Docker run command hangs on 6.3

Accelerated Computing Intelligent Video Analytics DeepStream SDK
,
1

• Hardware Platform (Jetson / GPU) GPU
• DeepStream Version 6.3
• JetPack Version (valid for Jetson only) N/A
• TensorRT Version 8.5.03
• NVIDIA GPU Driver Version (valid for GPU only) 535.129.03
• Issue Type( questions, new requirements, bugs) bug
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing) See below
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description) N/A

After upgrading from Deepstream 6.1 → 6.3 and following all steps in the documentation, I am unable to use docker run to run bash commands in a container from the nvcr.io/nvidia/deepstream:6.3-gc-triton-devel docker image. A simple test of:

docker run -t nvcr.io/nvidia/deepstream:6.3-gc-triton-devel ls

enters the container but does not run the command ls and hangs. However, running a container from any other docker image, including

docker run -t nvcr.io/nvidia/deepstream:6.1.1-devel ls

works as expected.

3
    1. I think this is a bug.

Because some images need to initialize Triton and other operations when docker run.
so entrypoint is added in Dockerfile.

As a result, the ls not run as bash command, and because you did not add the -i parameter, docker is not started in interactive mode, and there is no way to receive input.

So the hang occurs

2.Another option, modify this script as below.

/opt/nvidia/nvidia_entrypoint.sh $@

Then build the docker image by yourself.

I have tried, it doesn’t hang

4

Thanks @junshengy. Is there a deepstream:6.3-devel image available with similar entrypoints as 6.1.1-devel, so we don’t have to modify nvidia source code? Are there docs somewhere describing these entrypoint changes between 6.1 → 6.3 or do we need to try and reverse engineer 6.1?

Implementing the change you suggested above raises driver errors that are not present otherwise.

5

You can try to use nvcr.io/nvidia/deepstream:6.3-samples.

What kind of error is reported?

I think there should be no difference between building the image yourself and pulling the image from NGC.

Just add parameter passing

6

building the image myself results in the errors described in this page DeepStream6.3 Error while setting IOCTL

where there is no clear solution.

The driver is the correct version and nvidia-smi works as expected, inside and outside of an interactive docker session.

Is there a reason why docker run commands on the deepstream-6.3 images are not supported as they were in earlier versions?

7

Sorry @junshengy, modifying the entrypoint.sh in the way you mentioned works, indeed it was a driver mismatch that I was confused about. I can now execute docker run on an image that modifies that entrypoint with sed during the build phase, ex:

FROM nvcr.io/nvidia/deepstream:6.3-gc-triton-devel
RUN sed -ie '$d' /opt/nvidia/deepstream/deepstream-6.3/entrypoint.sh
RUN sed -ie '$a /opt/nvidia/nvidia_entrypoint.sh $@' /opt/nvidia/deepstream/deepstream-6.3/entrypoint.sh

thanks!

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.