How To Create a Kubernetes Cluster Using Kubeadm on Ubuntu 16.04

Your cluster will include the following physical resources:

• One master node

  The master node (a node in Kubernetes refers to a server) is responsible for managing the state of the cluster. It runs Etcd, which stores cluster data among components that schedule workloads to worker nodes.

• Two worker nodes

  Worker nodes are the servers where your workloads (i.e. containerized applications and services) will run. A worker will continue to run your workload once they’re assigned to it, even if the master goes down once scheduling is complete. A cluster’s capacity can be increased by adding workers.

After completing this guide, you will have a cluster ready to run containerized applications, provided that the servers in the cluster have sufficient CPU and RAM resources for your applications to consume. Almost any traditional Unix application including web applications, databases, daemons, and command line tools can be containerized and made to run on the cluster. The cluster itself will consume around 300-500MB of memory and 10% of CPU on each node.

Once the cluster is set up, you will deploy the web server Nginx to it to ensure that it is running workloads correctly.

• An SSH key pair on your local Linux/Mac OS/BSD machine. If you haven’t used SSH keys before, you can learn how to set them up by following this explanation of how to set up SSH keys on your local machine.

• Three servers running Ubuntu 16.04 with at least 2GB RAM and 2 vCPUs each. You should be able to SSH into each server as the root user with your SSH key pair.

• Ansible installed on your local machine. If you’re running Ubuntu 16.04 as your OS, follow the “Step 1 - Installing Ansible” section in How to Install and Configure Ansible on Ubuntu 16.04 to install Ansible. For installation instructions on other platforms like Mac OS X or CentOS, follow the official Ansible installation documentation.

• Familiarity with Ansible playbooks. For review, check out Configuration Management 101: Writing Ansible Playbooks.

• Knowledge of how to launch a container from a Docker image. Look at “Step 5 — Running a Docker Container” in How To Install and Use Docker on Ubuntu 16.04 if you need a refresher.

In this section, you will create a directory on your local machine that will serve as your workspace. You will also configure Ansible locally so that it can communicate with and execute commands on your remote servers. To do this, you will create a hosts file containing inventory information such as the IP addresses of your servers and the groups that each server belongs to.

Out of your three servers, one will be the master with an IP displayed as master_ip. The other two servers will be workers and will have the IPs worker_1_ip and worker_2_ip.

Create a directory named ~/kube-cluster in the home directory of your local machine and cd into it:

mkdir ~/kube-cluster
cd ~/kube-cluster

This directory will be your workspace for the rest of the tutorial and will contain all of your Ansible playbooks. It will also be the directory inside which you will run all local commands.

Create a file named ~/kube-cluster/hosts using nano or your favorite text editor:

nano ~/kube-cluster/hosts

Add the following text to the file, which will specify information about the logical structure of your cluster:

[masters]
master ansible_host=master_ip ansible_user=root

[workers]
worker1 ansible_host=worker_1_ip ansible_user=root
worker2 ansible_host=worker_2_ip ansible_user=root

[all:vars]
ansible_python_interpreter=/usr/bin/python3

You may recall that inventory files in Ansible are used to specify server information such as IP addresses, remote users, and groupings of servers to target as a single unit for executing commands. ~/kube-cluster/hosts will be your inventory file and you’ve added two Ansible groups (masters and workers) to it specifying the logical structure of your cluster.

In the masters group, there is a server entry named “master” that lists the master node’s IP (master_ip) and specifies that Ansible should run remote commands as the root user.

Similarly, in the workers group, there are two entries for the worker servers (worker_1_ip and worker_2_ip) that also specify the ansible_user as root.

The last line of the file tells Ansible to use the remote servers’ Python 3 interpreters for its management operations.

Save and close the file after you’ve added the text.

Having set up the server inventory with groups, let’s move on to installing operating system level dependencies and creating configuration settings.

In this section you will create a non-root user with sudo privileges on all servers so that you can SSH into them manually as an unprivileged user. This can be useful if, for example, you would like to see system information with commands such as top/htop, view a list of running containers, or change configuration files owned by root. These operations are routinely performed during the maintenance of a cluster, and using a non-root user for such tasks minimizes the risk of modifying or deleting important files or unintentionally performing other dangerous operations.

Create a file named ~/kube-cluster/initial.yml in the workspace:

nano ~/kube-cluster/initial.yml

Next, add the following play to the file to create a non-root user with sudo privileges on all of the servers. A play in Ansible is a collection of steps to be performed that target specific servers and groups. The following play will create a non-root sudo user:

- hosts: all
  become: yes
  tasks:
    - name: create the 'ubuntu' user
      user: name=ubuntu append=yes state=present createhome=yes shell=/bin/bash

    - name: allow 'ubuntu' to have passwordless sudo
      lineinfile:
        dest: /etc/sudoers
        line: 'ubuntu ALL=(ALL) NOPASSWD: ALL'
        validate: 'visudo -cf %s'

    - name: set up authorized keys for the ubuntu user
      authorized_key: user=ubuntu key="{{item}}"
      with_file:
        - ~/.ssh/id_rsa.pub

Here’s a breakdown of what this playbook does:

• Creates the non-root user ubuntu.

• Configures the sudoers file to allow the ubuntu user to run sudo commands without a password prompt.

• Adds the public key in your local machine (usually ~/.ssh/id_rsa.pub) to the remote ubuntu user’s authorized key list. This will allow you to SSH into each server as the ubuntu user.

Save and close the file after you’ve added the text.

Next, execute the playbook by locally running:

ansible-playbook -i hosts ~/kube-cluster/initial.yml

The command will complete within two to five minutes. On completion, you will see output similar to the following:

Output
PLAY [all] ****

TASK [Gathering Facts] ****
ok: [master]
ok: [worker1]
ok: [worker2]

TASK [create the 'ubuntu' user] ****
changed: [master]
changed: [worker1]
changed: [worker2]

TASK [allow 'ubuntu' user to have passwordless sudo] ****
changed: [master]
changed: [worker1]
changed: [worker2]

TASK [set up authorized keys for the ubuntu user] ****
changed: [worker1] => (item=ssh-rsa AAAAB3...
changed: [worker2] => (item=ssh-rsa AAAAB3...
changed: [master] => (item=ssh-rsa AAAAB3...

PLAY RECAP ****
master                     : ok=5    changed=4    unreachable=0    failed=0   
worker1                    : ok=5    changed=4    unreachable=0    failed=0   
worker2                    : ok=5    changed=4    unreachable=0    failed=0   

Now that the preliminary setup is complete, you can move on to installing Kubernetes-specific dependencies.

In this section, you will install the operating system level packages required by Kubernetes with Ubuntu’s package manager. These packages are:

• Docker - a container runtime. It is the component that runs your containers. Support for other runtimes such as rkt is under active development in Kubernetes.

• kubeadm - a CLI tool that will install and configure the various components of a cluster in a standard way.

• kubelet - a system service/program that runs on all nodes and handles node-level operations.

• kubectl - a CLI tool used for issuing commands to the cluster through its API Server.

Create a file named ~/kube-cluster/kube-dependencies.yml in the workspace:

nano ~/kube-cluster/kube-dependencies.yml

Add the following plays to the file to install these packages to your servers:

- hosts: all
  become: yes
  tasks:
   - name: install Docker
     apt:
       name: docker.io
       state: present
       update_cache: true

   - name: install APT Transport HTTPS
     apt:
       name: apt-transport-https
       state: present

   - name: add Kubernetes apt-key
     apt_key:
       url: https://packages.cloud.google.com/apt/doc/apt-key.gpg
       state: present

   - name: add Kubernetes' APT repository
     apt_repository:
      repo: deb http://apt.kubernetes.io/ kubernetes-xenial main
      state: present
      filename: 'kubernetes'

   - name: install kubelet
     apt:
       name: kubelet=1.14.0-00
       state: present
       update_cache: true

   - name: install kubeadm
     apt:
       name: kubeadm=1.14.0-00
       state: present

- hosts: master
  become: yes
  tasks:
   - name: install kubectl
     apt:
       name: kubectl=1.14.0-00
       state: present
       force: yes

The first play in the playbook does the following:

• Installs Docker, the container runtime.

• Installs apt-transport-https, allowing you to add external HTTPS sources to your APT sources list.

• Adds the Kubernetes APT repository’s apt-key for key verification.

• Adds the Kubernetes APT repository to your remote servers’ APT sources list.

• Installs kubelet and kubeadm.

The second play consists of a single task that installs kubectl on your master node.

Save and close the file when you are finished.

Next, execute the playbook by locally running:

ansible-playbook -i hosts ~/kube-cluster/kube-dependencies.yml

On completion, you will see output similar to the following:

Output
PLAY [all] ****

TASK [Gathering Facts] ****
ok: [worker1]
ok: [worker2]
ok: [master]

TASK [install Docker] ****
changed: [master]
changed: [worker1]
changed: [worker2]

TASK [install APT Transport HTTPS] *****
ok: [master]
ok: [worker1]
changed: [worker2]

TASK [add Kubernetes apt-key] *****
changed: [master]
changed: [worker1]
changed: [worker2]

TASK [add Kubernetes' APT repository] *****
changed: [master]
changed: [worker1]
changed: [worker2]

TASK [install kubelet] *****
changed: [master]
changed: [worker1]
changed: [worker2]

TASK [install kubeadm] *****
changed: [master]
changed: [worker1]
changed: [worker2]

PLAY [master] *****

TASK [Gathering Facts] *****
ok: [master]

TASK [install kubectl] ******
ok: [master]

PLAY RECAP ****
master                     : ok=9    changed=5    unreachable=0    failed=0   
worker1                    : ok=7    changed=5    unreachable=0    failed=0  
worker2                    : ok=7    changed=5    unreachable=0    failed=0  

After execution, Docker, kubeadm, and kubelet will be installed on all of the remote servers. kubectl is not a required component and is only needed for executing cluster commands. Installing it only on the master node makes sense in this context, since you will run kubectl commands only from the master. Note, however, that kubectl commands can be run from any of the worker nodes or from any machine where it can be installed and configured to point to a cluster.

All system dependencies are now installed. Let’s set up the master node and initialize the cluster.

In this section, you will set up the master node. Before creating any playbooks, however, it’s worth covering a few concepts such as Pods and Pod Network Plugins, since your cluster will include both.

A pod is an atomic unit that runs one or more containers. These containers share resources such as file volumes and network interfaces in common. Pods are the basic unit of scheduling in Kubernetes: all containers in a pod are guaranteed to run on the same node that the pod is scheduled on.

Each pod has its own IP address, and a pod on one node should be able to access a pod on another node using the pod’s IP. Containers on a single node can communicate easily through a local interface. Communication between pods is more complicated, however, and requires a separate networking component that can transparently route traffic from a pod on one node to a pod on another.

This functionality is provided by pod network plugins. For this cluster, you will use Flannel, a stable and performant option.

Create an Ansible playbook named master.yml on your local machine:

nano ~/kube-cluster/master.yml

Add the following play to the file to initialize the cluster and install Flannel:

- hosts: master
  become: yes
  tasks:
    - name: initialize the cluster
      shell: kubeadm init --pod-network-cidr=10.244.0.0/16 >> cluster_initialized.txt
      args:
        chdir: $HOME
        creates: cluster_initialized.txt

    - name: create .kube directory
      become: yes
      become_user: ubuntu
      file:
        path: $HOME/.kube
        state: directory
        mode: 0755

    - name: copy admin.conf to user's kube config
      copy:
        src: /etc/kubernetes/admin.conf
        dest: /home/ubuntu/.kube/config
        remote_src: yes
        owner: ubuntu

    - name: install Pod network
      become: yes
      become_user: ubuntu
      shell: kubectl apply -f https://raw.githubusercontent.com/coreos/flannel/a70459be0084506e4ec919aa1c114638878db11b/Documentation/kube-flannel.yml >> pod_network_setup.txt
      args:
        chdir: $HOME
        creates: pod_network_setup.txt

Here’s a breakdown of this play:

• The first task initializes the cluster by running kubeadm init. Passing the argument --pod-network-cidr=10.244.0.0/16 specifies the private subnet that the pod IPs will be assigned from. Flannel uses the above subnet by default; we’re telling kubeadm to use the same subnet.

• The second task creates a .kube directory at /home/ubuntu. This directory will hold configuration information such as the admin key files, which are required to connect to the cluster, and the cluster’s API address.

• The third task copies the /etc/kubernetes/admin.conf file that was generated from kubeadm init to your non-root user’s home directory. This will allow you to use kubectl to access the newly-created cluster.

• The last task runs kubectl apply to install Flannel. kubectl apply -f descriptor.[yml|json] is the syntax for telling kubectl to create the objects described in the descriptor.[yml|json] file. The kube-flannel.yml file contains the descriptions of objects required for setting up Flannel in the cluster.

Save and close the file when you are finished.

Execute the playbook locally by running:

ansible-playbook -i hosts ~/kube-cluster/master.yml

On completion, you will see output similar to the following:

Output
PLAY [master] ****

TASK [Gathering Facts] ****
ok: [master]

TASK [initialize the cluster] ****
changed: [master]

TASK [create .kube directory] ****
changed: [master]

TASK [copy admin.conf to user's kube config] *****
changed: [master]

TASK [install Pod network] *****
changed: [master]

PLAY RECAP ****
master                     : ok=5    changed=4    unreachable=0    failed=0  

To check the status of the master node, SSH into it with the following command:

ssh ubuntu@master_ip

Once inside the master node, execute:

kubectl get nodes

You will now see the following output:

Output
NAME      STATUS    ROLES     AGE       VERSION
master    Ready     master    1d        v1.14.0

The output states that the master node has completed all initialization tasks and is in a Ready state from which it can start accepting worker nodes and executing tasks sent to the API Server. You can now add the workers from your local machine.

Adding workers to the cluster involves executing a single command on each. This command includes the necessary cluster information, such as the IP address and port of the master’s API Server, and a secure token. Only nodes that pass in the secure token will be able join the cluster.

Navigate back to your workspace and create a playbook named workers.yml:

nano ~/kube-cluster/workers.yml

Add the following text to the file to add the workers to the cluster:

- hosts: master
  become: yes
  gather_facts: false
  tasks:
    - name: get join command
      shell: kubeadm token create --print-join-command
      register: join_command_raw

    - name: set join command
      set_fact:
        join_command: "{{ join_command_raw.stdout_lines[0] }}"


- hosts: workers
  become: yes
  tasks:
    - name: join cluster
      shell: "{{ hostvars['master'].join_command }} >> node_joined.txt"
      args:
        chdir: $HOME
        creates: node_joined.txt

Here’s what the playbook does:

• The first play gets the join command that needs to be run on the worker nodes. This command will be in the following format:kubeadm join --token <token> <master-ip>:<master-port> --discovery-token-ca-cert-hash sha256:<hash>. Once it gets the actual command with the proper token and hash values, the task sets it as a fact so that the next play will be able to access that info.

• The second play has a single task that runs the join command on all worker nodes. On completion of this task, the two worker nodes will be part of the cluster.

Save and close the file when you are finished.

Execute the playbook by locally running:

ansible-playbook -i hosts ~/kube-cluster/workers.yml

On completion, you will see output similar to the following:

Output
PLAY [master] ****

TASK [get join command] ****
changed: [master]

TASK [set join command] *****
ok: [master]

PLAY [workers] *****

TASK [Gathering Facts] *****
ok: [worker1]
ok: [worker2]

TASK [join cluster] *****
changed: [worker1]
changed: [worker2]

PLAY RECAP *****
master                     : ok=2    changed=1    unreachable=0    failed=0   
worker1                    : ok=2    changed=1    unreachable=0    failed=0  
worker2                    : ok=2    changed=1    unreachable=0    failed=0  

With the addition of the worker nodes, your cluster is now fully set up and functional, with workers ready to run workloads. Before scheduling applications, let’s verify that the cluster is working as intended.

A cluster can sometimes fail during setup because a node is down or network connectivity between the master and worker is not working correctly. Let’s verify the cluster and ensure that the nodes are operating correctly.

You will need to check the current state of the cluster from the master node to ensure that the nodes are ready. If you disconnected from the master node, you can SSH back into it with the following command:

ssh ubuntu@master_ip

Then execute the following command to get the status of the cluster:

kubectl get nodes

You will see output similar to the following:

Output
NAME      STATUS    ROLES     AGE       VERSION
master    Ready     master    1d        v1.14.0
worker1   Ready     <none>    1d        v1.14.0
worker2   Ready     <none>    1d        v1.14.0

If all of your nodes have the value Ready for STATUS, it means that they’re part of the cluster and ready to run workloads.

If, however, a few of the nodes have NotReady as the STATUS, it could mean that the worker nodes haven’t finished their setup yet. Wait for around five to ten minutes before re-running kubectl get node and inspecting the new output. If a few nodes still have NotReady as the status, you might have to verify and re-run the commands in the previous steps.

Now that your cluster is verified successfully, let’s schedule an example Nginx application on the cluster.

You can now deploy any containerized application to your cluster. To keep things familiar, let’s deploy Nginx using Deployments and Services to see how this application can be deployed to the cluster. You can use the commands below for other containerized applications as well, provided you change the Docker image name and any relevant flags (such as ports and volumes).

Still within the master node, execute the following command to create a deployment named nginx:

kubectl create deployment nginx --image=nginx

A deployment is a type of Kubernetes object that ensures there’s always a specified number of pods running based on a defined template, even if the pod crashes during the cluster’s lifetime. The above deployment will create a pod with one container from the Docker registry’s Nginx Docker Image.

Next, run the following command to create a service named nginx that will expose the app publicly. It will do so through a NodePort, a scheme that will make the pod accessible through an arbitrary port opened on each node of the cluster:

kubectl expose deploy nginx --port 80 --target-port 80 --type NodePort

Services are another type of Kubernetes object that expose cluster internal services to clients, both internal and external. They are also capable of load balancing requests to multiple pods, and are an integral component in Kubernetes, frequently interacting with other components.

Run the following command:

kubectl get services

This will output text similar to the following:

Output
NAME         TYPE        CLUSTER-IP       EXTERNAL-IP           PORT(S)        AGE
kubernetes   ClusterIP   10.96.0.1        <none>                443/TCP        1d
nginx        NodePort    10.109.228.209   <none>                80:nginx_port/TCP   40m

From the third line of the above output, you can retrieve the port that Nginx is running on. Kubernetes will assign a random port that is greater than 30000 automatically, while ensuring that the port is not already bound by another service.

To test that everything is working, visit http://worker_1_ip:nginx_port or http://worker_2_ip:nginx_port through a browser on your local machine. You will see Nginx’s familiar welcome page.

If you would like to remove the Nginx application, first delete the nginx service from the master node:

kubectl delete service nginx

Run the following to ensure that the service has been deleted:

kubectl get services

You will see the following output:

Output
NAME         TYPE        CLUSTER-IP       EXTERNAL-IP           PORT(S)        AGE
kubernetes   ClusterIP   10.96.0.1        <none>                443/TCP        1d

Then delete the deployment:

kubectl delete deployment nginx

Run the following to confirm that this worked:

kubectl get deployments

Output
No resources found.

In this guide, you’ve successfully set up a Kubernetes cluster on Ubuntu 16.04 using Kubeadm and Ansible for automation.

If you’re wondering what to do with the cluster now that it’s set up, a good next step would be to get comfortable deploying your own applications and services onto the cluster. Here’s a list of links with further information that can guide you in the process:

• Dockerizing applications - lists examples that detail how to containerize applications using Docker.

• Pod Overview - describes in detail how Pods work and their relationship with other Kubernetes objects. Pods are ubiquitous in Kubernetes, so understanding them will facilitate your work.

• Deployments Overview - this provides an overview of deployments. It is useful to understand how controllers such as deployments work since they are used frequently in stateless applications for scaling and the automated healing of unhealthy applications.

• Services Overview - this covers services, another frequently used object in Kubernetes clusters. Understanding the types of services and the options they have is essential for running both stateless and stateful applications.

Other important concepts that you can look into are Volumes, Ingresses and Secrets, all of which come in handy when deploying production applications.

Kubernetes has a lot of functionality and features to offer. The Kubernetes Official Documentation is the best place to learn about concepts, find task-specific guides, and look up API references for various objects.