dhi.io/mlflow-chart
MLflow is an open source platform for managing the end-to-end machine learning lifecycle, including experiment tracking, model registry, and deployment.
All examples in this guide use the public chart and images. If you've mirrored the repository for your own use (for example, to your Docker Hub namespace), update your commands to reference the mirrored chart instead of the public one.
For example:
dhi.io/<repository>:<tag><your-namespace>/dhi-<repository>:<tag>For more details about customizing the chart to reference other images, see the documentation.
To optionally mirror a chart to your own third-party registry, you can follow the instructions in How to mirror an image for either the chart, the image, or both.
The same regctl tool that is used for mirroring container images can also be used for mirroring Helm charts, as Helm
charts are OCI artifacts.
For example:
regctl image copy \
"${SRC_CHART_REPO}:${TAG}" \
"${DEST_REG}/${DEST_CHART_REPO}:${TAG}" \
--referrers \
--referrers-src "${SRC_ATT_REPO}" \
--referrers-tgt "${DEST_REG}/${DEST_CHART_REPO}" \
--force-recursive
The Docker Hardened Images that the chart uses require authentication. To allow your Kubernetes cluster to pull those images, you need to create a Kubernetes secret with your Docker Hub credentials or with the credentials for your own registry.
Follow the authentication instructions for DHI in Kubernetes.
For example:
kubectl create secret docker-registry helm-pull-secret \
--docker-server=dhi.io \
--docker-username=<Docker username> \
--docker-password=<Docker token> \
--docker-email=<Docker email>
To install the chart, use helm install. Make sure you use helm login to log in before running helm install.
Optionally, you can also use the --dry-run flag to test the installation without actually installing anything.
Note:
mlflow.backendStoreUriis a required value. It specifies where MLflow stores experiment and run metadata. For production use, point it to an external database (e.g. PostgreSQL). The example below uses SQLite, which is suitable for simple or single-node deployments.
helm install my-mlflow oci://dhi.io/mlflow-chart --version <version> \
--set "imagePullSecrets[0].name=helm-pull-secret" \
--set "mlflow.backendStoreUri=sqlite:////mlflow/mlflow.db" \
--set "mlflow.defaultArtifactRoot=/mlflow/artifacts" \
--set "storage.enabled=true"
Replace <version> accordingly. If the chart is in your own registry or repository, replace dhi.io with your own
registry and namespace. Replace helm-pull-secret with the name of the image pull secret you created earlier.
The deployment's pods should show up and running almost immediately:
$ kubectl get all
NAME READY STATUS RESTARTS AGE
pod/test-mlflow-chart-565f95ffff-mx9sp 1/1 Running 0 53s
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE
service/kubernetes ClusterIP 10.43.0.1 <none> 443/TCP 107s
service/test-mlflow-chart ClusterIP 10.43.201.117 <none> 5000/TCP 53s
NAME READY UP-TO-DATE AVAILABLE AGE
deployment.apps/test-mlflow-chart 1/1 1 1 53s
NAME DESIRED CURRENT READY AGE
replicaset.apps/test-mlflow-chart-565f95ffff 1 1 1 53s