I did a lot of testing, and it seems for concurrency on rasa SDK is a bottle neck. If you test action default vs a basic custom action that just prints hello, the custom action can only handle half the concurrent request as the built in action. Are there any plans on adding sanic or improving the SDK?
Did you try running the action server with a production WSGI runner? E.g. gunicorn, spawning multiple processes to handle the requests?
I thought rasa sdk is run through rasa which is running sanic rasa run actions --actions actions.actions & rasa run --log-file log.out --endpoints endpoints.yml
Should we instead run rasa sdk seperatately as its own function not using rasa run actions?