This repository contains Payload Processing plugins that will be connected to an AI Gateway via a pluggable BBR (Body Based Routing) framework developed as part of the Kubernetes Inference Gateway.
BBR plugins enable custom request/response mutations of both headers and body, allowing advanced capabilities such as promoting the model from a field in the body to a header and routing to a selected endpoint accordingly.
The target cluster must have ExternalModel CRD deployed.
If you're running this deployment after model-as-a-service, the CRD is already included.
if you're running this repo as a standalone, you need to deploy the CRD before running the helm chart.
-
If ExternalModel CRD is not deployed in your cluster, deploy it using the following:
kubectl apply -f https://raw.githubusercontent.com/opendatahub-io/models-as-a-service/refs/heads/main/deployment/base/maas-controller/crd/bases/maas.opendatahub.io_externalmodels.yaml
-
Set
GATEWAY_NAMEandGATEWAY_NAMESPACEvariables. The chart must be installed in the same namespace as the Gateway for the Istio EnvoyFiltertargetRefsto work:export GATEWAY_NAME=maas-default-gateway export GATEWAY_NAMESPACE=openshift-ingress
-
Install
payload-processinghelm chart:helm install payload-processing ./deploy/payload-processing \ --namespace ${GATEWAY_NAMESPACE} \ --dependency-update \ --set upstreamBbr.inferenceGateway.name=${GATEWAY_NAME} \ --set upstreamBbr.provider.istio.envoyFilter.anchorSubFilter=extensions.istio.io/wasmplugin/${GATEWAY_NAMESPACE}.kuadrant-${GATEWAY_NAME}
Important: The payload processing ext proc is attached to a Gateway. As a mandatory requirement,
--namespacemust match the namespace where the Gateway resource lives.
-
Uninstall
payload-processinghelm chart:helm uninstall payload-processing --namespace ${GATEWAY_NAMESPACE} -
Delete the ExternalModel CRD (optionally):
kubectl delete -f https://raw.githubusercontent.com/opendatahub-io/models-as-a-service/refs/heads/main/deployment/base/maas-controller/crd/bases/maas.opendatahub.io_externalmodels.yaml