rasa: http endpoints become unresponsive while nlu model is updating

Rasa version: Rasa 1.1.4

Python version: Python 3.7.2

Operating system: macOS 10.13.5

Issue: I am using rasa as an nlu service only. When configuring automatic model pulling, the /model/parse endpoint becomes unresponsible for a few seconds while the model is training.

From glancing at the code, the model pulling and update should happen in the background (using apscheduler) but it appears to still block standard request handling.

This issue also happens when using the PUT /model endpoint to load a model from e.g. S3 or when configuring Sanic to use multiple workers (app.run(..., workers=4)).

endpoints.yml

models:
  url: http://localhost:5001/model
  wait_time_between_pulls:  10

Command or request that led to error:

rasa run --enable-api -p 5000 --endpoints endpoints.yml --credentials credentials.yml