/
Configuring Kyvos Native Compute Cluster from Kyvos Manager

Configuring Kyvos Native Compute Cluster from Kyvos Manager

You can change from an external compute cluster to Kyvos Native for processing semantic models by following a three-step process through Kyvos Manager on the Compute Cluster page:

  1. Configure the Kyvos Native Cluster configurations.

  2. Change the default compute configurations.

Important

  • If using a shared Kubernetes cluster with a dedicated user node pool, ensure that the user node pool has dedicated taints. refer to the Adding Taints and Configure Tolerations in Kyvos worker pods section.

  • In Kyvos 2024.10,

    • For Azure, you can optionally use the K8s cluster on a shared basis. This functionality is not applicable to AWS and GCP.

    • For shared K8s cluster, provide the compute and monitoring namespace that has already been created. For AWS and GCP, namespace names are fixed; hence, fields are disabled.

  • If you select Kyvos compute type as Shared Query Engine, then K8s cluster-related information is not applicable.

  • From Kyvos 2024.9 onwards, you can now increase or decrease the Kubernetes node pool Maximum Compute Server Count. This is applicable only to dedicated K8s cluster.

  • If you have deployed Kyvos with the Kyvos Native compute cluster, you cannot switch to the External Compute cluster to process semantic models.

Prerequisites for the K8s cluster to be used with Kyvos

The following are the prerequisites for the K8s cluster to be used with Kyvos.

  • The Kubernetes cluster must be in the same region in all cloud environments.

  • For GCP, the Kubernetes must be in the same project.

  • For Azure, the Kubernetes cluster is supported even in a different Resource Group.

Note

To fetch the Kubernetes cluster from a different resource group, ensure that Kyvos Managed Identity must have reader on that Kubernetes.

  • For shared K8s cluster, provide the compute and monitoring namespace already created. Currently, only in the case of a dedicated (non-shared) K8s cluster will the system attempt to create a compute and monitoring namespace with the provided name on Azure Enterprise. For AWS and GCP, namespaces names are fixed, hence, those are disabled fields.

  • For AWS and GCP, the K8s cluster must be used in dedicated mode. Hence, the Shared cluster check box is disabled.

Configure Kyvos Native Cluster configurations

To configure the Compute cluster, perform the following steps.

  1. Click the cluster name > Kyvos and Ecosystem > Compute Cluster on the navigation pane to view this information.

  2. The Compute Cluster page displays the two tabs:

    • Kyvos Native: To process the semantic model using Kyvos native, click this tab to select the containerized Kubernetes or Shared Query Engines as a compute cluster.

    • External: Click this tab to view the External compute cluster details.

      image-20241018-110212.png

  3. Enter the details as:

  • Kyvos Native

Changing the default compute configurations

You can Change the Node Pool Instance Type for Kyvos Enterprise (AWS, Azure, and GCP).

  • This functionality is only available for dedicated user node pool.

  • The option to delete the older node pool is available only for dedicated (non-shared) user node pool. If node pool deletion permissions are there, after selecting node pool deletion, it will be deleted automatically to avoid manual deletion of node pool from external to system.

If you are not deleting the older node pool, ensure to remove the taints from older node pool. If you want to configure new taints and toleration, then refer the Adding Taints and Configure Tolerations in Kyvos worker pods

  • The instance type configured for a node pool must be associated with a single instance type, even if multiple instance types are supported for the node pool in any cloud environment.

  • Changing the instance type for a Kubernetes (K8s) cluster node pool should only be supported for the node pool where Kyvos jobs will execute. Therefore, there will be no support for changing the instance type of any other node pool not configured in Kyvos. For Azure, only the user node pool will support instance type changes, while the system node pool will not support such changes.

  • For GCP, the compute.machineTypes.list permission is required to change the Kubernetes node pool instance type.

  • For Azure, following permissions are required to change node pool instance type.

    • "Microsoft.ContainerService/managedClusters/agentPools/write",

    • "Microsoft.ContainerService/managedClusters/agentPools/read",

    • "Microsoft.ContainerService/managedClusters/agentPools/delete"

  • To support K8s cluster node pool instance type, change, add below required permissions to AWS Automated and Wizard based deployment IAM role.

    • Create Node Group

      "eks:CreateNodegroup"
      "eks:TagResource"
      "iam:GetRole"
      "ec2:DescribeKeyPairs"

    • Delete Node Group

      "eks:DeleteNodegroup"

Note

By default, each semantic model is processed using the default compute type. If you want to override the compute type for a particular semantic model, change the value for the kyvos.process.compute.type property from the advanced semantic model property.

To process all of your semantic models using a different compute type, you can change the default compute configurations for all semantic models at once, rather than modifying each one individually. This approach eliminates the need to configure the property for each model separately.

To change the default compute cluster for the Kyvos semantic model, perform the following steps.

  1. Click the cluster name > Kyvos and Ecosystem > Compute Cluster on the navigation pane to view this information.

  2. On the Compute Cluster page, click the Click to change link on the right of the page. The following dialog box is displayed.

    image-20241127-073612.png

  3. Select the required Default Compute Type from the list.

  4. Enter your Kyvos Manager password.

  5. Click Save. The default compute cluster is configured.

 

Copyright Kyvos, Inc. 2026. All rights reserved.