• Hardware Platform (Jetson / GPU)
• DeepStream Version
• JetPack Version (valid for Jetson only)
• TensorRT Version
**• Question **
I am facing a weird problem while running the deepstream reference app. The idea is to run the deepstream app controlled by other devices (windows pc) where I will be able to stop and start a rtsp feed (from a panasonic IP/network camera) by pressing buttons from the windows app.
The camera is connected through a PoE hub with a jetson nano and a windows pc. As for inference and other accelerated processes, I am not using any in this stage of development. I just want the camera feed for now.
The setup in jetson:
A python script for UDP connection. Upon receiving “start_camera_1” it will start a process through python subprocess. Something like this–>
subprocess.Popen('/home/nvidia/jet4_deepstream/start_camera_1.sh', shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE, stdin=subprocess.PIPE)
Within the shell script the command for running the deepstream app is–>
deepstream-app -c /home/nvidia/deepstream_test/start_cam_1.txt
Here the text file is the configuration file. And to terminate the process the shell script for subprocess contains the command below →
`killall -s SIGINT deepstream-app`
This setup works perfectly in the beginning, quickly close the camera feed and start again. But the problem rise when it is running for a long time. Upon receiving the kill request it can not terminate the deepstream app. Debugging it I found this error →
nvdc: open: too many files.
So to kill the process I ran this command from another terminal within the jetson →
This time it was able to kill the process and start live feed again but this method raised another issue. When this kill command is applied the feed stops immediately in the jetson, however the port in the PoE hub (connected to the camera) remains busy as the transferring data LED is still blinking. This also shows within the deepstream-app as it says →
error 503: service unavailable.
This also happens if I use other signal like (SIGUSR,SIGKILL, -9 etc). Only the SIGINT command works perfectly within shorter period of time where the reference app is running for less than 2 hours. (maybe even less).
My requirement is to find out a proper termination method of deepstream reference app. I am so sorry that I could not identify whether this problem is rooted in the reference app or the my way of starting/stopping the app. I am attaching the config file and a
ps ax result where the feed is running for more than 3 hours.
[NVIDIA_forum]start_cam_1.txt (3.0 KB)