Docker not installed before running containers with Puppet - puppet

Question

I am trying to run a docker container using the puppetlabs/docker module. However when the Puppet agent attempts to run the container I receive the error
Error: Failed to apply catalog: No such file or directory - docker
It would seem that neither the docker daemon nor the docker client are being installed before Puppet attempts to create the container.
An excerpt of the configuration is as follows:
Puppetfile
# frozen_string_literal: true
forge 'https://forge.puppet.com'
# Modules from the Puppet Forge
# ...
mod 'puppetlabs-docker', '4.4.0'
# ...
agent.pp
class profile::runner::agent (
Enum[present, absent] $ensure = present,
String $version = undef,
String $image = undef,
String $container_name = "${facts['group']}-agent",
Array[String] $container_environment = [],
) {
class { 'docker':
version => $version,
}
# ...
docker::run { $container_name:
ensure => $ensure,
image => $image,
env => $container_environment,
net => 'host',
restart => 'unless-stopped',
}
# ...
}
My understanding is that this configuration from the puppetlabs/docker module is supposed to ensure docker is installed before any container is started. I considered making a bug report on the module itself but this issue would surely be so common I'd be surprised if it had not already been reported. So I concluded that I must be doing something incorrect.
I have tried the usual metaparameters but none seem to have any effect and result in the same error. For example I have tried
docker::run { $container_name:
ensure => $ensure,
image => $image,
env => $container_environment,
net => 'host',
restart => 'unless-stopped',
require => Class['docker'], # and also Package['docker']
}
It may be worth mentioning that the Puppet agent is running on RockyLinux so therefore a RedHat OS family. If I remove the docker::run from the configuration and then run the Puppet agent then the catalog is applied successfully but of course the container does not run. Then adding the docker::run back to the configuration and running the agent again will run the container successfully. This is what has indicated to me that there is a dependancy issue that I have not been able to resolve.

Answer 1

Since neither chaining arrows nor metaparameters seem to have an affect in this specific situation I have had to resolve this issue using a guard clause that checks if Docker is installed using the facts provided by the puppetlabs/docker module
if $facts['docker_client_version'] != undef {
docker::run { $container_name:
ensure => $ensure,
image => $image,
env => $container_environment,
net => 'host',
restart => 'unless-stopped',
}
}
This is not ideal as it has the effect of requiring the catalog to be applied twice before the container will be run but it does solve the problem.

Answer 2

Declared classes are not automatically ordered relative to each other or relative to resources declared in the same context. There are good reasons for that, but they are tangential to the question.
However, you can use the same techniques to impose relative ordering on classes that you can use for resources: the before, require, notify, and subscribe metaparameters, and the chaining arrows. For your particular case, you might also be able to use the require function (even if you also use a resource like class declaration of class docker, but in the case, the resource-like declaration must precede the require).
For example, this is a pretty good way to make sure that class docker is applied before every docker::run instance, wherever declared:
class { 'docker':
version => $version,
}
-> Docker::Run<||>
That does, however, have the side effect of realizing any virtual docker::run instances you might have declared. If that's a problem then go with one of the other alternatives.