I have a question on Pyspark MLib Model want to deploy on docker, But the performance is out of expectation..

The docker is deployed with spark standalone, and load the trained pyspark model (randomforest).
But when I use gunicorn to start it up, it is very slow and give prediction.

Could I get your help or any suggestions on spark model deployment? Thanks!