Seldon Core

Seldon’s Core Model Deployment Technology

Seldon Core can be installed via a helm chart, kustomize or as part of kubeflow. For full use of Seldon Deploy features it should be installed with istio routing and request logging enabled.

SeldonDeployment custom resources are managed by the seldon core operator, typicall installed in the seldon-system namespace.

When a SeldonDeployment is deployed the operator creates the necessary Kubernetes pods. One of these pods will contain the seldon-container-engine, which reads the inference graph and orchestrates the HTTP calls to follow the sequence of the graph components.

Monitoring is via seldon-core-analytics, which is a grafana install that comes with OOTB dashboards.

Details on the request logging stack can be found in the Request Logging section.

Last modified October 21, 2019: Various updates (6dd5247)