Requirements for Attaching an Existing Cluster

Basic Requirements

To attach an existing cluster in the UI, the Application Management cluster must be able to reach the services and the api-server of the target cluster.

The cluster you want to attach can be a DKP-CLI-created cluster (which will become a Managed cluster upon attachment), or another Kubernetes cluster like AKS, EKS, or GKE (which will become an Attached cluster upon attachment).

DKP does not support attachment of K3s clusters.

For attaching existing clusters without networking restrictions, the requirements depend on which DKP version you are using. Each version of DKP supports a specific range of Kubernetes versions. You must ensure that the target cluster is running a compatible version.

As of this version of DKP, there is no official method of uninstalling Kommander.

Create a Default StorageClass

To deploy many of the services on the attached cluster, there must be a default StorageClass configured. Run the following command on the cluster you want to attach:

CODE

kubectl get sc

The output should look similar to this. Note the (default) after the name:

CODE

NAME               PROVISIONER       RECLAIMPOLICY   VOLUMEBINDINGMODE      ALLOWVOLUMEEXPANSION   AGE
ebs-sc (default)   ebs.csi.aws.com   Delete          WaitForFirstConsumer   false                  41s

If the StorageClass is not set as default, add the following annotation to the StorageClass manifest:

CODE

annotations:
  storageclass.kubernetes.io/is-default-class: "true"

Creating Projects and Workspaces

Before you attach clusters, you need to create one or more Workspaces, and we recommend that you also create Projects within your Workspaces. Workspaces give you a logical way to represent your teams and specific configurations. Projects let you define one or more clusters as a group to which Kommander pushes a common configuration. Grouping your existing clusters in Kommander projects and workspaces makes managing their platform services and resources easier and supports monitoring and logging.

Do not attach a cluster in the "Management Cluster Workspace" workspace. This workspace is reserved for your Application Management cluster only.

Platform Application Requirements

In addition to the basic cluster requirements, the platform services you want DKP to manage on those clusters will have an impact on the total cluster requirements. The specific combinations of platform applications will make a difference in the requirements for the cluster nodes and their resources (CPU, memory, and storage).

See this table of platform applications that DKP provides by default.

Attach Existing AKS, EKS and GKE clusters

Attaching an existing AWS cluster requires that the cluster be fully configured and running. You must create a separate service account when attaching existing AKS, EKS or Google GKE Kubernetes clusters. This is necessary because the kubeconfig files generated from those clusters are not usable out-of-the-box by Kommander. The kubeconfig files call CLI commands, such as azure, aws or gcloud, and use locally-obtained authentication tokens. Having a separate service account also allows you to keep access to the cluster specific and isolated to Kommander.

The suggested default cluster configuration includes a control plane pool containing three (3) m5.xlarge nodes and a worker pool containing four (4) m5.2xlarge nodes.

Consider the additional resource requirements for running the platform services you want DKP to manage, and ensure that your existing clusters comply.

To attach an existing EKS cluster, refer to the specific information in Attach Amazon EKS Cluster.

To attach an existing GKE cluster, refer to the specific information in Attach GKE Cluster.

Attach Clusters with an Existing cert-manager Installation

If you are attaching clusters that already have cert-manager installed, the cert-manager HelmRelease provided by DKP will fail to deploy, due to the existing cert-manager installation. As long as the pre-existing cert-manager functions as expected, you can ignore this failure. It will have no impact on the operation of the cluster.