bug: zombie process for running service

[oppokui]

trafficstars

Describe the bug

When I run "openllm start xxx" to play with it, I can easily use crtl+C to stop it, no http server left. But once I want to start a backend process by "nohup openllm start llama 2>&1 &", I will lose the control of backend process. It will launch the server as well as lot of http server (per core) at backend. When I kill the main process, those http listener process will be still alive, controlled by some circusd. I am frustrating about it. Everytime I need to reboot OS to restore the env.

How to make the server easy to stop?

To reproduce

No response

Logs

No response

Environment

openllm, 0.3.3 (compiled: no) Python (CPython) 3.10.4

System information (Optional)

No response

[aarnphm]

Ye, so we definitely want to improve on this

you can do

pgrep circusd | xargs kill -9

[aarnphm]

Will translate to bentoml since this is mainly on BentoML improvement

[sauyon]

I would advise against running kill 9 as a rule. How are you terminating the processes normally?