Deploy a Kubernetes cluster with AKS engine on Azure Stack Hub

You can deploy a Kubernetes cluster on Azure Stack Hub from a client VM running AKS engine. In this article, we look at writing a cluster specification, deploying a cluster with the apimodel.json file, and checking your cluster by deploying MySQL with Helm.

Define a cluster specification

You can specify a cluster specification in a document file using the JSON format called the API model. AKS engine uses a cluster specification in the API model to create your cluster.

You can find examples of the API model for your OS and AKS engine version number for recent releases at AKS engine and corresponding image mapping.

  1. Find your AKS engine version number, for example, v.0.63.0, in the table.
  2. In the API Model samples table, select and open the link for your OS.
  3. Select Raw. You can use the URL in the following instructions.

A URL to the API model may look like:

https://raw.githubusercontent.com/Azure/aks-engine-azurestack/master/examples/azure-stack/kubernetes-azurestack.json

For each of the following samples replace <URL for the API Model> with the URL.

Update the API model

This section looks at creating an API model for your cluster.

  1. Start by using an Azure Stack Hub API Model file for Linux or Windows. From the machine, you installed AKS engine, run:

    curl -o kubernetes-azurestack.json <URL for the API Model>
    

    Note

    If you are disconnected, you can download the file and manually copy it to the disconnected machine where you plan to edit it. You can copy the file to your Linux machine using tools such as PuTTY or WinSCP.

  2. To open the API model in an editor, you can use nano:

    nano ./kubernetes-azurestack.json
    

    Note

    If you don't have nano installed, you can install nano on Ubuntu: sudo apt-get install nano.

  3. In the kubernetes-azurestack.json file, find orchestratorRelease and orchestratorVersion. Select one of the supported Kubernetes versions; you can find the version table in the release notes. Specify the orchestratorRelease as x.xx and orchestratorVersion as x.xx.x. For a list of current versions, see Supported AKS engine Versions

  4. Find customCloudProfile and provide the URL to the tenant portal. For example, https://portal.local.azurestack.external.

  5. Add "identitySystem":"adfs" if you're using AD FS. For example,

        "customCloudProfile": {
            "portalURL": "https://portal.local.azurestack.external",
            "identitySystem": "adfs"
        },
    

    Note

    If you're using Microsoft Entra ID for your identity system, you don't need to add the identitySystem field.

  6. In masterProfile, set the following fields:

    Field Description
    dnsPrefix Enter a unique string that will serve to identify the hostname of VMs. For example, a name based on the resource group name.
    count Enter the number of masters you want for your deployment. The minimum for an HA deployment is 3, but 1 is allowed for non-HA deployments.
    vmSize Enter a size supported by Azure Stack Hub, example Standard_D2_v2.
    distro Enter aks-ubuntu-18.04 or aks-ubuntu-20.04.
  7. In agentPoolProfiles update:

    Field Description
    count Enter the number of agents you want for your deployment. The maximum count of nodes to use per subscription is 50. If you are deploying more than one cluster per subscription ensure that the total agent count doesn't go beyond 50. Make sure to use the configuration items specified in the sample API model JSON file.
    vmSize Enter a size supported by Azure Stack Hub, example Standard_D2_v2.
    distro Enter aks-ubuntu-18.04, aks-ubuntu-20.04 or Windows.
    Use Windows for agents that will run on Windows. For example, see kubernetes-windows.json
  8. In linuxProfile update:

    Field Description
    adminUsername Enter the VM admin user name.
    ssh Enter the public key that will be used for SSH authentication with VMs. Use ssh-rsa and then the key. For instructions on creating a public key, see Create an SSH key for Linux.

    If you're deploying to a custom virtual network, you can find instructions on finding and adding the required key and values to the appropriate arrays in the API Model in Deploy a Kubernetes cluster to a custom virtual network.

    Note

    AKS engine for Azure Stack Hub doesn't allow you to provide your own certificates for the creation of the cluster.

  9. If you're using Windows, in windowsProfile update the values of adminUsername: and adminPassword:

    "windowsProfile": {
    "adminUsername": "azureuser",
    "adminPassword": "",
    "sshEnabled": true
    }
    

More information about the API model

Add certificate when using ASDK

If you are deploying a cluster on the Azure Stack Development Kit (ASDK) and using Linux, you will need to add the root certificate to the trusted certificate store of the client VM running AKS engine.

  1. Find the root certificate in the VM at this directory: /var/lib/waagent/Certificates.pem.
  2. Copy the certificate file:
    sudo cp /var/lib/waagent/Certificates.pem /usr/local/share/ca-certificates/azurestacka.crt
    sudo update-ca-certificates
    

Deploy a Kubernetes cluster

After you have collected all the required values in your API model, you can create your cluster. At this point you should:

Ask your Azure Stack Hub operator to:

  • Verify the health of the system, suggest running Test-AzureStack and your OEM vendor's hardware monitoring tool.
  • Verify the system capacity including resources such as memory, storage, and public IPs.
  • Provide details of the quota associated with your subscription so that you can verify that there is still enough space for the number of VMs you plan to use.

Proceed to deploy a cluster:

  1. Review the available parameters for AKS engine on Azure Stack Hub CLI flags.

    Parameter Example Description
    azure-env AzureStackCloud To indicate to AKS engine that your target platform is Azure Stack Hub use AzureStackCloud.
    identity-system adfs Optional. Specify your identity management solution if you are using Active Directory Federated Services (AD FS).
    location local The region name for your Azure Stack Hub. For the ASDK, the region is set to local.
    resource-group kube-rg Enter the name of a new resource group or select an existing resource group. The resource name needs to be alphanumeric and lowercase.
    api-model ./kubernetes-azurestack.json Path to the cluster configuration file, or API model.
    output-directory kube-rg Enter the name of the directory to contain the output file apimodel.json and other generated files.
    client-id xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx Enter the service principal GUID. The Client ID identified as the Application ID when your Azure Stack Hub administrator created the service principal.
    client-secret xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx Enter the service principal secret. You set up the client secret when creating your service.
    subscription-id xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx Enter your Subscription ID. You must provide a subscription for the tenant. Deployment to the administrative subscription is not supported. For more information, see Subscribe to an offer

    Here is an example:

    Note

    For AKSe version 0.75.3 and above, the command to deploy an AKS engine cluster is aks-engine-azurestack deploy.

    aks-engine deploy \
    --azure-env AzureStackCloud \
    --location <for asdk is local> \
    --resource-group kube-rg \
    --api-model ./kubernetes-azurestack.json \
    --output-directory kube-rg \
    --client-id xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx \
    --client-secret xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx \
    --subscription-id xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx \
    --identity-system adfs # required if using AD FS
    
  2. If for some reason the execution fails after the output directory has been created, you can correct the issue and rerun the command. If you are rerunning the deployment and had used the same output directory before, AKS engine will return an error saying that the directory already exists. You can overwrite the existing directory by using the flag: --force-overwrite.

  3. Save the AKS engine cluster configuration in a secure, encrypted location.

    Locate the file apimodel.json. Save it to a secure location. This file will be used as input in all of your other AKS engine operations.

    The generated apimodel.json file contains the service principal, secret, and SSH public key you use in the input API model. The file also has all the other metadata needed by AKS engine to perform all other operations. If you lose the file, AKS engine won't be able configure the cluster.

    The secrets are unencrypted. Keep the file in an encrypted, secure place.

Verify your cluster

Check your cluster by connecting to kubectl, getting the info, and then getting the states of your nodes.

  1. Get the kubeconfig file to connect to the control plane.

    • If you already have kubectl installed, check the kubeconfig file for the newly created cluster in this directory path /kubeconfig/kubeconfig.json. You can add the /kubeconfig.json to the .kube directory and rename it to "config" to access your new cluster.
      If you have not installed kubectl, visit Install Tools to install the Kubernetes command-line tool. Otherwise, follow the instructions below to access the cluster from one of the control plane nodes.
  2. Get the public IP address of one of your control plane nodes using the Azure Stack Hub portal.

  3. From a machine with access to your Azure Stack Hub instance, connect via SSH to the new control plane node using a client such as PuTTY or MobaXterm.

  4. For the SSH username, use "azureuser" and the private key file of the key pair you provided for the deployment of the cluster.

  5. Check that the cluster endpoints are running:

    kubectl cluster-info
    

    The output should look similar to the following:

    Kubernetes master is running at https://democluster01.location.domain.com
    CoreDNS is running at https://democluster01.location.domain.com/api/v1/namespaces/kube-system/services/kube-dns:dns/proxy
    Metrics-server is running at https://democluster01.location.domain.com/api/v1/namespaces/kube-system/services/https:metrics-server:/proxy
    
  6. Then, review node states:

    kubectl get nodes
    

    The output should be similar to the following:

    k8s-linuxpool-29969128-0   Ready      agent    9d    v1.15.5
    k8s-linuxpool-29969128-1   Ready      agent    9d    v1.15.5
    k8s-linuxpool-29969128-2   Ready      agent    9d    v1.15.5
    k8s-master-29969128-0      Ready      master   9d    v1.15.5
    k8s-master-29969128-1      Ready      master   9d    v1.15.5
    k8s-master-29969128-2      Ready      master   9d    v1.15.5
    

Troubleshoot cluster deployment

When encountering errors while deploying a Kubernetes cluster using AKS engine, you can check:

  1. Are you using the correct Service Principal credentials (SPN)?
  2. Does the SPN have a "Contributors" role to the Azure Stack Hub subscription?
  3. Do you have a large enough quota in your Azure Stack Hub plan?
  4. Is the Azure Stack Hub instance having a patch or upgrade being applied?

For more information, see the Troubleshooting article in the Azure/aks-engine-azurestack GitHub repo.

Rotate your service principle secret

After the deployment of the Kubernetes cluster with AKS engine, the service principal (SPN) is used for managing interactions with the Azure Resource Manager on your Azure Stack Hub instance. At some point, the secret for this the service principal may expire. If your secret expires, you can refresh the credentials by:

  • Updating each node with the new service principal secret.
  • Or updating the API model credentials and running the upgrade.

Update each node manually

  1. Get a new secret for your service principal from your cloud operator. For instructions for Azure Stack Hub, see Use an app identity to access Azure Stack Hub resources.
  2. Use the new credentials provided by your cloud operator to update /etc/kubernetes/azure.json on each node. After making the update, restart both kubele and kube-controller-manager.

Update the cluster with aks-engine update

Alternatively, you can replace the credentials in the apimodel.json and run upgrade using the updated .json file to the same or newer Kubernetes version. For instructions on upgrading the model see Upgrade a Kubernetes cluster on Azure Stack Hub

Next steps