since until

Dock(erize) those Containers

I do a thing or two with containers from time to time, that involves building Mesosphere or Kubernetes setups. Often enough my OS of choice happens to be CoreOS, but as ashamed as I am I should mention that I often end up scratching my head and wondering what the hell that command was again :pensive:?!? Hope this helps me recall.


Provided, that you are using CoreOS as your VM OS, you may configure your boxes through the so-called cloud-config files. In several setups you will provide etcd details in the cloud config file. This section will just give you a general idea of where to look for your resources when introspecting CoreOS boxes.

One may query the unit-files known to a CoreOS box by running the followng command

systemctl list-unit-files

I, for one, often end up looking at the the service files of my etcd.service which happen to be somewhere in /run/systemd/system/etcd.service.d. A cloud-config containing

    name: etcd
    addr: $private_ipv4:4001
    peer-addr: $private_ipv4:7001
    cluster-active-size: 1
    etcd-http-read-timeout: 86400
    snapshot: true
    - name: etcd.service
      command: start

May result to the following conf in the etcd.service.d directory:


Some observation leads one to assume that there happens to be a one-to-one correspondence between the cloud-config content and the eventual configuration with the key names capitalized and prefixed with ETCD\_ and the $private\_ipv4 token substituted for the machines private IP address.

According to the CoreOS documentation one should find unit files in the /etc/systemd/system directory.


Fleet handles orchestration on CoreOS clusters. Somewhere I once read to “think of it as systemd for the cloud” and darn… that is spot on.

As you start daemons on your box with systemctl (controlling systemd) one may start services on the cluster with fleetctl (controlling fleetd).

The beautiful thing, I came to learn about fleet is that it stores the settings in etcd. I discovered it after, destroying some VM’s, creating them anew and discovering that somehow my services were magically spawned on the fresh boxes without my intervention. Introspecting the etcd store left me clueless but some basic sleuthing uncovered that fleet actually keeps track of this information in hidden etcd keys. There is actually a world of CoreOS-related etcd secrets to behold if you run etcdctl ls --recursive`).


The entire concept of clustering on CoreOS seems to require clustered machines to share a discovery repository. This is the way clustered machines seem to share information in the CoreOS world and etcd seems to be one of the popular solutions for making this happen.


Somewhere along the line a unit/service breaks and needs to be debugged.

CoreOS logs everything into /var/log/journal/* and journalctl allows one to view the content of these binary logs.

sudo journalctl --file=/var/log/journal/REF/FILE.log

I ran into the "Failed units: 1" notification which is presented at login. After reviewing which services were up with sudo systemctl list-units, I discovered that the gce-coreos-cloudinit.service had failed. Running journalctl and filtering for the logs regarding a specific service using the -u SERVICENAME parameters makes for much easier reading:

sudo journalctl -u gce-coreos-cloudinit.service --file=JOURNAL_FILE.log