Ingress Controller

Learn more about the Ingress controllers that Seldon Enterprise Platform supports.

An ingress controller functions as a reverse proxy and load balancer, implementing a Kubernetes Ingress. It adds an abstraction layer for traffic routing by receiving traffic from outside the Kubernetes platform and load balancing it to Pods running within the Kubernetes cluster.

Note: Seldon Enterprise Platform does not support NGINX as an alternative for Istio.

The Seldon Enterprise Platform supports Istio ingress controllers. Seldon Core 1 and Seldon Core 2 integrate most effectively with an Istio ingress controller in your Seldon Enterprise Platform ecosystem.

This table lists the features and the ingress controllers for Seldon Core 1 and Seldon Core 2.

Seldon Core 1
Seldon Core 2

The features of Seldon Core 1 does not function without the Istio ingress controller.

Seldon Core 2 is by design service mesh-agnostic and as a result you can choose any ingress controller you want. However, If you choose a non-Istio ingress controller it may restrict you to the more basic or limited functionality of Seldon Core 2.

Seldon Core uses Istio because it relies on a service mesh for traffic splitting and leverages Knative for drift and outlier detection, as well as inference event logging.

Seldon Core 2 offers features such as multi-model serving and overcommit functionality and uses Kafka as a message queue for data-centric pipelines. These features work best with Istio ingress controller.

Features of Istio ingress controller

  • Service mesh with advanced traffic management capabilities.

  • Required for effective use of Seldon Core 1.

  • Automatic configuration of ingress routes for Seldon Core 1 deployments.

  • Enables authorization of Seldon Core 1 deployments.

  • Manages traffic for rollout strategies, drift and outlier detection, and inference data logging for Seldon Core 1.

  • Supports both HTTP/1 and HTTP/2 (gRPC) traffic.

  • Provides TLS support for HTTPS and gRPCS.

  • Supports ingress routes for ML deployments.

Last updated