I encountered the same issue when trying to use the NVIDIA Nsight Tools JupyterLab Extension.
It seems that gitlab-master.nvidia.com is an NVIDIA private Git server, so replacing it with an open alternative might be the best course of action. After some searching, I found that xf86-video-dummy-0.3.8.tar.gz is also available on GitHub:
I patched the following files in nsight-systems to build an xf86-video-dummy container image and successfully set up a livestream of nsys-ui in JupyterLab:
I am not sure why autogen.sh, which is required to generate config.h.in, was not run in the original build-internal.sh, guessing the NVIDIA’s xf86-video-dummy repository already contains a pre-generated config.h.in.
This makes me wonder if there might be some missing parts in the xf86-video-dummy container image in my patched version.
NOTE:
In build.sh, the GIT_URL and GIT_TOKEN environment variables can be used to specify the URL of a Git repository for xf86-video-dummy and the corresponding Git access credential.
@fukui_in_ai_land a new version of jupyterlab-nvidia-nsight (0.7.0) is released.
The new version of the extension uses the latest public Nsight Streamer image (and suggests pulling it if not available locally), so building the image manually is not required anymore when using the latest jupyterlab-nvidia-nsight extension.
For reference, in my setup, Docker is hosted remotely rather than locally, and JupyterLab is launched as a container on the Docker host. In this setup, Docker host address field in the settings of jupyterlab-nvidia-nsight must be specified correctly (corresponding to the ui.dockerHost in the following JSON settings).
I guess this issue is no longer relevant in the latest versions and can be closed, because we don’t need to pull and build the Nsight Streamer code with GITLAB_TOKEN anymore.