Scott's Weblog The weblog of an IT pro specializing in cloud computing, virtualization, and networking, all with an open source view

Using Vagrant with CoreOS, etcd, fleet, and Docker

As a follow-up to my recent #vBrownBag session on “Docker and Friends,” I wanted to provide a quick and relatively easy way for VMware administrators to experiment with some of the technologies I demonstrated. Since not everyone has their own OpenStack cloud running in their basement, Vagrant seemed like a reasonable solution. So, in this post, I’ll show you how to use Vagrant to experiment with some of the technologies I demonstrated in the #vBrownBag session.

If you’d like to follow along on your own, I’ve uploaded the necessary files to the GitHub repo I created for the #vBrownBag session. Just have a look in the coreos-vagrant folder in that repository.

What You’ll Need

  • Vagrant (I tested with Vagrant 1.7.2)
  • VMware Fusion (I used Fusion 6.0.5 on OS X 10.9.5)
  • the VMware plugin for Vagrant
  • the CoreOS Vagrant box for the vmware_fusion provider (more on that in a moment)
  • the necessary supporting files (more on that in a moment as well)

I’m not going to cover installing Fusion, Vagrant, or the VMware provider for Vagrant, as those steps are reasonably well-documented by the respective vendors. I will, though, talk about getting the CoreOS box and the supporting files.

Getting the CoreOS Box

At the time of this writing, CoreOS supplied a Vagrant box for the vmware_fusion provider that you could install with a simple vagrant box add, like this:

vagrant box add http://stable.release.core-os.net/amd64-usr/current/coreos_production_vagrant_vmware_fusion.json

This command will download and install the Vagrant box you’ll need for use with VMware Fusion. Once it’s done, you’ll see the box listed when you run vagrant box list. I chose the stable branch for my testing; you could just as easily run builds from the alpha or beta branches by replacing “stable” in the URL above with the branch you want to use.

Getting the Supporting Files

There are two ways to get the supporting files you need:

  1. The GitHub repo I created for the “Docker and Friends” #vBrownBag session
  2. The GitHub repo created and maintained by CoreOS

You can choose to either clone these repos to your local system (using git clone and the GitHub HTTPS clone URL), or simply download a ZIP archive of the repository and unpack it locally. It’s up to you which approach you want to use.

Either repository will do what you need; I chose to create my own supporting files because I wanted to leverage some lessons learned in using YAML to create multi-machine Vagrant environments. For the rest of this post, I’ll assume you’re using the support files from my GitHub repo.

Preparation

Before you can just run vagrant up to turn up a cluster of systems on your laptop, there are a couple things you need to do first.

Provide VM Configuration Details

First, you’ll need to edit the servers.yml to provide Vagrant with the details on the VMs you’d like it to create. Here’s a snippet from the supplied servers.yml in the repository:

---
- name: coreos-01
  box: coreos-stable
  ram: 512
  vcpu: 1
  priv_ip: 192.168.254.101
  pub_ip: 172.16.5.11

Let’s walk through this real quick:

  • On the name line, supply the friendly name you’d like given to the VM spawned by Vagrant. This name will also be used as the hostname for the guest OS inside the VM.
  • On the box line, supply the name of the Vagrant box you downloaded earlier. You can verify the name by running vagrant box list; whatever’s listed there for the CoreOS box you downloaded, specify it here.
  • On the ram and vcpu lines, tell Vagrant how much RAM and how many vCPUs should be assigned to the VMs it will create.
  • On the priv_ip and pub_ip lines, supply IP addresses (both private and public) to be assigned to the CoreOS VMs.

This last item deserves a bit more discussion/explanation. CoreOS uses a variant of cloud-init (called coreos-cloudinit) that will read a YAML-formatted file (more on that in a moment in the “Provide Customization Information” section) to initialize (or customize) the CoreOS configuration. When used with Vagrant, coreos-cloudinit will read the IP address assigned in the Vagrantfile (not what is actually assigned to the interfaces) and put them in a file named /etc/environment. The values in this file are then used in a variety of other places, including in the etcd configuration. This has a couple of ramifications. First, you can’t leave the IP address out of the Vagrantfile, or else /etc/environment will get populated with the loopback address and nothing will work. Second, you can’t use DHCP on the networks (or at least the public network), or else there’s no guarantee that the address assigned to the interface will match the address specified in the Vagrantfile. (Unless you’d like to jump through a few hoops with static MAC addresses and DHCP configurations.) Remember, it’s the address supplied in the Vagrantfile that’s used by coreos-cloudinit, not the address actually assigned to the interface.

Repeat this block of information (making sure to include the “-” in front of “name” as per YAML syntax) for each VM you want Vagrant to spin up. The supplied servers.yml defines 3 VMs.

Provide Customization Information

In addition to providing the VM configuration details, you’ll also need to edit the user-data file that is used by coreos-cloudinit to customize CoreOS on startup. My GitHub repo has a sample user-data (just as it has a sample servers.yml), but you’ll need to supply an etcd discovery URL. This is documented inside the user-data file.

To obtain a discovery URL, just visit https://discovery.etcd.io/new. That URL will return a longer URL, which is your unique discovery URL. That’s the value you’ll supply in the user-data file where indicated by the comments.

Note that each cluster needs its own unique discovery URL, so everytime you vagrant destroy you’ll need to generate a new discovery URL.

Also, note here that the user-data file references a variable $public_ipv4. This variable references the static IP address assigned to the public network in the Vagrantfile. This is why the static IP address is so important—it’s used here in the configuration of both etcd and fleet, and if it is not correct or missing then everything breaks.

Once you’ve supplied the etcd discovery URL, you’re ready to start things up.

Starting up the VMs

In the same directory where the Vagrantfile, user-data, and servers.yml files reside, simply run vagrant up. Vagrant will instantiate and configure the VMs given the information in servers.yml, customize them via coreos-cloudinit using the information in user-data, and then communicate out to the etcd discovery URL to form an etcd cluster. Once Vagrant has completed, then you can use vagrant ssh <name> to connect to one of the VMs, then use commands like systemctl status etcd (to check on the status of etcd) or fleetctl list-machines (to see the membership of the cluster). Once fleet is working as expected, then you can also use fleetctl to launch Docker containers across the etcd cluster. (This is all stuff I demonstrated during the #vBrownBag session.)

Having an automated way to quickly and easily turn up a cluster of CoreOS systems running etcd, fleet, and Docker should make it much easier for newcomers to these technologies to begin to learn them and see how they could be put to work in their environments.

Additional Information

If you’d like some additional information on some of the technologies mentioned in this post, have a look at these articles:

A Quick Introduction to Vagrant
CoreOS Continued: etcd
CoreOS Continued: Fleet and Docker
Deploying CoreOS on OpenStack Using Heat
A Heat Template for Docker Containers

Metadata and Navigation

Be social and share this post!