Sglang[Diffusion]. Here, The server using a port that is bias from the user specified in command line [or yaml config] argument. that may introduce some chaos.
For example, I had made a server process on local tcp [instead of the default http proto] with launch_server based on a server_args. but because of the settle_port, I do not know which port is the truly port finally used, and I can not write my client code to connect to this port.
and, because of the settle_port will select an unused port in the machine, so , when the server based on a server_args launched, the configed port may be taken and go into used state. And so I can not use another server_args object base on the same config.yaml to write the client program, to connect to the server, because the internel scheduler_port of the client program must have been different.

All in all, Although the settle_port make sure the server can be launched when the desired port is taken [by using another unused port which is not configed], but it makes some chaos in the situation when you need to seperate server process and client processes. OR some other situations when we need to accurately control the scheduler_port finally used.

Maybe there had been some other design in sglang[diffusion] that can solve my situations, then a relative help is welcome. thanks. I know I can reset this property after obeject was created, it seems argly when you had specified the port in your config file and use the file to create a server_args obj, but then, you should set it again.