Install Kaptain on Konvoy

Install Kaptain on Konvoy


Before proceeding, verify that your environment meets the following basic requirements:

  • Control plane
    • min. 3 nodes
    • min. 4 cores per node
    • min. 200 GiB free disk space per node
    • min. 16 GiB RAM per node
  • Workers
    • min. 6 nodes
    • min. 8 cores per node
    • min. 200 GiB free disk space per node
    • min. 32 GiB RAM per node
  • GPUs (optional)
    • NVIDIA only
    • min. 200 GiB free disk space per instance
    • min. 64 GiB RAM per instance
    • min. 12 GiB GPU RAM per instance

Please note that these numbers are for the bare minimum. Running any realistic machine learning workloads on Kaptain bumps these requirements for nodes, CPUs, RAM, GPUs, and persistent disk. In particular, the number of CPU, GPU workers, and RAM, must be increased considerably. The amounts depend on the number, complexity, and size of the workloads, and the amount of metadata and log data stored with each run.

For on premise installations, horizontal scalability is limited by the overall size of the cluster and quotas therein. For cloud installations, scaling out can be limited by resource quotas.

  • Ensure the following Kubernetes base addons that are needed by Kaptain are enabled:

    - configRepository:
      configVersion: stable-1.20-4.1.0
        - name: istio
          enabled: true
        - name: dex
          enabled: true
        - name: cert-manager
          enabled: true
        - name: prometheus
          enabled: true
  • Add the Kaptain addon repository to your Konvoy cluster.yaml to install Kaptain dependencies, then follow the Konvoy documentation to deploy the addons:

    - configRepository:
      configVersion: stable-1.20-1.3.0
        - name: knative
          enabled: true
  • Install the kubectl-kudo CLI plugin

  • After the Konvoy cluster has been deployed (including Istio and Knative), install KUDO:

    kubectl kudo init --wait
  • Download kubeflow-1.3.0_1.2.0.tgz tarball.

  • Install Kaptain:

    NOTE: Starting with Kaptain 1.2.0, automatic profile creation on initial login is now disabled by default. See User Management for more details.

    kubectl kudo install --instance kaptain --namespace kubeflow --create-namespace ./kubeflow-1.3.0_1.2.0.tgz
  • If you would like to inject additional annotations to Kaptain’s default gateway kubeflow-ingressgateway, you can pass in the service annotations as parameters:

    kubectl kudo install --instance kaptain --namespace kubeflow --create-namespace ./kubeflow-1.3.0_1.2.0.tgz -p kubeflowIngressGatewayServiceAnnotations='{"foo": "abc","bar": "xyz"}'
  • Monitor the installation by running:

    kubectl kudo plan status --instance kaptain  -n kubeflow

Log in to Kaptain

Once all components have been deployed, you can log in to Kaptain:

  • Discover the cluster endpoint and copy it to the clipboard. If you are running Kaptain on-premises:

    kf_uri=$(kubectl get svc kubeflow-ingressgateway --namespace kubeflow -o jsonpath="{.status.loadBalancer.ingress[*].ip}") && echo "https://${kf_uri}"

    Or if you are running Kaptain on AWS:

    kf_uri=$(kubectl get svc kubeflow-ingressgateway --namespace kubeflow -o jsonpath="{.status.loadBalancer.ingress[*].hostname}") && echo "https://${kf_uri}"
  • Get the login credentials from Konvoy to authenticate:

    konvoy get ops-portal

Uninstall Kaptain

  • Use the following commands to uninstall Kaptain.

    kubectl kudo uninstall --instance kaptain --namespace kubeflow --wait
    kubectl delete kubeflow-1.3.0-1.2.0 --namespace kubeflow
    kubectl delete kubeflow --namespace kubeflow