Scale the Cluster

Procedure to scale-up or scale-down an existing DKP cluster

Scale up a Cluster

Follow these steps:

  1. Identify the existing PreprovisionedInventory that you wish to scale:

    kubectl get preprovisionedinventory -A
    
  2. Edit the PreprovisionedInventory to add additional IPs needed for additional worker nodes in the spec.hosts section:

    kubectl edit preprovisionedinventory <name> -n <namespace>
    
  3. (Optional) Add any additional host IPs that you require for adding to your cluster:

    spec:
      hosts:
        - address: <worker.ip.add.1>
        - address: <worker.ip.add.2>
    

    After you edit PreprovisionedInventory, fetch the MachineDeployment. The naming convention with md means that it is for worker machines backed by a MachineDeployment. For example:

    kubectl get machinedeployment -A
    
    NAME                     CLUSTER        AGE     PHASE     REPLICAS   READY   UPDATED   UNAVAILABLE
    machinedeployment-md-0   cluster-name   9m10s   Running   4          4       4
    
  4. Scale the worker nodes to the required number. In this example we are scaling from 4 to 6 worker nodes:

    kubectl scale --replicas=6 machinedeployment machinedeployment-md-0 -n default
    
    machinedeployment.cluster.x-k8s.io/machinedeployment-md-0 scaled
    
  5. Monitor the scaling with this command, by adding -w option to watch:

    kubectl get machinedeployment -n default -w
    
    NAME                     CLUSTER        AGE   PHASE       REPLICAS   READY   UPDATED   UNAVAILABLE
    machinedeployment-md-0   cluster-name   20m   ScalingUp   6          4       6         2
    
  6. Check the machine deployment to ensure it has successfully scaled. The output should resemble this example:

    kubectl get machinedeployment -n default
    
    NAME                     CLUSTER             AGE     PHASE     REPLICAS   READY   UPDATED   UNAVAILABLE
    machinedeployment-md-0   machinedeployment   3h33m   Running   6          6       6
    
  7. Alternately, you can use this command and verify the NODENAME column and you should see the additional worker nodes added and in Running state:

    kubectl get machines -A -o wide
    
    NAMESPACE   NAME                                    CLUSTER             AGE     PROVIDERID                          PHASE     VERSION   NODENAME
    default     machinedeployment-control-plane-sljgr   machinedeployment   113m    preprovisioned:////34.123.456.162   Running   v1.21.3   ip-10-0-245-186.us-west-2.compute.internal
    default     machinedeployment-control-plane-wn6pp   machinedeployment   108m    preprovisioned:////54.123.456.63    Running   v1.21.3   ip-10-0-21-113.us-west-2.compute.internal
    default     machinedeployment-control-plane-zpsh6   machinedeployment   119m    preprovisioned:////35.12.345.183    Running   v1.21.3   ip-10-0-43-72.us-west-2.compute.internal
    default     machinedeployment-md-0-d9b7658b-59ndc   machinedeployment   119m    preprovisioned:////18.123.456.224   Running   v1.21.3   ip-10-0-6-233.us-west-2.compute.internal
    default     machinedeployment-md-0-d9b7658b-5tbq9   machinedeployment   119m    preprovisioned:////35.12.345.237    Running   v1.21.3   ip-10-0-19-175.us-west-2.compute.internal
    default     machinedeployment-md-0-d9b7658b-9cgc8   machinedeployment   119m    preprovisioned:////54.123.45.76     Running   v1.21.3   ip-10-0-2-119.us-west-2.compute.internal
    default     machinedeployment-md-0-d9b7658b-9cgc7   machinedeployment   119m    preprovisioned:////55.123.45.76     Running   v1.21.3   ip-10-0-2-118.us-west-2.compute.internal
    default     machinedeployment-md-0-d9b7658b-9cgc6   machinedeployment   5m23s   preprovisioned:////56.123.45.76     Running   v1.21.3   ip-10-0-2-117.us-west-2.compute.internal
    default     machinedeployment-md-0-d9b7658b-9cgc5   machinedeployment   5m23s   preprovisioned:////57.123.45.76     Running   v1.21.3   ip-10-0-2-116.us-west-2.compute.internal
    

Scale Down a Cluster

Run the following command:

kubectl scale machinedeployment <name> -n <namespace> --replicas <new number>

For control plane nodes, currently there is an upstream bug that prevents scaling, but once the issue gets resolved and the fixed CAPI version gets used for DKP, you can perform the following command:

kubectl scale kubeadmcontrolplane <name> -n <namespace> --replicas <new number>

While this item (for control planes) is still being worked on, a workaround would be to perform a kubectl edit kubeadmcontrolplane command and change the replica value under the spec section.

Additional Notes for Scaling Down

It is possible for machines to get stuck in the provisioning stage when scaling down. You can utilize a delete operation to clear the stale machine deployment:

kubectl delete machine <name> -n <namespace>