MultiProcess problem

[muzhi1991]

When I check code about Chat Streaming, This variable seems to be shared by multiple processes, I also find this variable is assigned by another process. But is_end is local variable. There may be some problems with this code, or is it designed like this?

https://github.com/xlang-ai/OpenAgents/blob/880e26adfe380e999962fc645fc8fc80bd72f103/backend/utils/streaming.py#L266-L273

I am also confused that if this var changes to sharable, logic seems to be broken. Author's design should control this stream stop when chat_thread(this is a python process) not alive. When I use MacOS(m1, version 14.1.2 ) for test, this is_alive may not work, This may another problem.

[github-actions[bot]]

This issue is stale because it has been open 3 days with no activity. Remove stale label or comment or this will be closed in 4 days.

[Timothyxxx]

Hi, I believe we designed so. Could you point out what problems it could lead to?

[muzhi1991]

On my macOS, if tool selected enable, llm chain will not raise exception,(trace it raise exception because this mac python problems)，so i switch to spawn mode at begin of main.py (as code in __name__ == "__main__"). After the modification, the request is normal. However, another problem occurs, subprocess will never end so chat_thread.is_alive() is always true. Finally I found that When I use CODE_EXECUTION_MODE == 'docker' , subprocess will always trigger this start_kernel_publisher thread, and this thread will not stop. If I disable this mode, The child process exits normally.

[isperfee]

in this method multiprocess.Process will copy anther "executor" parameter for runing ,those address or id(executor) != id(executor)

[isperfee]

On mac os set multiprocess.set_start_method("fork", True)