TIL: How to pass extra gunicorn arguments to MLflow server

The MLflow server exposes the --workers flag to change the number of gunicorn workers, but if you want to pass other arguments to gunicorn then you need the --gunicorn-opts flag:

mlflow server --workers=8 --gunicorn-opts="--timeout=90"

It looks like --waitress-opts serves the same purpose if running MLflow on Windows.

If passing the --gunicorn-opts flag to a container running the server on Kubernetes, single quote its argument otherwise Kubernetes will not pass the argument to gunicorn correctly:

--gunicorn-opts='--timeout=90'