Request Logging

Log prediction requests and responses for analysis

In Seldon Core request logging is initiated by the seldon container engine. It sends a request-response pair to a knative broker (part of knative eventing) so that it can be processed asynchronously.

The knative broker has a listening trigger which can bring up a request logger. By default the request logger logs to stdout for fluentd to collect and send to elasticsearch, which is the primary backend integration:

requestloggingstack

The request logger component labels requests to distinguish text, tabular and image data. It also splits out batch requests so that they are individually logged. This allows requests to be displayed in the UI and selected for running explanations.

Request logging for KFServices is being tracked in github