Scaling out on multiple machines

Artikkel
07/11/2024

Now that AKS Edge Essentials is installed on your primary machine, this article describes how you can scale out your cluster to other secondary machines to create a multi-machine deployment.

Caution

Scaling to additional nodes is an experimental feature.

Prerequisites

Set up your scalable Kubernetes cluster.
Set up your secondary machines as described in the Set up machine article. You can't mix different Kubernetes distributions in your cluster. If the cluster on your primary machine is running K8s, you must install the K8s MSI on the secondary machines as well.

Step 1: get cluster configuration from your primary machine

On your primary machine on which you created your scalable deployment, run the following steps in an elevated PowerShell window to create the appropriate configuration file based on your requirements.

To scale by adding a Linux-only worker node, create the required configuration file using the following command and specify the NodeType as "Linux" and provide a unique and available IP address for this node:
```
New-AksEdgeScaleConfig -scaleType AddMachine -NodeType Linux -LinuxNodeIp x.x.x.x -outFile .\ScaleConfig.json | Out-Null
```
To scale by adding more Linux control plane node, specify the NodeType as "Linux", set the ControlPlane flag as true, and provide a unique IP address for the Linux node:
```
New-AksEdgeScaleConfig -scaleType AddMachine -NodeType Linux -LinuxNodeIp x.x.x.x -ControlPlane -outFile .\ScaleConfig.json | Out-Null
```
To scale by adding a Windows-only worker node, specify the NodeType as "Windows" and provide a unique IP address for the Windows node:
```
  New-AksEdgeScaleConfig -scaleType AddMachine -NodeType Windows -WindowsNodeIp x.x.x.x -outFile .\ScaleConfig.json | Out-Null
```

To add a Linux and Windows worker node, specify the NodeType as "LinuxAndWindows" and provide a unique IP address for both the Linux and Windows nodes:

New-AksEdgeScaleConfig -scaleType AddMachine -NodeType LinuxandWindows -LinuxNodeIp x.x.x.x -WindowsNodeIp x.x.x.x -outFile .\ScaleConfig.json | Out-Null

To add a Linux control plane node and Windows worker node, specify the NodeType as "LinuxAndWindows", set the ControlPlane flag as true, and provide a unique IP address for both the Linux and Windows nodes:
```
New-AksEdgeScaleConfig -scaleType AddMachine -NodeType LinuxandWindows -LinuxNodeIp x.x.x.x -WindowsNodeIp x.x.x.x  -ControlPlane -outFile .\ScaleConfig.json | Out-Null
```

These commands export the necessary data to join a cluster in the JSON format, return it as a JSON string, and store it in the file specified via outFile parameter.

Caution

Every time you run the New-AksEdgeScaleConfig command, the previously-created ClusterJoinToken becomes invalid, so you can't use a previously created ScaleConfig.json file. Additionally, ClusterJoinToken is only valid for 24 hours.

Step 2: validate the configuration parameters

The .\ScaleConfig.json configuration file includes the configuration from the primary machine. Review and update necessary sections, and provide details relevant to the machine you're scaling to.

Verify the NetworkConnection.AdapterName with reference to the secondary machine. If you've created an external switch on your Hyper-V on your secondary machine, you can choose to specify the vswitch details in your ScaleConfig.json file. If you don't create an external switch in Hyper-V manager and run the New-AksEdgeDeployment command, AKS Edge Essentials automatically creates an external switch named aksedgesw-ext and uses that for your deployment.

Note

In this release, there is a known issue in automatic creation of external switches with the New-AksEdgeDeployment command if you are using a Wi-fi adapter for the switch. In this case, first create the external switch using the Hyper-V manager - Virtual Switch Manager and map the switch to the Wi-fi adapter. Then provide the switch details in your configuration JSON as described in this section.
The Network.NetworkPlugin is flannel by default. Flannel is the default CNI for a K3S cluster. For a K8S cluster, change the NetworkPlugin to calico.
Verify the resource configuration for the secondary nodes. You can modify these parameters as needed. Ensure that you reserve enough memory for each node. If you specified MacAddress on your primary machine, verify and provide the right MAC address relevant to the secondary machine.
An odd number of control plane nodes is the only supported setting. Therefore, if you want to scale up/down your control plane, make sure you have one, three, or five control plane nodes.

Step 3: bring up a node on your secondary machine

Now you're ready to bring up nodes on your secondary machines and add them to the cluster.

To deploy the corresponding node on the secondary machine, you can now use the ScaleConfig.json file created in the previous step:

New-AksEdgeDeployment -JsonConfigFilePath .\ScaleConfig.json

Step 4: validate your cluster setup

On any node in the cluster, run the following cmdlet:

kubectl get nodes -o wide

You should be able to see all the nodes of the cluster.

Step 5: add more nodes

You can generate a new ScaleConfig file based on the nodeType required by repeating steps 1-4. Ensure that you provide IP addresses that are available in your network each time you add a node.

Step 6: add the second node (Linux/Windows) on a machine that already has a node (optional)

You can add another node to an existing machine that already has a node. For example, if your machine is running a Linux node, you can add a Windows node to it:

New-AksEdgeScaleConfig -ScaleType AddNode -NodeType Windows -WindowsNodeIp "xxx" -outFile .\ScaleConfig.json | Out-Null

You can also specify parameters such as CpuCount and/or MemoryInMB for your Windows VM.

Note

Run New-AksEdgeScaleConfig only on machines that have the Linux node with ControlPlane role.

You can use the generated configuration file and run the following command to add the Windows node:

Add-AksEdgeNode -JsonConfigFilePath .\ScaleConfig.json

You can also specify parameters such as CpuCount and/or MemoryInMB for your Windows VM here.

Del via