How does ray server implement variable sharing with fastapi?

Service.deploymen(args) is implemented in ray server, and args can be obtained in the interface of fastapi, but it should be necessary to share variables between multiple processes. I would like to ask how ray server implements this part? thanks

The class instance variables are local to each process, they are not shared. Every replica start with a fresh copy and they operate on their own. The main code is here: ray/http_util.py at master · ray-project/ray · GitHub