Now that we have a model with a machine we can get to work. What we’re going to do is manually place a few components, then let the process take care of the rest!
We’ll need to add our other VMs. During this step you can add as many VMs as you’d like, the process is the same. In the following sections I’ll address how to scale out the components beyond this very small deployment.
In order to do so, we’ll need to add the other machines so Juju knows where we want to put our components. To do this, run the following command for each additional machine we’ve not yet told Juju about:
juju add-machine ssh:<user>@<ip>
As with before, replace
<ip> with the proper values from your setup.
juju status to verify you have all machines added and registered.
Kubernetes is comprised of a handful of components: etcd, easyrsa, kubernetes-master, kubernetes-worker, and flannel. When you complete a deployment of Kubernetes using conjure-up these components are installed, configured, and connected for you. Conjure-up uses Juju as the driver for these instructions and we’re doing this “manual” deployment manually with the Juju pieces directly.
First, we need to deploy EasyRSA and ETCD onto the machine. However, we don’t want to just smash them together, we’ll use LXD to separate and isolate these components.
juju deploy ~containers/easyrsa --to lxd:0
juju deploy ~containers/etcd --to 0
Depending on your networking, this will take a few moments to create LXC machines and setup the software. Eventually you’ll end up with a state where etcd is blocked. You don’t need to wait for this to complete before issuing the following commands:
juju deploy ~containers/kubernetes-master --to 0
juju deploy ~containers/kubernetes-worker --to 1
This will combine these two components on the single machine. We’re not going to use LXD for these components since it won’t be routable from outside the VM without messing with the network configuration. As such, we’re deploying
--to machine 0, the components will be directly accessible through the VMs IP address.
After a few moments, you’ll find something like the following in
As you can see, there’s still items executing. We could wait for these to complete, but if you’re as impatient as I am, then be thankful we live in an asynchronous world and press forward! The final step is to glue all these components together (and deploy the SDN). To do that, we’ll take the
kubernetes-core bundle, which is a super light weight Kubernetes cluster, and deploy that now. It’ll skip over any component you’ve already deployed, add any components not yet deployed, and execute all the required relationships.
juju deploy kubernetes-core
The output for this is pretty verbose, and should look something like the following:
This is to be expected. We see in several places Juju skips over components we’ve already deployed, adds things (flannel) that we’re missing, and finally adds all the relations for these components. This is how we resolve the etcd “blocked” message that it’s missing a certificate authority. You’ll notice that
etcd:certificates is connected to
easyrsa:client which will provider certs for etcd!
Eventually, after running
juju status for a few mins you should end up with the following. A completely deployed Kubernetes cluster.
From this point forward, we’ll need to get the credentials for the cluster. This is done automatically for you with
conjure-up. With this method you’ll just need to issue the following
juju scp kubernetes-master/0:config ~/.kube/config
If you already have a Kubernetes config file, choose another path, like
~/.kube/config.cdk and make sure you use
export KUBECONFIG=$HOME/.kube/config.cdk to use the new configuration file.
For the final touch, I wanted to show how to scale this up. Ideally, you’d want to use a public cloud, private cloud (VMWare, OpenStack), or MAAS for bare metal. The manual provider is just that — very manual. That said, if you have more VMs you can add them and scale the applications to spread across them. I’m going to add another machine and use it for both etcd and kubernetes-worker.
juju add-machine ssh:email@example.com
juju add-unit -n2 etcd --to 1,2
juju add-unit etcd --to 2
The result will be a three node etcd and two nodes for Kubernetes workloads. Again,
juju status will show you the status of the cluster at anytime. Eventually everything will converge on
idle. Once this is done you’ve scaled out the deployment. From here you can continue to add VMs, redeploy everything again, or actually start using Kubernetes!