Lead Image © choreograph, 123RF.com

Lead Image © choreograph, 123RF.com

Linking Kubernetes clusters

Growing Pains

Article from ADMIN 68/2022
When Kubernetes needs to scale applications, it searches for free nodes that meet a container's CPU and main memory requirements; however, when the existing hardware is at full capacity, the Kubernetes Cluster Federation project (KubeFed) takes the pain out of adding clusters.

In legacy IT, dynamic scaling usually meant that you had to start new virtual machines manually. Then, you had to integrate the application into an existing cluster and possibly reconfigure the load balancers. Kubernetes takes all of this work off your plate. The

kubectl scale deployment/webserver --replicas=5

command tells Kubernetes to launch five instances of the web server container and makes sure they are running and reachable. Houston, we have load balancing.

The downside is that this only works as long as enough resources are left in the cluster. If the resources are exhausted, you need to add new nodes. Although this might be easy to set up, it means purchasing new hardware and facing a wait, during which the service is not available at the desired performance level. Moreover, if you use autoscaling, Kubernetes can reach its hardware limits without anyone noticing.

One way to bridge the resource gap is to migrate to a public cloud. In the simplest case, you just lease the required resources in the form of virtual machines (VMs) and link them to your cluster. To do this, the existing cluster and the nodes in the cloud need to be able to reach each other directly, either by public addresses or over a VPN connection. Another option is to federate Kubernetes clusters, which gives you the option of docking resources from a second, standalone cluster onto your own and running your applications on both. The software for this is available from the Kubernetes Cluster Federation project (KubeFed) [1].


In the following example, I assume you already operate a Kubernetes (K8s) cluster. The example uses a managed K8s cluster in the AWS cloud as the extension, but apart from the details of accessing it, the installation described here will work with any other Kubernetes cluster.

The KubeFed documentation describes how to install with the use of the Helm Charts packaging format. In the first step, add the KubeFed repository to the locally installed Helm Charts and check the results:

# helm repo add kubefed-charts https://raw.githubusercontent.com/kubernetes-sigs/kubefed/master/charts
# helm repo list
kubefed-charts https://raw.githubusercontent.com/kubernetes-sigs/kubefed/master/charts

If everything worked, Helm now has multiple versions of the chart. The command

# helm search repo kubefed

delivered version 0.9.0 in our lab, which is used as a parameter when importing the chart:

# helm --namespace kube-federation-system upgrade -i kubefed kubefed-charts/kubefed --version=0.9.0 --create-namespace

The next step is to add the first cluster to the federation. To manage the federation, download the kubefedctl [2] tool and copy it to /usr/local/bin/ for ease of use.

kubeconfig Files

The Kubernetes site [3] states that "… a kubeconfig file … is a generic way of referring to configuration files. It does not mean that there is a file named kubeconfig." Access to the Kubernetes cluster(s) is controlled by the default kubeconfig file, ~/.kube/config. This YAML file contains quite a bit of information for each managed cluster, including the cluster itself with a certificate authority (CA) certificate, a name, and an endpoint for access (the cluster entry). Additionally, the user data includes either a certificate and private key, a token for a service account, a script, or a username and password (the user entry).

Last but not least, a context binds one user entry and one cluster entry together. The file also contains the current-context entry, which describes which of the contexts is the default if you do not specify one when calling kubectl. You can easily edit the file with a text editor or use kubectl to do so. The names used in the file are for mutual reference only.

A command like

# kubefedctl join earth --host-cluster-context=earth

adds an initial cluster to the federation. In this example, earth is the name of a cluster context from the ~/.kube/config file. Next, check the status of the cluster:

# kubectl -n kube-federation-system get kubefedclusters
earth  41h    True

While researching this article, I encountered a minor problem in the lab environment. Because the user was named kubeadm and the cluster kubernetes , the cluster context was named kubeadm@kubernetes . However, the at sign interfered with kubefedctl, and the join did not work until I changed the name manually.

Second Cluster

With the major public cloud providers such as Amazon Web Service (AWS) or Google Compute Engine (GCE), you usually have two options for running Kubernetes. Either normal virtual machines are used, on which you install K8s yourself, or the provider has ready-made K8s clusters available as part of a platform-as-a-service offering. Virtual machines also launch in the background, but you see the cluster as the transfer point.

An AWS-managed offering named Elastic Kubernetes Service (EKS) was used for this example. Although creating this cluster basically means pressing a button, you do need to set up some environment parameters beforehand, including:

  • the subnets in the availability zones where the cluster's working nodes will run (at least two nets in two zones),
  • an authorization role for the nodes in AWS – within the AWS infrastructure, this role defines the other resources the Kubernetes nodes can use and how they can use them, and
  • a security group with firewall rules that control the traffic coming into the Kubernetes cluster.

The Kubernetes cluster comprises two components: the control plane (i.e., the cluster) and the node group, to which the containers running in Kubernetes are distributed. An additional link runs between the local data center and the AWS cloud. In principle, the entire cluster could be addressed directly over the Internet, but this option would just add attack vectors.

In a business setting, you would set up a site-to-site VPN connection between your data center and the AWS environment. Accordingly, the Kubernetes cluster and all applications running in it only have private IP addresses and can only be reached within the AWS environment or over the VPN. The security groups for network access also only contain the private IP addresses of the AWS environment and the local data center as sources.

I do not run the Kubernetes cluster permanently in AWS but only enable it when needed. The AWS orchestration tool CloudFormation and Ansible help you create all the components as a stack in AWS, providing the local firewall with a VPN tunnel to the environment and adjusting the routing on the local Kubernetes node so that users can reach the cluster in AWS. Figure 1 shows the infrastructure with this setup. On the AWS side, IP addresses are assigned from the subnet. The Kubernetes cluster itself needs service addresses from a non-local network – in this specific case, The thick black line in the figure represents the VPN tunnel over the Internet.

Figure 1: Architecture of the clusters in AWS and at the data center.

To be able to access the cluster in AWS in a configuration with kubectl, or with the API, you need to add matching entries to the .kube/config file. The AWS command-line interface (CLI) gives you a separate command for this,

aws eks update-kubeconfig <cluster-name>

which adds the required entries. On the management machine, besides the Kubernetes tools, the AWS CLI also needs to be installed and configured to have access to the AWS account.

The entries in the .kube/config file for the EKS cluster look similar. However, instead of logging in with a token, you log in with the output of a command. In the background, kubectl runs the

aws eks get-token --cluster-name vulkan

command, where vulkan is the name of the Kubernetes cluster in AWS in this example. In the Kubernetes config file syntax, the whole user entry then looks like Listing 1.

Listing 1

User Entry

- name: kubernetes-eks-user
      apiVersion: client.authentication.k8s.io/v1alpha1
      command: aws
      - eks
      - get-token
      - --cluster-name
      - vulcan

Because the arguments of the command form a field, you write them as a YAML list one below the other. The cluster context is also named vulcan . If something goes wrong, it is best to scroll to the beginning of the command output, which usually has the most relevant part of the information. The remainder is a stack trace, some of which can be very long-winded. If everything works, a call to kubectl,

# kubectl -n kube-federation-system get kubefedclusters
earth 2h True
vulcan 5m True

confirms that the federation now comprises two clusters.

Buy this article as PDF

Express-Checkout as PDF
Price $2.95
(incl. VAT)

Buy ADMIN Magazine

Get it on Google Play

US / Canada

Get it on Google Play

UK / Australia

Related content

  • A multicluster management tool for Kubernetes
    Multiple Kubernetes clusters with different distributions need a central management tool. One candidate is the open source Open Cluster Manager.
  • Kubernetes clusters within AWS EKS
    Automated deployment of the AWS-managed Kubernetes service EKS helps you run a production Kubernetes cluster in the cloud with ease.
  • Secure access to Kubernetes
    Kubernetes comes with a sophisticated system for ensuring secure access by users and system components through an API. We look at the options for authentication, authorization, and access control.
  • Nested Kubernetes with Loft
    Kubernetes has limited support for multitenancy, so many admins prefer to build multiple standalone Kubernetes clusters that eat up resources and complicate management. As a solution, Loft launches any number of clusters within the same control plane.
  • Rancher Kubernetes management platform
    Rancher has set up shop as an agile alternative to Red Hat OpenShift as an efficient way to manage Kubernetes clusters. In terms of the architecture, a Rancher setup differs significantly from classic Kubernetes.
comments powered by Disqus
Subscribe to our ADMIN Newsletters
Subscribe to our Linux Newsletters
Find Linux and Open Source Jobs

Support Our Work

ADMIN content is made possible with support from readers like you. Please consider contributing when you've found an article to be beneficial.

Learn More”>


		<div class=