgrade process. If your recipes all work on the old client, and all work on the new client, and the upgrader works, then the sum of the parts should work as well (and again, if you really deeply care about the edge conditions where that might not work -- then test on real production-like images and not with test-kitchen).

Use of 'exec' in production

This is highly discouraged since the exec will not clean up the supervising process. You're very likely to see it upgrade successfully and then see the old chef-client process continue to run and fork off copies of the old chef-client to run again. Or for the upgrade process to hang, or for other issues to occur causing failed upgrades.

You can use 'exec' in production if you are running from cron or some other process manager and firing off single-shot --no-fork chef-client processes without using the --interval option. This will have the advantage that the new chef-client kicks off immediately after the upgrade giving fast feedback on any failures under the new chef-client. The utility of this approach is most likely is not enough to justify the hassle.

Maintainers

This cookbook is maintained by Chef's Community Cookbook Engineering team. Our goal is to improve cookbook quality and to aid the community in contributing to cookbooks. To learn more about our team, process, and design goals see our team documentation. To learn more about contributing to cookbooks like this see our contributing documentation, or if you have general questions about this cookbook come chat with us in #cookbok-engineering on the Chef Community Slack

License

Copyright:: 2016-2018, Chef Software, Inc

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.