Skip to content

Latest commit

 

History

History
117 lines (85 loc) · 4.29 KB

USAGE.md

File metadata and controls

117 lines (85 loc) · 4.29 KB
Raw

Using the node-rapids docker images

Quick links

About our docker images

We publish standalone docker images to the rapidsai/node repository to the GitHub container registry on each PR merge.

These images are based on the nvidia/cuda runtime image. They are intended to be used directly, extended with app code, or to serve as examples for building your own deployment images. The sources for each image are in dev/dockerfiles/runtime.

Our tag template scheme is as follows:

ghcr.io/rapidsai/node:{{ RAPIDS_VERSION }}-runtime-node{{ NODE_VERSION }}-cuda{{ CUDA_VERSION }}-ubuntu{{ UBUNTU_VERSION }}-{{ LIBRARY_NAME }}

The latest manifest is available at the GitHub container registry page.

Retrieving the runtime images

The following will retrieve the docker image with each library (+ its native and JS dependencies) installed into the image's /home/node/node_modules:

REPO=ghcr.io/rapidsai/node

VERSIONS="22.12.2-runtime-node16.15.1-cuda11-ubuntu20.04"
docker pull $REPO:$VERSIONS-cudf
docker pull $REPO:$VERSIONS-cuml
docker pull $REPO:$VERSIONS-cugraph
docker pull $REPO:$VERSIONS-cuspatial

VERSIONS="22.12.2-runtime-node16.15.1-cuda11-ubuntu20.04"
docker pull $REPO:$VERSIONS-glfw

# Includes all the above RAPIDS libraries in a single image
docker pull $REPO:$VERSIONS-main

# Includes all the above RAPIDS libraries + demos in a single image
docker pull $REPO:$VERSIONS-demo

Running code in the runtime images

Like the official node images, the default command in the runtime images is node. You can run the following test to validate the GPU is available to your containers:

REPO=ghcr.io/rapidsai/node
VERSIONS="22.12.2-runtime-node16.15.1-cuda11-ubuntu20.04"

# Be sure to pass either the `--runtime=nvidia` or `--gpus` flag!
docker run --rm --gpus=0 $REPO:$VERSIONS-cudf \
    -p "const {Series, DataFrame} = require('@rapidsai/cudf');\
        new DataFrame({ a: Series.new([0, 1, 2]) }).toString()"
>   a
> 0.0
> 1.0
> 2.0

You can mount your host's X11 socket and $DISPLAY envvar, then launch demos that render via OpenGL:

REPO=ghcr.io/rapidsai/node
VERSIONS="22.12.2-runtime-node16.15.1-cuda11-ubuntu20.04"

# Be sure to pass either the `--runtime=nvidia` or `--gpus` flag!
docker run --rm \
    --runtime=nvidia \
    -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
Click here to see a demo of the above command:

Extracting the packaged artifacts

We also publish the container of npm-packed .tgz artifacts installed into each runtime image.

You can use the following technique to install the npm-packed modules into another container or bare-metal cloud instance, provided you also install any runtime dependencies (a gcc-9-compatible libstdc++, matching CUDA Toolkit, etc.) needed by the module.

REPO=ghcr.io/rapidsai/node
VERSIONS="22.12.2-devel-node16.15.1-cuda11-ubuntu20.04"

# Pull the latest image of the packaged .tgz artifacts
docker pull $REPO:$VERSIONS-packages

# Copy the packaged .tgz artifacts, including any of its `@rapidsai/*` dependencies
docker run --rm -v "$PWD:/out" \
    $REPO:$VERSIONS-packages \
    bash -c "cp \
        /opt/rapids/rapidsai-core-*.tgz \
        /opt/rapids/rapidsai-cuda-*.tgz \
        /opt/rapids/rapidsai-rmm-*.tgz  \
        /opt/rapids/rapidsai-cudf-*.tgz \
        /out/"

# Install the npm-packed .tgz (+ npm dependencies) into `./node_modules`
npm install --force --production --legacy-peer-deps \
    --omit dev --omit peer --omit optional \
    *.tgz