If you don't have a centralized ACM managing your edge cluster(s), you can follow the steps in this document to deploy the inference service container image that was built using the MLOps Pipeline.
Just like in the case with ACM, Argo CD will need to be installed on each target edge cluster that the GitOps-managed inference service container will be deployed on. As a user with the ability to install operators from the OperatorHub on the target edge cluster(s), find and install the OpenShift GitOps operator in the OpenShift Console, in Menu > Operators > OperatorHub.
In your GitOps repo, where the manifests for your application are stored, make sure that the namespace definition has the correct label so that OpenShift GitOps can manage resources in it once it creates it.
To allow the default configuration of OpenShift GitOps to deploy into the namespace, it will need to have the following label:
argocd.argoproj.io/managed-by: openshift-gitops
For more information on this label, see the OpenShift GitOps docs.
Once the operator has been installed from the earlier step, the Application
CRD will be available on the edge
cluster(s), allowing the creation of Application CRs in the GitOps namespace.
On each edge cluster, create an Argo CD Application CR to point the Argo CD instance at the location in the GitOps repo where the manifests are located.
Examples using kustomize
can be found in the byo-mgmt directory.
In the root of this repository, the following can be run to roll out the example inference container applications that exist in this project:
oc apply -k byo-mgmt/registration/near-edge/overlays/tensorflow-housing-app/
oc apply -k byo-mgmt/registration/near-edge/overlays/bike-rental-app/
On each of the edge clusters, you can enable monitoring for user-defined projects:
oc -n openshift-monitoring edit configmap cluster-monitoring-config
- Set variable
enableUserWorkload
totrue
.
If you forward metrics from each edge cluster to a central location, you can find a list of example metrics from the OpenVINO and Seldon model servers from our examples in metrics_list that you may wish to forward.