Deepstream-app not working on DS7 docker image on JP5

• Hardware Platform
Jetson with JetPack 5.1
• DeepStream Version
7.0
• TensorRT Version
as in nvcr.io/nvidia/deepstream:7.0-gc-triton-devel
• Issue Type
bug
• How to reproduce the issue ?

On jetson with JetPack 5.1, running:

docker container run --rm --privileged -it --gpus=all --runtime=nvidia nvcr.io/nvidia/deepstream:7.0-triton-multiarch deepstream-app -c samples/configs/deepstream-app/source2_1080p_dec_infer-resnet_demux_int8.txt

Ends up with:

...
**PERF:  54.87 (2.53)   54.87 (2.53)
**PERF:  0.00 (1.23)    0.00 (1.23)
**PERF:  0.00 (0.81)    0.00 (0.81)
**PERF:  0.00 (0.61)    0.00 (0.61)
**PERF:  0.00 (0.49)    0.00 (0.49)
**PERF:  0.00 (0.40)    0.00 (0.40)
**PERF:  0.00 (0.35)    0.00 (0.35)
**PERF:  0.00 (0.30)    0.00 (0.30)
...

and does not stop. Then, ctr-c and ctr-z can’t even stop the app.

Same command on jetson with Jetpack 6 works fine.

1 Like

Since some libraries are shared between host and docker on jetson, such as cuda/bsp and so on.

This means that the deepstream version of the host and docker must be the same.

So If you want to use deepstream-7.0 docker image, must burn the Jetpack 6.0.

@junshengy

You directly mentioned in your docs, that these libraries are no longer shared with host since 7.0 version.
https://docs.nvidia.com/metropolis/deepstream/dev-guide/text/DS_docker_containers.html#a-docker-container-for-jetson

The DeepStream container no longer expects CUDA, TensorRT to be installed on the Jetson device, because it is included within the container image.

What is the point of docker if container libraries are not separated from host?

But the BSP is still shared, such as codecs.

You can find shared libraries at here. /etc/nvidia-container-runtime/host-files-for-container.d/drivers.csv

Jetson is different from dGPU. dGPU can work by just installing GPU driver. Jetson driver and BSP are integrated in Jetpack

The current role of docker on jetson is only to be deployed on the same version of Jetpack.

There’s nothing I can do about it

I think you should at least mention it in your docs about jetson dockers. Because this information is nowhere to be found.