We recommend a VCK deployment per namespace. The namespace for the deployment of the controller and the namespace being watched by the controller should be the same.
Note: For the purposes of this document, a stock dataset or model are the most frequently used datasets and models in the cluster.
We recommend that the stock datasets and models are pre-populated in each node of the cluster if there are no restrictions on the disk space. Each node can be pre-populated with stocks datasets and models using the S3 source type and setting the replicas equal to the number of nodes in the cluster.
After the pre-population, persistent volume claim (PVC) names and the corresponding dataset or model it represents can be tabulated. The user has to only interact with the PVC name depending upon the dataset or model of their choice. A PVC can be used by any number of objects in Kubernetes.
Recommendations around RBAC, service accounts and namespaces will be added soon.