Currently requests simulating production go to the metric server directly, bypassing the prediction server. So we do not have a prediction from the model available in the metric server.
We can solve this by either
going through the prediction server, adding the prediction and forwarding to the metric server (cleaner, but more effort) or
by adding the prediction in the script simulating the request (somewhat hacky, but probably much easier to do)
Currently requests simulating production go to the metric server directly, bypassing the prediction server. So we do not have a prediction from the model available in the metric server.
We can solve this by either