GLFW in docker container through ssh

Question

I'm trying to resurrect some old code, which relies on script with GLFW.

Source code for script can be viewed here

The problem is, I need to run it inside the nvidia docker container (nvcr.io/nvidia/tensorrt:21.06-py3), which is deployed on the server without any Display connected to it. The only way to access server, as usual, is ssh connection.

When calling the script I keep having the same exception: EXCEPTION: GLFW initialization failed., which is the result of

 if (glfwInit() != GLFW_TRUE) { 
    throw std::runtime_error("GLFW initialization failed."); 
}

calling glfwGetError(&description) results in

X11: Failed to open display unix:0

The problem is - there is no default value for $DISPLAY on the server.

I've tried to manually set it on host machine via export DISPLAY=:0, and then share host x11 socket with running container:

docker run --gpus all -e DISPLAY=unix$DISPLAY -it --rm --shm-size 1G  \
        -v /tmp/.X11-unix:/tmp/.x11-unix \
        <image>

after connecting to the server with x11 forwarding (ssh -X bla@bla)

But it didn't solve the problem

So my question basically is - how can I make it work? Should I recompile GLFW with some special options, so it will be able to work without set $DISPLAY. Or maybe there is way to invoke mock display in docker container ?

I really need this script to work independently i.e. without any active ssh connections and hence without any x11 forwarding

Answer 1

I added #include <stdexcept> to facewarp.cpp.

Or maybe there is way to invoke mock display in docker container?

Yes there is: use a framebuffer (xvfb) to emulate X11.

🗎 Dockerfile

FROM nvcr.io/nvidia/tensorrt:21.06-py3

ARG DEBIAN_FRONTEND=noninteractive

RUN apt-get update -q && \
    apt-get install -y -qq \
        g++ \
        libglfw3-dev \
        libstb-dev \
        xvfb

COPY facewarp.cpp .
RUN g++ -I/usr/include/stb/ facewarp.cpp -lGL -lglfw -o facewarp

ENV DISPLAY=:99

ENTRYPOINT []
CMD Xvfb :99 -screen 0 1024x768x16 & ./facewarp