Feedback aggregator

Feedback aggregator is a service that provides a Model Feedback API and gathers input and output prediction requests

API-provided URLs

Model Feedback API provide just single endpoint that allow you send feedback on a prediction request:

POST /api/v1/feedback

Information about this URL can be viewed using the auto-generated, interactive Swagger page. It is located at <api-address>/swagger/index.html. You can read all of the up-to-date documentation and invoke this endpoint (allowed for your account) right from this web page.

Authentication and authorization

Feedback aggregator distributed in odahu-flow-core helm chart with enabled authorization and pre-defined OPA policies. If Security Subsystem is installed, then all requests to Model Feedback API service will be enforced using pre-defined OPA policies.

Implementation details

Feedback aggregator contains two major subcomponents

  • REST Server provides Model Feedback API and sends them to configured fluentd server
  • Envoy Proxy tap filter catches all requests and responses of deployed models and sends this info to configured fluentd server
Technologies used GoLang, Envoy Proxy
Distribution representation Docker Image
Source code location packages/operator
Can be used w/o Odahu Platform? No
Does it connect to other services? Yes (Fluentd, Envoy Proxy)
Can it be deployed locally? If a local Kubernetes cluster is present
Does it provide any interface? Yes (HTTP REST API)