How to Use Virtual Node Addon to Improve Cluster Elasticity

By Xian Wei

It is possible to manually run a yaml file to add virtual nodes to a Kubernetes cluster. However, this is not user friendly and the nodes cannot be continuously upgraded and managed as components. In this blog post, we discuss how virtual nodes work and how you can use Helm to simplify the deployment and management of ack-virtual-node.

What Are Virtual Nodes?

First, let’s briefly review how virtual nodes work.

Virtual nodes come from the community’s virtual kubelet technology. They make the seamless connection between Kubernetes and an Elastic Container Instance (ECI) possible, so that Kubernetes clusters can easily obtain a high level of elasticity without being limited by the computing capacity of cluster nodes.

ECI-based virtual nodes support a variety of functions. They not only enhance the elasticity of Kubernetes clusters, but also provide a wide range of capability extensions, such as GPU container instances, EIP mounting, and large container instances, so that users can easily manage multiple computing workloads in a Kubernetes cluster to meet the needs of a variety of scenarios.

In a hybrid cluster, pods on real nodes are interconnected with ECI pods on virtual nodes.

Note that ECI pods on virtual nodes can be charged on demand, which is different from that on real nodes. For the ECI billing rules, see: Pricing overview. For ECI pod specification configuration support from 0.25c to 64c, see: Limits

Use Cases

Virtual nodes and Serverless Kubernetes are built based on ECI and are Serverless Containers. They are suitable for a variety of Serverless workload scenarios, and can greatly reduce O&M costs, reduce the overall computing costs of users, and improve computing efficiency.

Installing the ack-virtual-node Plug-in

Now let’s look at how you can use Helm to simplify the deployment and management of ack-virtual-node.

To install the ack-virtual-node plug-in, you can follow these steps:

1. Log on to the Container Service console, and create a hosted Kubernetes cluster. Select ack-virtual-node on the Application Directory page.

https://cs.console.aliyun.com/#/k8s/catalog/detail/incubator_ack-virtual-node

2. Configure virtual node parameters, including Region, AK information, vswitchId, and securityGroupId, with the same configurations as the Kubernetes cluster (network configuration information can be viewed on the Cluster Information page).

3. After the Chart is successfully installed, a node, virtual-kubelet, is added to the Node page.

4. With the kubectl command, the node and Helm deployment status can be viewed. Later, ack-virtual-node can also be upgraded and managed through Helm.

Creating a Pod Based on a Virtual Node

When a virtual node exists in the cluster, a pod can be scheduled to the virtual node, and the VK will create a corresponding ECI pod.Currently, a ECI pod can be created by using one of three ways:

1. Set nodeSelector and Tolerations for the Pod

Create the following nginx pod, and set the correct nodeSelector and tolerations to ensure that the pod is scheduled to the virtual node.

apiVersion: v1
kind: Pod
metadata:
  name: nginx
spec:
  containers:
  - image: nginx
    imagePullPolicy: Always
    name: nginx
  nodeSelector:
    type: virtual-kubelet
  tolerations:
  - key: virtual-kubelet.io/provider
    operator: Exists

2. Set nodeName for the Pod

apiVersion: v1
kind: Pod
metadata:
  name: nginx
spec:
  containers:
  - image: nginx
    imagePullPolicy: Always
    name: nginx
  nodeName: virtual-kubelet

3. Configure the Tag for the Namespace to Specify That All Pods in the Namespace are Scheduled to Virtual Nodes.

By adding the tag “virtual-node-affinity-injection=enabled” to the namespace, the admission controller in the system automatically adds nodeAffinity and tolerations to pods in the namespace without requiring users to manually configure tolerations for the pods, which greatly simplifies the use of virtual nodes and does not require the yaml file to adapt to virtual nodes.

kubectl create ns vk
kubectl label namespace vk virtual-node-affinity-injection=enabled
kubectl -n vk run nginx --image nginx

In this way, an ingress application can be created on a virtual node, and the yaml file can be run in the specified namespace without any modification.

# kubectl -n vk apply -f https://raw.githubusercontent.com/AliyunContainerService/serverless-k8s-examples/master/ingress/ingress-cafe-demo.yaml
deployment "coffee" created
service "coffee-svc" created
deployment "tea" created
service "tea-svc" created
ingress "cafe-ingress" created
# kubectl -n vk get pod -o wide
NAME                      READY     STATUS    RESTARTS   AGE       IP              NODE
coffee-56668d6f78-7mdvc   1/1       Running   0          2m        192.168.1.170   virtual-kubelet
coffee-56668d6f78-tpslg   1/1       Running   0          2m        192.168.1.169   virtual-kubelet
tea-85f8bf86fd-5fl2v      1/1       Running   0          2m        192.168.1.172   virtual-kubelet
tea-85f8bf86fd-8n9n8      1/1       Running   0          2m        192.168.1.171   virtual-kubelet
tea-85f8bf86fd-jv7kj      1/1       Running   0          2m        192.168.1.173   virtual-kubelet
# kubectl -n vk get ing
NAME           HOSTS              ADDRESS       PORTS     AGE
cafe-ingress   cafe.example.com   120.55.8.82   80        2m
# curl -H "Host:cafe.example.com" 120.55.8.82/tea 
Server address: 192.168.1.173:80
Server name: vk-tea-85f8bf86fd-jv7kj
Date: 13/May/2019:05:54:32 +0000
URI: /tea
Request ID: 84d2afa2d3a74d7af38f94de21d11d37
# curl -H "Host:cafe.example.com" 120.55.8.82/coffee
Server address: 192.168.1.169:80
Server name: vk-coffee-56668d6f78-tpslg
Date: 13/May/2019:05:54:36 +0000
URI: /coffee
Request ID: 280df5e9f29d22d8174540f8dfe77861

Original Source

How to Use Virtual Node Addon to Improve Cluster Elasticity