ZenML, an extensible, open-source MLOps framework for crafting portable, production-ready machine learning pipelines, is a popular tool in the MLOps domain and would be a great extension to the Charmed Kubeflow ecosystem.
Encouraged by the ease of defining and implementing Kubernetes components through Juju and Charmcraft, and inspired by the Charmed MLFlow Operator project, I decided to create a Charmed ZenML Operator.
This project produces a bundle containing ZenML Server with MySQL backend.
To deploy it onto Kubernetes run:
juju deploy zenml --channel edge
By default, the ZenML Server service is configured as a NodePort exposing port 31375 and has the default username default and no password set.
After deployment, the dashboard can be accessed under http://<node_ip>:31375/ and the zenml client can connected by running:
zenml connect --uri http://<node_ip>:31375/ --username default --password ''
Check out the project Git repository on how to integrate Charmed ZenML Server with Charmed Kubeflow Pipelines and other components:
Feel free to review and contribute!