Installation

This section details how to set up and configure Model Registry on your Kubernetes cluster with Kubeflow.

Prerequisites

These are the minimal requirements to install Model Registry:

• Kubernetes >= 1.27
• Kustomize >= 5.0.3 (see more)

Installing Model Registry

Kubeflow Model registry may be installed as part of a Kubeflow Platform, or as a standalone component. The best option for you will depend on your specific requirements.

Installing on Kubeflow Platform

Kubeflow Model Registry is available as an opt-in alpha component in Kubeflow Platform 1.9+, see Installing Kubeflow to learn more about deploying the Kubeflow Platform.

These instructions assume that you’ve installed Kubeflow from the manifests, if you’re using a distribution consult its documentation instead.

Clone the model-registry repository:

git clone --depth 1 -b v0.2.14 https://github.com/kubeflow/model-registry.git

Switch to the manifests/kustomize for the remaining commands in this section:

cd model-registry/manifests/kustomize

Kubeflow Central Dashboard uses Profiles to handle user namespaces and permissions. By default, the manifests deploy the Model Registry in the kubeflow namespace, to install a compatible version of Model Registry for Kubeflow, you should instead deploy to your profile namespace. Use the following command the modify the manifests for your profile:

PROFILE_NAME=<your-profile>
for DIR in options/istio overlays/db ; do (cd $DIR; kustomize edit set namespace $PROFILE_NAME); done

Now apply the manifests:

kubectl apply -k overlays/db
kubectl apply -k options/istio
kubectl apply -k options/ui/overlays/istio

It may take a few minutes for the pods to be ready, you can check the status of the pods in your profile namespace:

kubectl get pods -n $PROFILE_NAME -w

Finally, configure a Model Registry link in the Kubeflow Dashboard:

kubectl get configmap centraldashboard-config -n kubeflow -o json | jq '.data.links |= (fromjson | .menuLinks += [{"icon": "assignment", "link": "/model-registry/", "text": "Model Registry", "type": "item"}] | tojson)' | kubectl apply -f - -n kubeflow

Standalone installation

It is also possible to install Model Registry without separately from Kubeflow.

By default, the manifests deploy the Model Registry in the kubeflow namespace; you must ensure the kubeflow namespace is available (for example: kubectl create namespace kubeflow) or modify the kustomization file to your desired namespace.

See the list of available versions on the GitHub releases of the kubeflow/model-registry repository. To install a specific release of the Model Registry, modify the following commands with the desired ref=<GIT_TAG>.

Run the following command to install the v0.2.14 release of Model Registry:

MODEL_REGISTRY_VERSION=0.2.14
kubectl apply -k "https://github.com/kubeflow/model-registry/manifests/kustomize/overlays/db?ref=v${MODEL_REGISTRY_VERSION}"

If your Kubernetes cluster uses Istio, you MUST apply the Istio-compatibility manifests (e.g. when using a full Kubeflow Platform). However, these are NOT required for non-Istio clusters.

MODEL_REGISTRY_VERSION=0.2.14
kubectl apply -k "https://github.com/kubeflow/model-registry/manifests/kustomize/options/istio?ref=v${MODEL_REGISTRY_VERSION}"

If you want Kserve to be able to support model-registry:// URI formats, you must apply the cluster-scoped CustomStorageContainer CR.

MODEL_REGISTRY_VERSION=0.2.14
kubectl apply -k "https://github.com/kubeflow/model-registry/manifests/kustomize/options/csi?ref=v${MODEL_REGISTRY_VERSION}"

Check Model Registry setup

You can check the status of the Model Registry deployment with your Kubernetes tooling, or for example with:

kubectl wait --for=condition=available -n kubeflow deployment/model-registry-deployment --timeout=1m
kubectl logs -n kubeflow deployment/model-registry-deployment

Optionally, you can also manually forward the REST API container port of Model Registry and interact with the REST API, for example with:

kubectl port-forward svc/model-registry-service -n kubeflow 8081:8080
# in another terminal:
curl -X 'GET' \
  'http://localhost:8081/api/model_registry/v1alpha3/registered_models?pageSize=100&orderBy=ID&sortOrder=DESC' \
  -H 'accept: application/json' | jq

If you are not receiving a 2xx response, it might be the case you are trying to consume a different version (v1alphaX) of the REST API than intended.

Perform the check from within a Notebook

To check the connection to the Model Registry from a Notebook instead, start a Terminal from the Notebook environment, then you can dry-run the connection with the following command:

curl model-registry-service.kubeflow.svc.cluster.local:8080/api/model_registry/v1alpha3/registered_models

or, alternatively, with the following command:

wget -nv -O- model-registry-service.kubeflow.svc.cluster.local:8080/api/model_registry/v1alpha3/registered_models

If the command executes without any error, you will get a JSON response from Model Registry, indicating the connection and request was successful.

You can use the same commands in a Jupyter Notebook cell by prefixing the command with ! (e.g.: ! curl ...).

Next steps

• Follow the Getting Started guide to learn how to use Model Registry.