I found a solution. One way seems to be to use databricks volumes, those volumes can be accessed from the worker. So by reading the volume you can update parameters on the workers.