Edit

Secure Model Endpoints

Learn how to secure model endpoints in Seldon Core, including authentication, authorization, and TLS configuration.

In enterprise use cases, you may need to control who can access the endpoints for deployed models or pipelines. You can leverage existing authentication mechanisms in your cluster or environment, such as service mesh-level controls, or use cloud provider solutions like Apigee on GCP, Amazon API Gateway on AWS, or a provider-agnostic gateway like Gravitee. Seldon Core 2 integrates with various service meshes that support these requirements. Though Seldon Core 2 is service-mesh agnostic, the example on this page demonstrates how to set up authentication and authorization to secure a model endpoint using the Istio service mesh.

Securing Endpoints with Istio

Service meshes offer a flexible way of defining authentication and authorization rules for your models. With Istio, for example, you can configure multiple layers of security within an Istio Gateway, such as a TLS for HTTPS at the gateway level, mutual TLS (mTLS) for secure internal communication, as well as AuthorizationPolicies and RequestAuthentication policies to enforce both authentication and authorization controls.

Prerequisites

Note There are many types of authorization policies that you can configure to enable access control on workloads in the mesh.

In the following example, you can secure the endpoint such that any requests to the endpoint without the access token are denied.

To secure the endpoints of a model, you need to:

  1. Create a RequestAuthentication resource named ingress-jwt-auth in the istio-system namespace. Replace <OIDC_TOKEN_ISSUER> and <OIDC_TOKEN_ISSUER_JWKS> with your OIDC provider's specific issuer URL and JWKS (JSON Web Key Set) URI.

apiVersion: security.istio.io/v1beta1
kind: RequestAuthentication
metadata:
  name: ingress-jwt-auth
  namespace: istio-system  # This is the namespace where Istio Ingress Gateway usually resides
spec:
  selector:
    matchLabels:
      istio: istio-ingressgateway  # Apply to Istio Ingress Gateway pods
  jwtRules:
    - issuer: <OIDC_TOKEN_ISSUER>
      jwksUri: <OIDC_TOKEN_ISSUER_JWKS>

Create the resource using kubectl apply -f ingress-jwt-auth.yaml.

  1. Create an authorization policy deny-empty-jwt in the namespace istio-system.

Create the resource using kubectl apply -f deny-empty-jwt.yaml.

  1. To verify that the requests without an access token are denied send this request:

The output is similar to:

Now, send the same request with an access token:

The output is similar to:

PreviousTraefikNextConfluent

Last updated

Was this helpful?