Reading Time: 11 minutes
In my first two articles of this series, you looked at deploying a cluster of nodes and running a simple two-service application across them with Docker Compose on the Civo platform. You then took a look at some rather rough cloud theory in the third installment, providing some foundation upon which to consider your application development and deployment in order to create scaleable, fault-tolerant services.
In this article, you’ll take your exploration of cloud deployments to another level, by redeploying the previous simple application using Google’s now famous Kubernetes platform.
Let’s start by creating three small Ubuntu nodes on Civo using the same steps from Part I of this series. Give them the names
kube-worker2. For the purposes of this article, be sure to give them a
root user. You may want to do this differently with your real applications, but it makes life a little easier for this demonstration.
Once they have resolved their IP addresses, go ahead and provision Docker to each server as before.
Since you’re not going to be using Docker Swarm this time, do not link your instances together. Kubernetes provides its own means to do this, which you’ll see shortly.
Now that your cluster of servers is up and accessible, go ahead and SSH into the
kube-manager instance using Docker Machine.
# docker-machine ssh kube-manager
Once connected, the first thing you’ll need to do is prepare the server for the Kubernetes install. You do this with:
root@kube-manager:~# sudo apt-get update && sudo apt-get install -y apt-transport-https
This ensures that the current Ubuntu install is up to date and enables
apt secure repository support.
sudoon each call here to minimize difficulties. Since you’re logged in as the root user, you shouldn’t need to do this, but it also doesn’t hurt.
Adding the Kubernetes Repository
By default, your Ubuntu instance has no idea how or where the Kubernetes packages live. You’ll need to fix that. Since Ubuntu doesn’t simply allow packages to be installed from just any location, you’ll first need to add the Kubernetes repository authentication key.
root@kube-manager:~# sudo curl -s https://packages.cloud.google.com/apt/doc/apt-key.gpg | apt-key add
Next, you need to add the Kubernetes repository path to Ubuntu’s repository source list. You can do this by first creating a new list file in the
apt sources directory and supplying it with the package location. To do this, run
nano (or your favorite text editor) with the path of the new list file.
root@kube-manager:~# sudo nano /etc/apt/sources.list.d/kubernetes.list
Then add the following package path as its content:
deb http://apt.kubernetes.io/ kubernetes-xenial main
Be sure to save and exit this file.
apt so that it loads this new repository list, like so:
root@kube-manager:~# sudo apt-get update
Installing the Kubernetes Binaries
You’re now ready to install Kubernetes itself. Don’t worry, the process is surprisingly painless.
root@kube-manager:~# sudo apt-get install -y kubelet kubeadm kubectl kubernetes-cni
That’s it! Kubernetes is installed! Was that so hard?
The previous line installed each of the Kubernetes interactive tools, as well as the Container Network Interface. Everything you need to run your app is now available; you simply need to configure it.
Before you do this, however, you’ll first need to repeat the above process to install Kubernetes on each of the workers. Go ahead and do that now.
Configuring the Master Node
As with Docker Swarm, Kubernetes uses a master/slave node formation. The master node is responsible for sending commands to each of the slave nodes (and to itself). When first setting up your cluster, you need to manually elect a master before doing anything else. While SSH’d into the
kube-manager server, go ahead and issue the following command:
root@kube-manager:~# sudo kubeadm init --pod-network-cidr 10.244.0.0/16
kubeadm init command prepares the machine as a Kubernetes master node. The
--pod-network-cidr flag is required specifically for this article, as you’ll be using the Flannel pod network. If you don’t use Flannel in your actual projects, you won’t need to supply it. Don’t worry about this for now, as all will be explained later.
Running the above command should print out something similar to the following:
Your Kubernetes master has initialized successfully! To start using your cluster, you need to run (as a regular user): mkdir -p $HOME/.kube sudo cp -i /etc/kubernetes/admin.conf $HOME/.kube/config sudo chown $(id -u):$(id -g) $HOME/.kube/config You should now deploy a pod network to the cluster. Run “kubectl apply -f [podnetwork].yaml” with one of the options listed at: http://kubernetes.io/docs/admin/addons/ You can now join any number of machines by running the following on each node as root: kubeadm join –token lzyq4s.jjlcg7hm8oqk9fzl --discovery-token-ca-cert-hash sha256:63796b2147c2450f.[snip].d3d33
Wonderful! Kubernetes has written half of this article for me!
Let’s go ahead and finish preparing the master node.
Node Configuration Directory
Kubernetes uses a configuration storage directory for placing data about your environment. As the previous output stated, this must be done as a user other than the root user. To do this, you’ll need to create one while ensuring it has sudo capabilities.
root@kube-manager:~# adduser kubeuser root@kube-manager:~# usermod -aG sudo kubeuser
When adding the user, you’ll be prompted to supply a password and some details. You’ll need the password whenever you call
sudo while in the user context. The other details can simply be skipped by pressing Enter, if you wish. Next, change your user context to this new user:
root@kube-manager:~# su – kubeuser
You should see the user’s name form part of the command prompt, like this:
You can now issue the remaining commands to set up your Kubernetes master.
kubeuser@kube-manager:~# mkdir -p $HOME/.kube kubeuser@kube-manager:~# sudo cp -i /etc/kubernetes/admin.conf $HOME/.kube/config kubeuser@kube-manager:~# sudo chown $(id -u):$(id -g) $HOME/.kube/config
That was pretty much it for the master node. However, there is still one outstanding requirement to configure your Kubernetes installation before you can establish your cluster: the pod network.
If you recall from Part 2 of this series, you deployed two containers using a Docker Compose file. The YAML inside that file detailed a service called
service (rather stupidly in retrospect). Later in the same article, an inspection of the deployed service contained a
Network array which detailed a network also called
When deploying Docker containers into a Swarm cluster, the cluster manager creates an overlay network, placing each container within it. An overlay network is a virtual network of sorts that allow containers to communicate in the same address space. Other types of network can also be created, if you know what you’re doing.
Using Swarm and Docker Compose, it is possible and recommended to create custom networks that group services together. Each container can belong to more than one network if necessary. The result is then a means to protect those containers that do not need to be accessed from the outside world, while remaining contactable by the rest of your service architecture.
With Kubernetes, the same network configuration requirement applies. The considerable difference is that, while the Docker Engine manages networks for Docker Swarm, a Kubernetes cluster typically requires the use of third-party solutions.
Flannel is a third-party network solution for Kubernetes. Of the available options for providing a pod network, Flannel is by far the simplest. That’s why you’ll use it here, for the purpose of this article.
Installing Flannel onto your master node is super simple. Simply run the following:
kubeuser@kube-manager:~# sudo kubectl apply -f https://raw.githubusercontent.com/coreos/flannel/master/Documentation/kube-flannel.yml kubeuser@kube-manager:~# sudo kubectl apply -f https://raw.githubusercontent.com/coreos/flannel/master/Documentation/k8s-manifests/kube-flannel-rbac.yml
These commands run containers within Kubernetes
pods that manage traffic to your own pods. If you inspect your current running Kubernetes pod list, you should see something like this:
kubeuser@kube-manager:~# sudo kubectl get pods --all-namespaces NAMESPACE NAME READY STATUS RESTARTS AGE kube-system etcd-kube-manager 1/1 Running 0 1m kube-system kube-apiserver-kube-manager 1/1 Running 0 1m kube-system kube-controller-manager-kube-manager 1/1 Running 0 1m kube-system kube-dns-86f4d74b45-48bmx 0/3 ContainerCreating 0 2m kube-system kube-flannel-ds-zdz59 1/1 Running 0 32s kube-system kube-proxy-txjh8 1/1 Running 0 2m kube-system kube-scheduler-kube-manager 1/1 Running 0 1m
Don’t worry if it’s not exactly the same, so long as you see pods listed.
kube-flannel pod was started as part of the Flannel setup. Flannel uses
etcd to keep track of your pods and their relationships to one another. The result is a simple overlay network that works with minimal fuss.
Creating the Cluster
Now you’re ready to join your workers to the manager. Using the
kubeadm join command output when the master node was initialized, SSH into both worker machines and run it in the console. You should be presented with a
In order to check that all is well, simply call this line in the master node, ensuring you are in the kubeuser context.
kubeuser@kube-manager:~# kubectl get nodes
The resulting output should look a little like this:
NAME STATUS ROLES AGE VERSION kube-manager Ready master 24m v1.10.0 kube-worker1 Ready <none> 27m v1.10.0 kube-worker2 Ready <none> 27m v1.10.0
You may be tempted to simply spin up one machine and use it as a master with no slaves. This will not work. Kubernetes will not allow pods to be scheduled without at least one additional slave node attached to the cluster.
Deploying the Application
Just as with Docker Swarm, Kubernetes also makes use of YAML files to deploy applications. While you can easily deploy a whole cluster with a single YAML file for Docker Swarm, it is preferable to break Kubernetes deployments into smaller chunks, providing greater flexibility and making it easier to focus on smaller details.
Deploying the Backend Application
For the backend application, you’ll deploy the
phpinf image from Part I as a Kubernetes pod. Pods are an abstraction of one or more containers, providing port, volume, and other configuration data. This is a more advanced concept than a simple container, as it means multiple containers can exist together as a pseudo container, providing greater flexibility.
To deploy the
phpinf application, you’ll need a pod configuration file. On the master server, create a new document called
backend.yaml and populate it with the following content:
apiVersion: apps/v1 kind: Deployment metadata: name: service spec: selector: matchLabels: app: php-service srv: backend replicas: 3 template: metadata: labels: app: php-service srv: backend spec: containers: - name: php-service image: "leesylvester/phpinf" ports: - name: http containerPort: 80
This seems quite a bit more complicated than the Docker Swarm example, I’m sure. The important parts to take note of here are the
replica count, the
image, and the
In Kubernetes, much of the configuration you’ll work with involves the use of labels. Here, the container’s port value is associated with the label
http. Typically, container port specification in configuration files presents no additional functionality, but by naming the port, this value can be referenced in other configurations by name, keeping deployments truly dynamic.
With this file created and saved, let’s deploy the image to the cluster.
kubeuser@kube-manager:~# kubectl create -f backend.yaml
You can check for its status by recalling the current pod list.
kubeuser@kube-manager:~# kubectl get pods NAME READY STATUS RESTARTS AGE service-6c75bd4c8c-lhplx 1/1 Running 0 2m service-6c75bd4c8c-2q76f 1/1 Running 0 2m service-6c75bd4c8c-47jw9 1/1 Running 0 2m
If your status says something like
ContainerCreating, then the pod is still downloading and preparing to be run. Wait a few seconds and try again.
So you now have three instances of the backend pod running and ready to receive requests, but in its current state, there is no way to do so. Kubernetes pods are inaccessible until a
service is deployed, providing configuration for port access.
Once again, to create a service, you’ll need to provide a YAML configuration file. Do this now by creating a file called
service.yaml and populating it with the following:
apiVersion: v1 kind: Service metadata: name: service spec: selector: app: php-service srv: backend ports: - protocol: TCP port: 80 targetPort: http
Again, you can launch the service by simply entering:
kubeuser@kube-manager:~# kubectl create -f service.yaml
A list of running services can be displayed by entering:
kubeuser@kube-manager:~# kubectl get services
Deploying the Frontend Application
Just as with Docker Swarm in Part I, the
backend service now has an internal port ready to receive requests. What you need now is the
frontend load balancer app to forward calls from the public internet.
As with the backend application, you’ll need a YAML file. This time, it will include both the deployment and service configuration, together.
apiVersion: v1 kind: Service metadata: name: frontend spec: selector: app: nginx-lb srv: frontend ports: - protocol: "TCP" port: 80 targetPort: lbhttp nodePort: 30001 type: NodePort --- apiVersion: apps/v1 kind: Deployment metadata: name: frontend spec: selector: matchLabels: app: nginx-lb srv: frontend replicas: 3 template: metadata: labels: app: nginx-lb srv: frontend spec: containers: - name: nginx image: "leesylvester/civolb" ports: - containerPort: 80 name: lbhttp lifecycle: preStop: exec: command: ["/usr/sbin/nginx","-s","quit"]
If you create this service and query the pods, you should now see both sets of containers running.
kubeuser@kube-manager:~# kubectl create -f frontend.yaml kubeuser@kube-manager:~# kubectl get pods NAME READY STATUS RESTARTS AGE frontend-64bcc57c9d-nvrbg 1/1 Running 0 1m frontend-64bcc57c9d-rqmlb 1/1 Running 0 1m frontend-64bcc57c9d-5mpgc 1/1 Running 0 1m service-6c75bd4c8c-lhplx 1/1 Running 0 14m service-6c75bd4c8c-2q76f 1/1 Running 0 14m service-6c75bd4c8c-47jw9 1/1 Running 0 14m
Now listing the services will show both the frontend and backend overlay networks.
kubeuser@kube-manager:~# kubectl get services NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE frontend NodePort 10.108.194.216 <none> 80:30001/TCP 3m kubernetes ClusterIP 10.96.0.1 <none> 443/TCP 16m service ClusterIP 10.110.192.42 <none> 80/TCP 15m
That’s it! Your application is now running.
To see it in action, call up any one of the external IPs from your cluster and visit port 30001 in the browser. As with the Docker Swarm implementation, refreshing the browser should show a different service id and IP each time.
When requests are made to a node (a given server in the cluster), the frontend service routes the request to the NGINX load balancer. This request may not go to the load balancer on the same server. Likewise, the request from the load balancer to the backend application may also go to a different server via the backend service. It is the service’s job to identify a suitable pod to handle the request, whichever node it happens to be running on.
There are ways to ensure that requests passed to a server remain within the same node if that is your preference, provided the server contains the service it is trying to contact. However, that is beyond the scope of this article. If you wish to read up on that further, try searching for
Kubernetes is rich with confguration capabilities, of which
externalTrafficPolicy is but a small part, so I would advise getting a good book on the subject if you really want to master it. I would thoroughly recommend Kubernetes in Action by Marko Lukša, as this provides a wealth of real-world information that is easy to understand and follow.
Despite all that Kubernetes can do, however, there is still something you can do external to your cluster to ensure your application is load balanced and fault tolerant: implement an external load balancer.
The Civo Load Balancer
Although Kubernetes ensures requests to each pod service is load balanced, the actual request to the cluster still needs to hit a node. If you assign a domain name to the master node and send all traffic to it, then the frontend service on that node will be hit with 100 percent of all traffic. What’s more, if the node fails, the whole application becomes unavailable, which is not good for business.
The solution is to leverage Civo’s Load Balancer service.
By supplying a domain and assigning each of the cluster nodes, they can be sufficiently balanced using either a
IP address hash policy. The service also provides a means to ping a node’s health via a given endpoint, which you can expose in your application.
This way, if a node fails, the load balancer can be sure not to route any further traffic to it until it comes back online.
In this article, you took a simple Docker Swarm application and transitioned it to big-boy deployment on the Civo platform. While you have only just scratched the surface of what Kubernetes can do, I hope it has at least provided a sufficient starting point for your own projects and shown that Kubernetes is not so complicated and scary to work with.
In the next article of this series, you’ll take a look at some of the additional benefits of Kubernetes.
You can find the previous articles here: