• thockin title: Manually Deploying Kubernetes on Ubuntu Nodes

{% capture overview %} This document describes how to deploy Kubernetes on ubuntu nodes, 1 master and 3 nodes involved in the given examples. You can scale to any number of nodes by changing some settings with ease. The original idea was heavily inspired by @jainvipin 's ubuntu single node work, which has been merge into this document. {% endcapture %}

The scripting referenced here can be used to deploy Kubernetes with networking based either on Flannel or on a CNI plugin that you supply. This document is focused on the Flannel case. See kubernetes/cluster/ubuntu/config-default.sh for remarks on how to use a CNI plugin instead.

Cloud team from Zhejiang University will maintain this work.

{% capture prerequisites %}


  1. The nodes have installed docker version 1.2+ and bridge-utils to manipulate linux bridge.
  2. All machines can communicate with each other. Master node needs to be connected to the Internet to download the necessary files, while worker nodes do not.
  3. These guide is tested OK on Ubuntu 14.04 LTS 64bit server, but it can not work with Ubuntu 15 which uses systemd instead of upstart.
  4. Dependencies of this guide: etcd-2.2.1, flannel-0.5.5, k8s-1.2.0, may work with higher versions.
  5. All the remote servers can be ssh logged in without a password by using key authentication.
  6. The remote user on all machines is using /bin/bash as its login shell, and has sudo access. {% endcapture %}

{% capture steps %}

Starting a Cluster

Set up working directory

Clone the Kubernetes github repo locally

$ git clone --depth 1 https://github.com/kubernetes/kubernetes.git

Configure and start the Kubernetes cluster

The startup process will first download all the required binaries automatically. By default etcd version is 2.2.1, flannel version is 0.5.5 and k8s version is 1.2.0. You can customize your etcd version, flannel version, k8s version by changing corresponding variables ETCD_VERSION , FLANNEL_VERSION and KUBE_VERSION like following.

$ export KUBE_VERSION=1.2.0
$ export FLANNEL_VERSION=0.5.0
$ export ETCD_VERSION=2.2.0


For users who want to bring up a cluster with k8s version v1.1.1, controller manager may fail to start due to a known issue. You could raise it up manually by using following command on the remote master server. Note that you should do this only after api-server is up. Moreover, this issue is fixed in v1.1.2 and later.

$ sudo service kube-controller-manager start

Note that we use flannel here to set up overlay network, yet it's optional. Actually you can build up k8s cluster natively, or use flannel, Open vSwitch or any other SDN tool you like.

An example cluster is listed below:

| IP Address  |   Role   |
||   node   |
||   node   |
|| both master and node|

First configure the cluster information in cluster/ubuntu/config-default.sh, following is a simple sample.

export nodes="vcap@ vcap@ vcap@"

export roles="ai i i"

export NUM_NODES=${NUM_NODES:-3}



The first variable nodes defines all your cluster nodes, master node comes first and separated with blank space like <user_1@ip_1> <user_2@ip_2> <user_3@ip_3>

Then the roles variable defines the roles of above machine in the same order, "ai" stands for machine acts as both master and node, "a" stands for master, "i" stands for node.

The NUM_NODES variable defines the total number of nodes.

The SERVICE_CLUSTER_IP_RANGE variable defines the Kubernetes service IP range. Please make sure that you do have a valid private ip range defined here, because some IaaS provider may reserve private ips. You can use below three private network range according to rfc1918. Besides you'd better not choose the one that conflicts with your own private network range.        -  (10/8 prefix)      -  (172.16/12 prefix)     - (192.168/16 prefix)

The FLANNEL_NET variable defines the IP range used for flannel overlay network, should not conflict with above SERVICE_CLUSTER_IP_RANGE. You can optionally provide additional Flannel network configuration through FLANNEL_BACKEND and FLANNEL_OTHER_NET_CONFIG, as explained in cluster/ubuntu/config-default.sh.

The default setting for ADMISSION_CONTROL is right for the latest release of Kubernetes, but if you choose an earlier release then you might want a different setting. See the admission control doc for the recommended settings for various releases.

Note: When deploying, master needs to be connected to the Internet to download the necessary files. If your machines are located in a private network that need proxy setting to connect the Internet, you can set the config PROXY_SETTING in cluster/ubuntu/config-default.sh such as:

 PROXY_SETTING="http_proxy=http://server:port https_proxy=https://server:port"

After all the above variables being set correctly, we can use following command in cluster/ directory to bring up the whole cluster.

$ KUBERNETES_PROVIDER=ubuntu ./kube-up.sh

The scripts automatically copy binaries and config files to all the machines via scp and start Kubernetes service on them. The only thing you need to do is to type the sudo password when promoted.

Deploying node on machine
[sudo] password to start node: 

If everything works correctly, you will see the following message from console indicating the k8s cluster is up.

Cluster validation succeeded

Test it out

You can use kubectl command to check if the newly created cluster is working correctly. The kubectl binary is under the cluster/ubuntu/binaries directory. You can make it available via PATH, then you can use the below command smoothly.

For example, use $ kubectl get nodes to see if all of your nodes are ready.

$ kubectl get nodes
NAME            STATUS   AGE   VERSION   Ready    3d    v1.6.0+fff5156   Ready    3d    v1.6.0+fff5156   Ready    3d    v1.6.0+fff5156

Also you can run Kubernetes guest-example to build a redis backend cluster.

Deploy addons

Assuming you have a starting cluster now, this section will tell you how to deploy addons like DNS and UI onto the existing cluster.

The configuration of DNS is configured in cluster/ubuntu/config-default.sh.





The DNS_SERVER_IP is defining the ip of dns server which must be in the SERVICE_CLUSTER_IP_RANGE. The DNS_REPLICAS describes how many dns pod running in the cluster.

By default, we also take care of kube-ui addon.


After all the above variables have been set, just type the following command.

$ cd cluster/ubuntu
$ KUBERNETES_PROVIDER=ubuntu ./deployAddons.sh

After some time, you can use $ kubectl get pods --namespace=kube-system to see the DNS and UI pods are running in the cluster.

On going

We are working on these features which we'd like to let everybody know:

  1. Run Kubernetes binaries in Docker using kube-in-docker, to eliminate OS-distro differences.
  2. Tearing Down scripts: clear and re-create the whole stack by one click.


Generally, what this approach does is quite simple:

  1. Download and copy binaries and configuration files to proper directories on every node.
  2. Configure etcd for master node using IPs based on input from user.
  3. Create and start flannel network for worker nodes.

So if you encounter a problem, check etcd configuration of master node first.

  1. Check /var/log/upstart/etcd.log for suspicious etcd log
  2. You may find following commands useful, the former one to bring down the cluster, while the latter one could start it again.
$ KUBERNETES_PROVIDER=ubuntu ./kube-down.sh
$ KUBERNETES_PROVIDER=ubuntu ./kube-up.sh
  1. You can also customize your own settings in /etc/default/{component_name} and restart it via $ sudo service {component_name} restart.

Upgrading a Cluster

If you already have a Kubernetes cluster, and want to upgrade to a new version, you can use following command in cluster/ directory to update the whole cluster or a specified node to a new version.

$ KUBERNETES_PROVIDER=ubuntu ./kube-push.sh [-m|-n <node id>] <version>

It can be done for all components (by default), master(-m) or specified node(-n). Upgrading a single node is currently experimental. If the version is not specified, the script will try to use local binaries. You should ensure all the binaries are well prepared in the expected directory path cluster/ubuntu/binaries.

$ tree cluster/ubuntu/binaries
├── kubectl
├── master
│   ├── etcd
│   ├── etcdctl
│   ├── flanneld
│   ├── kube-apiserver
│   ├── kube-controller-manager
│   └── kube-scheduler
└── minion
    ├── flanneld
    ├── kubelet
    └── kube-proxy

You can use following command to get a help.

$ KUBERNETES_PROVIDER=ubuntu ./kube-push.sh -h

Here are some examples:

  • upgrade master to version 1.0.5: $ KUBERNETES_PROVIDER=ubuntu ./kube-push.sh -m 1.0.5
  • upgrade node vcap@ to version 1.0.5 : $ KUBERNETES_PROVIDER=ubuntu ./kube-push.sh -n 1.0.5
  • upgrade master and all nodes to version 1.0.5: $ KUBERNETES_PROVIDER=ubuntu ./kube-push.sh 1.0.5

The script will not delete any resources of your cluster, it just replaces the binaries.

Test it out

You can use the kubectl command to check if the newly upgraded Kubernetes cluster is working correctly.

To make sure the version of the upgraded cluster is what you expect, you will find these commands helpful.

  • upgrade all components or master: $ kubectl version. Check the Server Version.
  • upgrade node vcap@ $ ssh -t vcap@ 'cd /opt/bin && sudo ./kubelet --version'* {% endcapture %}

Support Level

IaaS Provider Config. Mgmt OS Networking Docs Conforms Support Level
Bare-metal custom Ubuntu flannel docs Community (@resouer, @WIZARD-CXY)

For support level information on all solutions, see the Table of solutions chart.

{% include templates/task.md %}