Customizing Kubeflow on GKE

Version v0.3 of the documentation is no longer actively maintained. The site that you are currently viewing is an archived snapshot. For up-to-date documentation, see the latest version.

This guide describes how to customize your deployment of Kubeflow on Google Kubernetes Engine (GKE).

Before you start

This guide assumes you have already set up Kubeflow with GKE. If you haven’t done so, follow the guide to getting started with Kubeflow on GKE.

Customizing Kubeflow

You can use ksonnet to customize Kubeflow.

The deployment process is divided into two steps, generate and apply, so that you can modify your deployment before actually deploying.

To customize GCP resources (such as your Kubernetes Engine cluster), you can modify the deployment manager configs in ${KFAPP}/gcp_config.

Many changes can be applied to an existing configuration in which case you can run:

cd ${KFAPP}
${KUBEFLOW_REPO}/scripts/ apply platform

or using Deployment Manager directly:

. ${KFAPP}/
cd ${KFAPP}/gcp_config
gcloud deployment-manager --project=${PROJECT} deployments update ${DEPLOYMENT_NAME} --config=cluster-kubeflow.yaml
  • We source to define the environment variables ${PROJECT} and ${DEPLOYMENT_NAME} for this app.

Some changes (such as the VM service account for Kubernetes Engine) can only be set at creation time; in this case you need to tear down your deployment before recreating it:

cd ${KFAPP}
${KUBEFLOW_REPO}/scripts/ delete all
${KUBEFLOW_REPO}/scripts/ apply all

To customize the Kubeflow resources running within the cluster you can modify the ksonnet app in ${KFAPP}/ks_app. For example, to mount additional physical volumes (PVs) in Jupyter:

cd ${KF_APP}/ks_app
ks param set jupyterhub disks "kubeflow-gcfs"

You can then redeploy using kfctl:

cd ${KFAPP}
${KUBEFLOW_REPO}/scripts/ apply k8s

or using ksonnet directly:

cd ${KFAPP}/ks_app
ks apply default

Common customizations

Add GPU nodes to your cluster:

  • Set gpu-pool-initialNodeCount here.

To use VMs with more CPUs or RAM:

  • Change the machineType.
  • There are two node pools:
    • one for CPU only machines here.
    • one for GPU machines here.
  • When making changes to the node pools you also need to bump the pool-version here before you update the deployment.

To grant additional users IAM permissions to access Kubeflow:

  • Add the users here.

After making the changes you need to recreate your deployment:

cd ${KFAPP}
${KUBEFLOW_REPO}/scripts/ apply all

For more information please refer to the Deployment Manager docs.

More customizations

Refer to the navigation panel on the left of these docs for more customizations, including using your own domain, setting up Cloud Filestore, and more.