Add containerd metrics configurability

[bogn83]

containerd provides its own metrics server exposing metrics in prometheus format if you configure an address for it in /etc/containerd/config.toml. That file is watched and salt triggers a restart of the containerd systemd unit if it's changed. reload: true is not used on the service because the containerd unit is not reloadable.

[bogn83]

Just found out that for user namespace remap scenarios the main difference compared to cadvisor is that instead of a cgroup slice (the case for cadvisor) we get the real container ID but sadly not the container name.

[chr4]

Thanks for the pull request! I'm definitly in favor of adding this to the formula!

I think it makes sense to deploy the complete /etc/containerd/config.toml instead of using file.append. The default file seems to be not very large:

disabled_plugins = ["cri"]

#root = "/var/lib/containerd"
#state = "/run/containerd"
#subreaper = true
#oom_score = 0

#[grpc]
#  address = "/run/containerd/containerd.sock"
#  uid = 0
#  gid = 0

#[debug]
#  address = "/run/containerd/debug.sock"
#  uid = 0
#  gid = 0
#  level = "info"

Maybe we can put the complete content into a YAML pillar parameter, like so (this should be the default):

docker:
  containerd:
    config: |
      disabled_plugins = ["cri"]

That would make all options configurable, while being simple. Furthermore it enables consistent control over the complete file, which can't be guaranteed with file.append.

What do you think?