Migrate from Kubeflow Distributions¶
Learn about migrating to deployKF from other Kubeflow Distributions like Kubeflow Manifests, Kubeflow on AWS, Kubeflow on GCP and Charmed Kubeflow.
About Migrating¶
Here are some common questions about migrating from other Kubeflow Distributions to deployKF. To learn how to migrate, please see the Steps to Migrate section below.
What is deployKF?¶
deployKF builds machine learning platforms on Kubernetes.
To learn more about deployKF itself, please see the introduction page:
How are Kubeflow and deployKF related?¶
Kubeflow and deployKF are two different but related projects. By using deployKF, you get everything that Kubeflow offers, plus a lot more.
To learn more about the differences, please see the comparison page:
Why migrate to deployKF?¶
We have seen many users struggle with the complexity of deploying Kubeflow. Many users spend days or weeks trying to get Kubeflow working, only to give up in frustration. This is why we created deployKF.
Unlike other distributions, deployKF behaves like a Helm Chart for Kubeflow, with centralized configuration values for all aspects of the platform, so you should never need to edit Kubernetes YAML files or deal with Kustomize patches.
Other Questions¶
Does deployKF support ArgoCD?
Yes, deployKF actually requires ArgoCD.
In the future, we may add support for other GitOps tools or implement our own. Also note, the deployKF ArgoCD Plugin can optionally be used, which makes deployKF behave even more like a Helm Chart.
Is deployKF a Helm Chart for Kubeflow?
No. The very short answer is that Kubeflow is too complex to be deployed as a single Helm Chart, it's closer to an entire cloud platform than a single app.
The slightly longer answer is that Kubeflow is a cluster-wide platform of many different components and dependencies. Helm lacks the sequencing and dependency management features required to deploy Kubeflow in a single chart. deployKF addresses these challenges by being a collection of Helm Charts (and some Kustomize apps) which are configured by a single set of values. You may think of them like "global" Helm values as they control multiple internal apps.
Steps to Migrate¶
Once you have decided to migrate to deployKF, you will need to follow these steps.
1. Create a New Deployment¶
The best way to migrate from Kubeflow Manifests to deployKF is to spin up deployKF in a separate Kubernetes cluster, and then migrate your data manually.
To create a new deployment of deployKF, follow the Getting Started guide.
Warning
Kubeflow Manifests and deployKF can NOT be deployed concurrently in the same Kubernetes cluster, doing so will result in unexpected behavior.
2. Migrate your data¶
Once you have a new deployment of deployKF, you can migrate the data from specific Kubeflow tools to their deployKF equivalents.
For example, you will likely need to migrate your existing Kubeflow Pipelines (scheduled runs, pipeline definitions) and Kubeflow Notebooks (user volume data) to the new deployment.
Created: 2024-05-09