Expert Training in Kubernetes and Rancher
Join our free online training sessions to learn more about Kubernetes, containers, and Rancher.Sign up here
Kubernetes installation is considered one of the toughest problems for operators and DevOps. Because Kubernetes can run on various platforms and operating systems, there are so many factors that should be considered during the installation process. In this post, I am going to introduce a new, lightweight tool for installing Kubernetes that supports installation on bare-metal and virtualized servers. Rancher Kubernetes Engine (RKE) is a Kubernetes installer written in Golang. It’s easy to use and doesn’t require a lot of preparation from the user to get started.
You can install RKE from the official GitHub repository. You can run RKE from both Linux and MacOS machines. After installation, run the following code to make sure that you have the latest version:
./rke --version rke version v0.0.6-dev ./rke --help NAME: rke - Rancher Kubernetes Engine, Running kubernetes cluster in the cloud USAGE: rke [global options] command [command options] [arguments...] VERSION: v0.0.6-dev AUTHOR(S): Rancher Labs, Inc. COMMANDS: up Bring the cluster up remove Teardown the cluster and clean cluster nodes version Show cluster Kubernetes version config, config Setup cluster configuration help, h Shows a list of commands or help for one command GLOBAL OPTIONS: --debug, -d Debug logging --help, -h show help --version, -v print the version
RKE is a container-based installer, which means it requires Docker to be installed on the remote server, and it currently requires Docker version 1.12 to be installed on the servers. RKE works by connecting to each server via SSH and setting up a tunnel to the Docker socket on this server, which means that the SSH user must have access to the Docker engine on this server. To enable access to the SSH user, you can add this user to the Docker group:
usermod -aG docker
These are the only preparations the remote servers require to start the Kubernetes installation.
This example assumes that the user provisioned three servers:
node-1: 192.168.1.5 node-2: 192.168.1.6 node-3: 192.168.1.7
The Cluster Config File
By default, RKE looks for a file called cluster.yml, which contains information about the remote servers and services that will run on servers. The minimum file should look like this:
--- nodes: - address: 192.168.1.5 user: ubuntu role: [controlplane] - address: 192.168.1.6 user: ubuntu role: [worker] - address: 192.168.1.7 user: ubuntu role: [etcd] services: etcd: image: quay.io/coreos/etcd:latest kube-api: image: rancher/k8s:v1.8.3-rancher2 kube-controller: image: rancher/k8s:v1.8.3-rancher2 scheduler: image: rancher/k8s:v1.8.3-rancher2 kubelet: image: rancher/k8s:v1.8.3-rancher2 kubeproxy: image: rancher/k8s:v1.8.3-rancher2
The cluster configuration file contains a nodes list. Each node should contain at least these values:
- Address — The SSH IP/FQDN of the server
- User — An SSH user to connect to the server
- Role — A list of the host roles: worker, controlplane, or etcd
The other section is Services, which contains information about the Kubernetes components that will be deployed on the remote servers. There are three types of roles for which a host can be used:
- etcd — These host(s) can be used to hold the data for the cluster.
- controlplane — These hosts(s) can be used to hold the Kubernetes API server and other components that are required to run K8s.
- worker These are the hosts on which your applications can deploy.
To run RKE, make sure to have cluster.yml in the same directory and run:
➜ ./rke up
To point to the other configuration file, run:
➜ ./rke up --config /tmp/config.yml
You should see output like this:
➜ ./rke up --config cluster-aws.yml INFO Building Kubernetes cluster INFO [ssh] Checking private key INFO [ssh] Start tunnel for host [192.168.1.5] INFO [ssh] Start tunnel for host [192.168.1.6] INFO [ssh] Start tunnel for host [192.168.1.7] INFO [certificates] Generating kubernetes certificates INFO [certificates] Generating CA kubernetes certificates INFO [certificates] Generating Kubernetes API server certificates …. INFO [addons] User addon deployed successfully.. INFO Finished building Kubernetes cluster successfully
Connecting to the Cluster
RKE will deploy a local file in the same directory that has the config
file, which contains kube config information to connect to the newly
generated cluster. By default, the kube config file is called
.kube_config_cluster.yml. Copy this file to your local
~/.kube/config to start using
kubectl locally. Note that the
deployed local kube config name is relative to the cluster config file.
For example, if you used a config filed called mycluster.yml, the
local kube config will be named .kube_config_mycluster.yml.
➜ kubectl get nodes NAME STATUS ROLES AGE 192.168.1.5 Ready master 4m v1.8.3-rancher1 192.168.1.6 Ready 4m v1.8.3-rancher1
A Peek Under the Hood
RKE uses the x509 authentication method by default to set authentication between Kubernetes components and for users too. RKE first generates certificates for every component and for the user components as well.
INFO [certificates] Generating kubernetes certificates INFO [certificates] Generating CA kubernetes certificates INFO [certificates] Generating Kubernetes API server certificates INFO [certificates] Generating Kube Controller certificates INFO [certificates] Generating Kube Scheduler certificates INFO [certificates] Generating Kube Proxy certificates INFO [certificates] Generating Node certificate INFO [certificates] Generating admin certificates and kubeconfig INFO [certificates] Deploying kubernetes certificates to Cluster nodes
After generating certificates, RKE deploys the generated certificates to /etc/kubernetes/ssl to the servers and saves the local kube config file, which contains the master user certificate and can be used later with RKE to remove or upgrade the cluster. RKE then deploys each service component as containers that can communicate with each other. RKE also saves the cluster state in Kubernetes as a config map for later use. RKE is an idempotent tool that can run several times, generating the same output. It can also deploy one of the following network plugins:
- Flannel (default)
To use different network plugins, you can specify that in the config file:
network: plugin: calico
RKE supports pluggable add-ons on cluster bootstrap. Users can specify the add-on YAML in the cluster.yml file. RKE deploys the add-ons YAML after the cluster starts. RKE first uploads this YAML file as a config map in the Kubernetes cluster and then runs a Kubernetes job that mounts this config map and deploys the add-ons.
Note that RKE doesn’t yet support the removal of add-ons. Once they are deployed the first time, you can’t change them using RKE.
To start using add-ons, use the
addons: option in the cluster config
file. For example:
addons: |- --- apiVersion: v1 kind: Pod metadata: name: my-nginx namespace: default spec: containers: - name: my-nginx image: nginx ports: - containerPort: 80
Note that we are using
|-</code because addons is a multi-line string
option, in which you can specify multiple YAML files and separate them
The RKE tool is HA ready. You can specify more than one control plane
host in the cluster config file, and RKE will deploy master components
on all of them. By default, the kubelets are configured to connect to
127.0.0.1:6443, which is the address of nginx-proxy service that proxies
requests to all master nodes. To start an HA cluster, specify more than
one host with the role
controlplane, and start the cluster normally.
RKE supports adding/removing nodes for
hosts. To add additional nodes, you need to update only the cluster
config file with additional nodes and run the cluster config with the
same file. To remove nodes, just remove them from the nodes list in the
cluster configuration file, and re-run the
rke up command.
The Cluster Remove Command
RKE supports the
rke remove command. The command does the following:
- Connects to each host and removes the Kubernetes services deployed on it.
- Cleans each host from the directories left by the services: