3.5 KiB

Raw Permalink Blame History

Deploying the Model Registry UI in a local cluster

For this guide, we will be using kind for locally deploying our cluster. See the Model registry server set up guide for prerequisites on setting up kind and deploying the model registry server.

Setup

1. Create a kind cluster

Create a local cluster for running the MR UI using the following command:

kind create cluster

2. Create kubeflow namespace

Create a namespace for model registry to run in, by default this is kubeflow, run:

kubectl create namespace kubeflow

3. Build a standalone image for the UI

Right now, the default image is targeted for the KF Central Dashboard. To build a standalone image for the UI, run:

make docker-build-standalone
make docker-push-standalone

Note: You will need to set up IMG_UI_STANDALONE in your .env.local file to push the image to your own registry.

4. Deploy Model Registry UI to cluster

You can now deploy the UI and BFF to your newly created cluster using the kustomize configs in the root manifest directory:

First you need to set up your new image

export IMG_UI_STANDALONE="<img url that was set above>"
cd manifests/kustomize/options/ui/base
kustomize edit set image model-registry-ui=$IMG_UI_STANDALONE}

Now you can set the namespace to kubeflow and apply the manifests:

cd ../overlays/standalone
kustomize edit set namespace kubeflow
kubectl apply -k .

After a few seconds you should see 1 pod running:

kubectl get pods -n kubeflow

NAME                                  READY   STATUS    RESTARTS   AGE
model-registry-ui-58755c4754-zdrnr    1/1     Running   0          11s

5. Access the Model Registry UI running in the cluster

Now that the pods are up and running you can access the UI.

First you will need to port-forward the UI service by running the following in it's own terminal:

kubectl port-forward service/model-registry-ui-service 8080:8080 -n kubeflow

You can then access the UI running in your cluster locally at http://localhost:8080/

You can now make API requests to the BFF endpoints like:

curl http://localhost:8080/api/v1/model-registry

{
    "model_registry": null
}

Troubleshooting

Running on macOS

When running locally on macOS you may find the pods fail to deploy, with one or more stuck in the pending state. This is usually due to insufficient memory allocated to your docker / podman virtual machine. You can verify this by running:

kubectl describe pods -n kubeflow

If you're experiencing this issue you'll see an output containing something similar to the following:

Events:
  Type     Reason            Age   From               Message
  ----     ------            ----  ----               -------
  Warning  FailedScheduling  29s   default-scheduler  0/1 nodes are available: 1 Insufficient memory. preemption: 0/1 nodes are available: 1 No preemption victims found for incoming pod.

To fix this, you'll need to increase the amount of memory available to the VM. This can be done through either the Podman Desktop or Docker Desktop GUI. 6-8GB of memory is generally a sufficient amount to use.

Running with Kubeflow and Istio

Alternatively, if you'd like to run the UI and BFF pods with an Istio configuration for the KF Central Dashboard, you can apply the manifests by running:

kubectl apply -k manifests/kustomize/options/ui/overlays/istio -n kubeflow

3.5 KiB Raw Permalink Blame History