Artifacts in support of ODH Edge use cases that integration with Red Hat Advanced Cluster Management(Open Cluster Management)
Components | Version |
---|---|
OpenShift | 4.13 |
Open Data Hub | 2.x |
Red Hat Advanced Cluster Management | 2.8 |
OpenShift Pipelines | 1.11.x |
Quay Registry | 2.8 |
The main objective is to showcase that a user can take a trained model, use a pipeline to package it with all the dependencies and deploy it at the near edge location(s) in a centralized way.
- Provision OpenShift Cluster
- Configure the default Identity Provider
- Install Red Hat Advanced Cluster Management
- Register the clusters
ACM Application manifests are located in acm/registration to register and configure the target environments required for the AI at the Edge use cases. The files can be applied to the ACM hub cluster manually:
$ oc apply -k acm/registration
- Core - Cluster host the ODH Core components that will be used in the MLOps Engineer workflow to train, build and push the model. This cluster is not required to be co-located with the ACM Hub but we group them together to simplify the use case
- Near Edge - Cluster(s) that will host the running model at the edge. This is the target environment after a new model is available for use
- Deploy Open Data Hub to the Core cluster and register any configurations to support pushing models to the edge cluster
- GitOps repos
- Image repos
- Manage the edge cluster environments to support deployment of models and support for monitoring
- Configure ACM Observability
- Deploy the Model container
- Develop the model in an ODH Jupyter notebook
- Build the model from the notebook using Data Science Pipelines
- Push the model to the image registry accessible by the near edge cluster(s)
- Update the GitOps config for the near edge cluster
- Core cluster
- Login to the core cluster and run
make install/observability-core
to setup acm-observability on the core cluster.
- Login to the core cluster and run
- Edge cluster(s)
- Login to edge cluster
- Enable userWorkloadMonitoring
oc edit cm cluster-monitoring-config
- Set variable
enableUserWorkload
totrue
- Run
make install/observability-edge
to create the ConfigMap required for metric whitelisting.