Error when creating a window in Docker WLS2

[chrisgervang]

I've been following the USAGE.md guide to set up node rapids with windows, and run into an issue with the windowed examples.

Windowed examples in docker WLS2, such as below, throw an exception in glfw.

REPO=ghcr.io/rapidsai/node
VERSIONS="22.12.2-runtime-node16.15.1-cuda11-ubuntu20.04"
docker run --rm -it --gpus all $REPO:$VERSIONS-demo npx @rapidsai/demo-graph

I've also tried variations of this based on this:

docker run --rm -it --gpus all
    -e "DISPLAY=$DISPLAY" \
    -v "/etc/fonts:/etc/fonts:ro" \
    -v "/tmp/.X11-unix:/tmp/.X11-unix:rw" \
    -v "/usr/share/fonts:/usr/share/fonts:ro" \
    -v "/usr/share/icons:/usr/share/icons:ro" \
    $REPO:$VERSIONS-demo \
    npx @rapidsai/demo-graph

I've also tried a couple different gpu flags. --gpus=0, --gpus all, and --runtime=nvidia. The gpus flags throw the same exception shown below.

Stack Trace for `--gpus`

Error creating GLFW window:
Error: api unavailable EGL: Failed to initialize EGL: EGL is not or could not be initialized
    at /opt/rapids/node/modules/glfw/src/window.cpp:62
    at createGLFWWindow (/home/node/node_modules/@rapidsai/jsdom/build/js/polyfills/glfw.js:599:39)
    at /home/node/node_modules/@rapidsai/jsdom/build/js/polyfills/glfw.js:697:23
    at /home/node/node_modules/@rapidsai/jsdom/build/js/index.js:83:42
    at Array.reduce (<anonymous>)
    at Object.beforeParse (/home/node/node_modules/@rapidsai/jsdom/build/js/index.js:83:19)
    at new JSDOM (/home/node/node_modules/jsdom/lib/api.js:40:13)
    at new RapidsJSDOM (/home/node/node_modules/@rapidsai/jsdom/build/js/index.js:59:9)
    at Function.fromReactComponent (/home/node/node_modules/@rapidsai/jsdom/build/js/index.js:40:23)
    at Module.module.exports (/home/node/node_modules/@rapidsai/demo-graph/index.js:39:18)
    at Object.<anonymous> (/home/node/node_modules/@rapidsai/demo-graph/index.js:58:10)
Uncaught Exception
Origin: uncaughtException
Exception: Error: api unavailable EGL: Failed to initialize EGL: EGL is not or could not be initialized
    at /opt/rapids/node/modules/glfw/src/window.cpp:62
    at createGLFWWindow (/home/node/node_modules/@rapidsai/jsdom/build/js/polyfills/glfw.js:599:39)
    at /home/node/node_modules/@rapidsai/jsdom/build/js/polyfills/glfw.js:697:23
    at /home/node/node_modules/@rapidsai/jsdom/build/js/index.js:83:42
    at Array.reduce (<anonymous>)
    at Object.beforeParse (/home/node/node_modules/@rapidsai/jsdom/build/js/index.js:83:19)
    at new JSDOM (/home/node/node_modules/jsdom/lib/api.js:40:13)
    at new RapidsJSDOM (/home/node/node_modules/@rapidsai/jsdom/build/js/index.js:59:9)
    at Function.fromReactComponent (/home/node/node_modules/@rapidsai/jsdom/build/js/index.js:40:23)
    at Module.module.exports (/home/node/node_modules/@rapidsai/demo-graph/index.js:39:18)
    at Object.<anonymous> (/home/node/node_modules/@rapidsai/demo-graph/index.js:58:10)

This exception points to:

node/modules/glfw/src/window.cpp

Lines 62 to 63 in d28a85e

	GLFW_TRY(env,
	window = GLFWAPI::glfwCreateWindow(width, height, title.c_str(), monitor, root_win));

Is there something I should do differently when invoking docker?

Stack Trace for `--runtime=nvidia`

docker: Error response from daemon: unknown or invalid runtime name: nvidia.

Do I need to resolve this missing runtime issue?

---

Non-windowed node-rapids examples do work, such as this:

REPO=ghcr.io/rapidsai/node
VERSIONS="22.12.2-runtime-node16.15.1-cuda11-ubuntu20.04"
docker run --rm -it --gpus all $REPO:$VERSIONS-cudf -p "const {Seris, DataFrame} = require('@rapidsai/cudf'); new DataFrame({ a: Serienew([0,1,2]) }).toString()"