Jetpack 4.2 python-runtime seg fault
rbrigden opened this issue · comments
We start a streaming sender using the python example in stereolabs/zed-examples in the stereolabs/zed:3.1-py-runtime-jetson-jp4.3 and encounter a segfault on our Jetson Nano.
More specifically, we build an image on the Nano with this docker file
FROM stereolabs/zed:3.1-py-runtime-jetson-jp4.3
COPY /main.py /main.py
CMD ["python3", "/main.py"]
where main.py is an exact copy of this
We build and run the image as so
docker build . -t zed-sender
docker run --rm zed-sender
Segmentation fault (core dumped)
I am able to reproduce it. It seems there's a bug loading the Python API in docker on Nano.
We'll investigate and keep you informed
So trying different things, I think it boils down to the docker image launch options, without --gpus all on jetson the sample segfault (and without --privileged no camera will be detected).
Not sure yet why it doesn't output a cleaner error code. For now, you should try with :
docker run --gpus all --privileged --rm zed-sender
Thanks! Forgot to add that flag...
This solved the sender crash issue, but we still were not able to read the stream from the receiver. I re-flashed the device with Jetpack 4.3 and found this command to work perfectly. Using nvidia-docker2 here
sudo docker run -it --rm --runtime nvidia -v streaming_sender.py:/main.py --privileged --network=host stereolabs/zed:3.1-py-runtime-jetson-jp4.3 python3 /main.py
--network=host was the next trick to make it work.
Yes, by default Docker isolate as much as possible, I'll add the clarification in the readme about the network.