Description

Add post_predict() to the Predictor API for any actions that should be taken after the request has been responded to.

Questions

• What information should be passed in? Probably the input and return value of predict().

Motivation

Some potential user tasks (e.g. sending metrics, updating in-memory cache, etc) do not have to be on the critical path of responding to the request, so running them after the response is sent will decrease total latency.