avouacr
commented
3 months ago First remark : it was not a typo but it was not clear either, I reformulated
Second remark : it's because of what I tried to explain in the paragraph just above, but maybe it wasn't clear enough. We could deploy directly the built image on Kubernetes with custom manifests - hence the dotted line - but it's not the best approach for reasons desctibed in the paragraph. Thus, we use a GitOps repo with the manifests, and the operator is in charge of tracking any change happening there. When it detects a change - notably, a change in the tag of the docker image to be deployed - it modifies the manifests on the Kubernetes cluster, and it is Kubernetes that fetches and deploys the last image. There is no direct link between the operator and the container registry.
just a typo > Using a MLflow model registry enables to integrate the pre-processing step before each prediction to the API
Fig 12 is not very clear to me : why is'nt there a line from the container registry to the operator, in order to represent the part "any change in the application manifests is directly propagated to the deployment on the cluster" ? And does the fact that the line from the container registry to the orchestrator is dotted means something, eg. a manual action ? The place of the operator is not obvious to me in this drawing.