Scale Your Cluster With KubeKey: Best Practices To Make Life Easier for Kubernetes Cluster Operators

KubeKey is a lightweight, powerful installer for Kubernetes and add-ons, and it is also an efficient tool you can use to scale your Kubernetes cluster.

Sherlock Xu

Jan. 22, 21 · Open Source Zone · Tutorial

Like (3)

Save

2.32K Views

In my last post, I demonstrated how to set up a three-node Kubernetes cluster using KubeKey. As I mentioned in the article, KubeKey is a lightweight, powerful installer that is able to install Kubernetes as well as its related add-ons in a quick and convenient way. In fact, KubeKey can do way more than that as it is also an efficient tool to scale your Kubernetes cluster.

On some cloud platforms, you can directly scale your cluster by increasing or decreasing the number of nodes. Usually, this does not entail complex operations as these platforms will do almost everything for you and you only need to click a few buttons. However, in some on-premises environments, you may need to manually change the number of nodes. In this article, I am going to demonstrate how to scale out and scale in your cluster using KubeKey. The steps are listed as follows:

Download KubeKey
Use KubeKey to retrieve cluster information with a configuration file that's created automatically
Verify cluster information and apply the configuration to scale out the cluster
Scale in the cluster

Prepare Hosts

Here is my node information of the existing Kubernetes cluster:

Host IP	Host Name	Role	System
192.168.0.2	master	master, etcd	CentOS 7.5, 8 Core, 8 G Memory, 50 G Disk
192.168.0.3	worker1	worker	CentOS 7.5, 8 Core, 8 G Memory, 50 G Disk
192.168.0.4	worker2	worker	CentOS 7.5, 8 Core, 8 G Memory, 50 G Disk

    CSS
   
xxxxxxxxxx

$ kubectl get nodes
NAME      STATUS   ROLES    AGE     VERSION
master    Ready    master   9m50s   v1.17.9
worker1   Ready    worker   9m19s   v1.17.9
worker2   Ready    worker   9m20s   v1.17.9

Here is the node that will be added to the cluster first and then removed from the cluster:

Host IP	Host Name	Role	System
192.168.0.5	worker3	worker	CentOS 7.5, 8 Core, 8 G Memory, 50 G Disk

For more information about requirements for nodes, networks, and dependencies, see my last post.

Download KubeKey

Download KubeKey from its GitHub Release Page or use the following command to download KubeKey version 1.0.1. You only need to download KubeKey to one of your machines, that which serves as the taskbox for scaling, such as the master node.

    CSS
   
xxxxxxxxxx

curl -sfL https://get-kk.kubesphere.io | VERSION=v1.0.1 sh -

Note: You can change the version number in the command to download a specific version.

The above command downloads KubeKey and unzips the file. Your folder now contains a file called kk. Make it executable:

    CSS
   
xxxxxxxxxx

chmod +x kk

Scale Out the Cluster

Now that I have KubeKey ready, I need to create a configuration file that contains the information of all nodes, which is the same as setting up a Kubernetes cluster through KubeKey. If your cluster is installed through KubeKey, you may still have that configuration file on your machine. In this case, you can edit it directly. Otherwise, execute the following command to retrieve your cluster information:

    CSS
   
xxxxxxxxxx

./kk create config --from-cluster

The flag --from-cluster is used to get the exiting cluster’s information.

The above command creates a configuration file, which is sample.yaml by default. Open the file and you can see some fields are pre-populated with values. Add the information of the new node and verify if other fields are set correctly.

    CSS
   
xxxxxxxxxx

vi sample.yaml

    CSS
   
xxxxxxxxxx

apiVersion: kubekey.kubesphere.io/v1alpha1
kind: Cluster
metadata:
  name: sample
spec:
  hosts:
  # You should complete the ssh information of the hosts
  - {name: master, address: 192.168.0.2, internalAddress: 192.168.0.2, user: root, password: Testing123}
  - {name: worker1, address: 192.168.0.3, internalAddress: 192.168.0.3, user: root, password: Testing123}
  - {name: worker2, address: 192.168.0.4, internalAddress: 192.168.0.4, user: root, password: Testing123}
  - {name: worker3, address: 192.168.0.5, internalAddress: 192.168.0.5, user: root, password: Testing123}
  roleGroups:
    etcd:
    - master
    master:
    - master
    worker:
    - worker1
    - worker2
    - worker3
  controlPlaneEndpoint:
    # If loadbalancer was used, 'address' should be set to loadbalancer's ip.
    domain: lb.kubesphere.local
    address: ""
    port: 6443
  kubernetes:
    version: v1.17.9
    imageRepo: kubesphere
    clusterName: cluster.local
    proxyMode: ipvs
    masqueradeAll: false
    maxPods: 110
    nodeCidrMaskSize: 24
  network:
    plugin: calico
    kubePodsCIDR: 10.233.64.0/18
    kubeServiceCIDR: 10.233.0.0/18
  registry:
    privateRegistry: ""

Note: You are not allowed to modify the host name of existing nodes (e.g. master) when adding new nodes. For more information about different parameters in the configuration file, see my last article.

Save the file and execute the following command to apply the configuration:

    CSS
   
xxxxxxxxxx

./kk add nodes -f sample.yaml

You can see the output below when scaling finishes.

    CSS
   
xxxxxxxxxx

Congratulations! Scaling cluster is successful.

Execute the following command to check the status of namespaces:

    CSS
   
xxxxxxxxxx

kubectl get pod --all-namespaces

    CSS
   
xxxxxxxxxx

NAMESPACE     NAME                                       READY   STATUS    RESTARTS   AGE
kube-system   calico-kube-controllers-59d85c5c84-48dbg   1/1     Running   0          50m
kube-system   calico-node-8f9tr                          1/1     Running   0          50m
kube-system   calico-node-ttl4c                          1/1     Running   0         3m56s
kube-system   calico-node-wkfld                          1/1     Running   0          50m
kube-system   calico-node-x9ks2                          1/1     Running   0          50m
kube-system   coredns-74d59cc5c6-6jszk                   1/1     Running   0          50m
kube-system   coredns-74d59cc5c6-mzcxp                   1/1     Running   0          50m
kube-system   kube-apiserver-master                      1/1     Running   0          50m
kube-system   kube-controller-manager-master             1/1     Running   0          50m
kube-system   kube-proxy-fw5cf                           1/1     Running   0          50m
kube-system   kube-proxy-qrv5p                           1/1     Running   0          50m
kube-system   kube-proxy-r82sh                           1/1     Running   0         3m56s
kube-system   kube-proxy-v4rnf                           1/1     Running   0          50m
kube-system   kube-scheduler-master                      1/1     Running   0          50m
kube-system   nodelocaldns-4wfsv                         1/1     Running   0         3m56s
kube-system   nodelocaldns-fbsbl                         1/1     Running   0          50m
kube-system   nodelocaldns-mzpvt                         1/1     Running   0          50m
kube-system   nodelocaldns-zj5jz                         1/1     Running   0          50m

Execute the following command to check your nodes:

    CSS
   
xxxxxxxxxx

kubectl get nodes

    CSS
   
xxxxxxxxxx

NAME      STATUS   ROLES    AGE     VERSION
master    Ready    master   52m     v1.17.9
worker1   Ready    worker   51m     v1.17.9
worker2   Ready    worker   51m     v1.17.9
worker3   Ready    worker   5m12s   v1.17.9

As you can see above, all the nodes are up and running.

Scale in the Cluster

To delete a node, you also need to prepare a configuration file in advance. If you jump to this step directly, or you have not downloaded KubeKey, refer to the above content to download KubeKey and retrieve your cluster information. After that, edit the configuration file.

    CSS
   
xxxxxxxxxx

vi sample.yaml

Check that all the fields are set correctly, make changes if necessary, and save the file.

    CSS
   
xxxxxxxxxx

apiVersion: kubekey.kubesphere.io/v1alpha1
kind: Cluster
metadata:
  name: sample
spec:
  hosts:
  # You should complete the ssh information of the hosts
  - {name: master, address: 192.168.0.2, internalAddress: 192.168.0.2, user: root, password: Testing123}
  - {name: worker1, address: 192.168.0.3, internalAddress: 192.168.0.3, user: root, password: Testing123}
  - {name: worker2, address: 192.168.0.4, internalAddress: 192.168.0.4, user: root, password: Testing123}
  - {name: worker3, address: 192.168.0.5, internalAddress: 192.168.0.5, user: root, password: Testing123}
  roleGroups:
    etcd:
    - master
    master:
    - master
    worker:
    - worker1
    - worker2
    - worker3
  controlPlaneEndpoint:
    # If loadbalancer was used, 'address' should be set to loadbalancer's ip.
    domain: lb.kubesphere.local
    address: ""
    port: 6443
  kubernetes:
    version: v1.17.9
    imageRepo: kubesphere
    clusterName: cluster.local
    proxyMode: ipvs
    masqueradeAll: false
    maxPods: 110
    nodeCidrMaskSize: 24
  network:
    plugin: calico
    kubePodsCIDR: 10.233.64.0/18
    kubeServiceCIDR: 10.233.0.0/18
  registry:
    privateRegistry: ""

Execute the following command to scale in your cluster (worker3 is removed in this example):

    CSS
   
xxxxxxxxxx

./kk delete node <nodeName> -f sample.yaml

Note: replace <nodeName> with your own node name.

Execute the following command to verify the result:

    CSS
   
xxxxxxxxxx

kubectl get nodes

    CSS
   
xxxxxxxxxx

NAME      STATUS   ROLES    AGE    VERSION
master    Ready    master   3h7m   v1.17.9
worker1   Ready    worker   3h6m   v1.17.9
worker2   Ready    worker   3h7m   v1.17.9

You can see that the node worker3 has been successfully removed.

Summary

Adding and deleting nodes remains an important part of cluster maintenance. In reality, you need to adjust the number of nodes based on the actual workloads running on your cluster for optimal resource utilization. KubeKey, fortunately, turns out to be one of the most efficient tools in this regard.

Topics:

kubernetes, opensource, kubekey, cluster analysis, cluster autoscaler, kubesphere, kubernetes cluster

Published at DZone with permission of Sherlock Xu. See the original article here.

Opinions expressed by DZone contributors are their own.

Comments