BLD: Dockerfile for cpu and cuda

.github/workflows/docker-cd.yaml

@@ -17,7 +17,8 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: [ "3.9", "3.10", "3.11" ]
+        python-version: [ "3.10", "3.11", "3.12" ]
+        cuda-version: [ "none", "12.0", "12.5" ]
     steps:
       - name: Check out code
         uses: actions/checkout@v3
@@ -36,7 +37,8 @@ jobs:
         if: ${{ github.repository == 'xorbitsai/xorbits' }}
         env:
           DOCKER_ORG: ${{ secrets.DOCKERHUB_USERNAME }}
-          PY_VERSION: ${{ matrix.python-version }}
+          PYTHON_VERSION: ${{ matrix.python-version }}
+          CUDA_VERSION: ${{ matrix.cuda-version }}
         run: |
           if [[ "$GITHUB_REF" =~ ^"refs/tags/" ]]; then
             export GIT_TAG=$(echo "$GITHUB_REF" | sed -e "s/refs\/tags\///g")
@@ -65,21 +67,28 @@ jobs:
               git checkout $branch
               export IMAGE_TAG="nightly-$branch"
             fi
-            docker build -t "xprobe/xorbits:base-py$PY_VERSION" --progress=plain -f python/xorbits/deploy/docker/Dockerfile.base . --build-arg PYTHON_VERSION=$PY_VERSION
-            docker build -t "$DOCKER_ORG/xorbits:${IMAGE_TAG}-py${PY_VERSION}" --progress=plain -f python/xorbits/deploy/docker/Dockerfile . --build-arg PYTHON_VERSION=$PY_VERSION
-            docker push "xprobe/xorbits:base-py$PY_VERSION"
-            docker push "$DOCKER_ORG/xorbits:${IMAGE_TAG}-py${PY_VERSION}"
+            if [[ "$CUDA_VERSION" == "none" ]]; then
+              # Build CPU image
+              docker build -t "$DOCKER_ORG/xorbits:base-py${PYTHON_VERSION}" --progress=plain -f python/xorbits/deploy/docker/Dockerfile.cpu.base . --build-arg PYTHON_VERSION=$PYTHON_VERSION
+              docker push "$DOCKER_ORG/xorbits:base-py${PYTHON_VERSION}"
+              docker build -t "$DOCKER_ORG/xorbits:${IMAGE_TAG}-py${PYTHON_VERSION}" --progress=plain -f python/xorbits/deploy/docker/Dockerfile.cpu . --build-arg PYTHON_VERSION=$PYTHON_VERSION
+              docker push "$DOCKER_ORG/xorbits:${IMAGE_TAG}-py${PYTHON_VERSION}"
+            else
+              # Build GPU image
+              docker build -t "$DOCKER_ORG/xorbits:${IMAGE_TAG}-cuda${CUDA_VERSION}-py${PYTHON_VERSION}" --progress=plain -f python/xorbits/deploy/docker/Dockerfile.cuda . --build-arg PYTHON_VERSION=$PYTHON_VERSION --build-arg CUDA_VERSION=$CUDA_VERSION
+              docker push "$DOCKER_ORG/xorbits:${IMAGE_TAG}-cuda${CUDA_VERSION}-py${PYTHON_VERSION}"
+            fi
           done
 
       - name: Set default image
         shell: bash
-        if: matrix.python-version == '3.10'
+        if: matrix.python-version == '3.11'
         env:
           DOCKER_ORG: ${{ secrets.DOCKERHUB_USERNAME }}
-          PY_VERSION: ${{ matrix.python-version }}
+          PYTHON_VERSION: ${{ matrix.python-version }}
         run: |
           if [[ "$GITHUB_REF" =~ ^"refs/tags/" ]]; then
             export GIT_TAG=$(echo "$GITHUB_REF" | sed -e "s/refs\/tags\///g")
-            docker tag "$DOCKER_ORG/xorbits:${GIT_TAG}-py${PY_VERSION}" "$DOCKER_ORG/xorbits:${GIT_TAG}"
+            docker tag "$DOCKER_ORG/xorbits:${GIT_TAG}-py${PYTHON_VERSION}" "$DOCKER_ORG/xorbits:${GIT_TAG}"
             docker push "$DOCKER_ORG/xorbits:${GIT_TAG}"
           fi

.github/workflows/python.yaml

@@ -92,16 +92,16 @@ jobs:
           - { os: ubuntu-20.04, module: hadoop, python-version: 3.9 }
           - { os: ubuntu-latest, module: vineyard, python-version: 3.11 }
           - { os: ubuntu-latest, module: external-storage, python-version: 3.11 }
-          # always test compatibility with the latest version
-          # - { os: ubuntu-latest, module: compatibility, python-version: 3.9 }
           - { os: ubuntu-latest, module: doc-build, python-version: 3.9 }
           - { os: self-hosted, module: gpu, python-version: 3.11}
           - { os: ubuntu-latest, module: jax, python-version: 3.9 }
+          - { os: ubuntu-latest, module: datasets, python-version: 3.9 }
+          - { os: ubuntu-latest, module: kubernetes, python-version: 3.11 }
           # a self-hosted runner which needs computing resources, activate when necessary
           # - { os: juicefs-ci, module: kubernetes-juicefs, python-version: 3.9 }
           # - { os: ubuntu-latest, module: slurm, python-version: 3.9 }
-          - { os: ubuntu-latest, module: datasets, python-version: 3.9 }
-          - { os: ubuntu-latest, module: kubernetes, python-version: 3.11 }
+          # always test compatibility with the latest version
+          # - { os: ubuntu-latest, module: compatibility, python-version: 3.9 }
     steps:
     - name: Check out code
       uses: actions/checkout@v3

doc/source/development/contributing_environment.rst

@@ -106,7 +106,7 @@ For the image tag prefixes, ``nightly-main`` represents the latest code from `Xo
 while ``v<release_version>`` represents version numbers for each release.
 You can choose to pull the image based on your specific requirements.
 
-If you indeed need to manually build Xorbits image, Xorbits provides a ``DockerFile`` in the ``python/xorbits/deploy/docker`` directory to build a Docker image
+If you indeed need to manually build Xorbits image, Xorbits provides a ``Dockerfile`` in the ``python/xorbits/deploy/docker`` directory to build a Docker image
 with a full Xorbits development environment.
 
 **Docker Commands**
@@ -117,7 +117,7 @@ Build the Docker image::
     $ cd xorbits
 
     # Build the image
-    docker build -t xorbits-dev --progress=plain -f python/xorbits/deploy/docker/Dockerfile . --build-arg PYTHON_VERSION=<your_python_version>
+    docker build -t xorbits-dev --progress=plain -f python/xorbits/deploy/docker/Dockerfile.cpu . --build-arg PYTHON_VERSION=<your_python_version>
 
 Run Container::
 
@@ -130,7 +130,7 @@ Run Container::
 
 **Visual Studio Code**
 
-You can use the DockerFile to launch a remote session with Visual Studio Code,
+You can use the Dockerfile to launch a remote session with Visual Studio Code,
 a popular free IDE, using the ``.devcontainer.json`` file.
 See https://code.visualstudio.com/docs/remote/containers for details.
 

doc/source/getting_started/installation.rst

@@ -10,19 +10,23 @@ Xorbits can be installed via pip from `PyPI <https://pypi.org/project/xorbits>`_
 
     pip install xorbits
 
-
-.. _install.version:
+It will install the latest version of Xorbits and dependencies like ``pandas``, ``numpy``, etc.
+We recommend you to use environment management tools like ``conda`` or ``venv`` to create 
+a new environment. ``conda`` will install the pre-compiled packages, while ``pip`` will
+install the wheel (which is pre-compiled) or compile the packages from source code if no wheel
+is available.
 
 Python version support
 ----------------------
 
-Officially Python 3.9, 3.10, 3.11, and 3.12.
+Officially support Python 3.9, 3.10, 3.11, and 3.12.
 
 Packages support
 ----------------
 
 Xorbits partitions large datasets into chunks and processes each individual 
-chunk using single-node packages (such as pandas). Currently, our latest version strives 
+chunk using single-node packages (such as pandas). 
+Currently, our latest version strives 
 to be compatible with the latest single-node packages. The table below lists the highest 
 versions of the single-node packages that Xorbits are compatible with. If you are using 
 an older version of pandas, you should either upgrade your pandas or downgrade Xorbits.
@@ -40,6 +44,37 @@ Xorbits Python              `NumPy`_ `pandas`_ `xgboost`_ `lightgbm`_ `datasets`
 .. _`lightgbm`: https://lightgbm.readthedocs.io
 .. _`datasets`: https://huggingface.co/docs/datasets/index
 
+GPU support
+-----------
+
+Xorbits can also scale GPU-accelerated data science tools like `CuPy`_ and `cuDF`_. To enable GPU support, you need to install
+GPU-accelerated packages. As GPU software stacks (i.e.,GPU driver, CUDA, etc.)
+are complicated from CPU, you need to make sure NVIDIA driver and CUDA toolkit are properly installed.
+We recommend you to use ``conda`` to install ``cuDF`` first, it will install both ``cudf`` and ``cupy``,
+and then install ``xorbits`` with ``pip``. 
+``conda`` will help resolve the dependencies of ``cuDF`` and provides supporting software like CUDA.
+Refer to `RAPIDS_INSTALL_DOCS`_ for more details about how to install ``cuDF``.
+
+When using Xorbits with GPU, you need to add the :code:`gpu=True` parameter to the data loading method.
+For example:
+
+.. code-block:: python
+
+    import xorbits.pandas as pd
+    df = pd.read_parquet(path, gpu=True)
+
+======= =================== ======== =========
+Xorbits Python              `CuPy`_  `cuDF`_  
+======= =================== ======== =========
+0.8.1   3.10,3.11,3.12      13.3.0    24.10   
+======= =================== ======== =========
+
+If you find installing GPU-accelerated packages too complicated, you can use our docker images
+with pre-installed GPU drivers and CUDA toolkit. Please refer to :ref:`docker` for more details.
+
+.. _`Cupy`: https://cupy.dev
+.. _`cuDF`: https://docs.rapids.ai/api/cudf/stable/
+.. _`RAPIDS_INSTALL_DOCS`: https://docs.rapids.ai/install/
 
 Dependencies
 ------------
@@ -89,3 +124,13 @@ The following extra dependencies will be installed.
   library.
 
 * `fsspec <https://github.com/fsspec/filesystem_spec>`__: for cloud data accessing.
+
+.. _docker:
+Docker image
+------------
+
+To simplify the installation of Xorbits, we provide docker images with pre-installed
+Xorbits and its dependencies.
+
+* CPU image: ``xprobe/xorbits:v{version}-py{python_version}``, e.g., ``xprobe/xorbits:v0.8.0-py3.12``
+* GPU image: ``xprobe/xorbits:v{version}-cuda{cuda_version}-py{python_version}``, e.g., ``xprobe/xorbits:v0.8.0-cuda12.0-py3.12``

doc/source/locale/zh_CN/LC_MESSAGES/development/contributing_environment.po

@@ -223,7 +223,7 @@ msgstr ""
 #: ../../source/development/contributing_environment.rst:112
 msgid ""
 "If you indeed need to manually build Xorbits image, Xorbits provides a "
-"``DockerFile`` in the ``python/xorbits/deploy/docker`` directory to build"
+"``Dockerfile`` in the ``python/xorbits/deploy/docker`` directory to build"
 " a Docker image with a full Xorbits development environment."
 msgstr ""
 "如果你真的需要手动构建镜像，Xorbits 在 ``python/xorbits/deploy/docker`` 目录下提供了一个 Dockerfile 用以构建包含完备开发环境的镜像。"
@@ -254,7 +254,7 @@ msgstr ""
 
 #: ../../source/development/contributing_environment.rst:136
 msgid ""
-"You can use the DockerFile to launch a remote session with Visual Studio "
+"You can use the Dockerfile to launch a remote session with Visual Studio "
 "Code, a popular free IDE, using the ``.devcontainer.json`` file. See "
 "https://code.visualstudio.com/docs/remote/containers for details."
 msgstr ""

doc/source/locale/zh_CN/LC_MESSAGES/user_guide/deployment_kubernetes.po

@@ -161,14 +161,14 @@ msgstr "自 ``v0.2.1`` 起，Xorbits 镜像不再支持 Python ``3.7``，同时
 #: ../../source/user_guide/deployment_kubernetes.rst:82
 msgid ""
 "If you need to build an image from source, the related Dockerfiles exists"
-" at `this position <https://github.com/xprobe-"
-"inc/xorbits/tree/main/python/xorbits/deploy/docker>`_ for reference. You "
+" at `this position <https://github.com/xorbitsai"
+"/xorbits/tree/main/python/xorbits/deploy/docker>`_ for reference. You "
 "can follow the `Docker document "
 "<https://docs.docker.com/engine/reference/commandline/build/>`_ to build "
 "your own Xorbits image."
 msgstr ""
-"如果你希望从源码制作一个镜像，可以参考我们的 `Dockerfile <https://github.com/xprobe-"
-"inc/xorbits/tree/main/python/xorbits/deploy/docker>`_ 和 `Docker 构建文档 "
+"如果你希望从源码制作一个镜像，可以参考我们的 `Dockerfile <https://github.com/xorbitsai"
+"/xorbits/tree/main/python/xorbits/deploy/docker>`_ 和 `Docker 构建文档 "
 "<https://docs.docker.com/engine/reference/commandline/build/>`_ 进行制作。"
 
 #: ../../source/user_guide/deployment_kubernetes.rst:85
@@ -192,15 +192,15 @@ msgstr "安装 Python 包"
 
 #: ../../source/user_guide/deployment_kubernetes.rst:94
 msgid ""
-"Refer `DockerFile <https://github.com/xprobe-"
-"inc/xorbits/blob/main/python/xorbits/deploy/docker/Dockerfile.base>`_ for"
+"Refer `Dockerfile <https://github.com/xorbitsai"
+"/xorbits/blob/main/python/xorbits/deploy/docker/Dockerfile.cpu>`_ for"
 " the python packages included in the Xorbits image. If you want to "
 "install additional python packages in your Xorbits K8s cluster, use "
 "``pip`` and ``conda`` options of the "
 ":meth:`xorbits.deploy.kubernetes.client.new_cluster` api."
 msgstr ""
-"Xorbits 的发布镜像中已经包含了一些 Python 包，参考 `DockerFile <https://github.com/xprobe-"
-"inc/xorbits/blob/main/python/xorbits/deploy/docker/Dockerfile.base>`_ "
+"Xorbits 的发布镜像中已经包含了一些 Python 包，参考 `Dockerfile <https://github.com/xorbitsai"
+"/xorbits/blob/main/python/xorbits/deploy/docker/Dockerfile.cpu>`_ "
 "中安装内容。如果你想安装额外的 Python 包或者改变其中某些包的版本，使用 "
 ":meth:`xorbits.deploy.kubernetes.client.new_cluster` 接口中的 ``pip`` 和 "
 "``conda`` 选项即可。"

doc/source/user_guide/deployment_kubernetes.rst

@@ -91,7 +91,7 @@ Finally, specify your own image during the deployment process through the ``imag
 
 Install Python Packages
 -----------------------
-Refer `DockerFile <https://github.com/xorbitsai/xorbits/blob/main/python/xorbits/deploy/docker/Dockerfile.base>`_ for the python packages included in the Xorbits image.
+Refer `Dockerfile <https://github.com/xorbitsai/xorbits/blob/main/python/xorbits/deploy/docker/Dockerfile.cpu>`_ for the python packages included in the Xorbits image.
 If you want to install additional python packages in your Xorbits K8s cluster, use ``pip`` and ``conda`` options of the :meth:`xorbits.deploy.kubernetes.client.new_cluster` api.
 
 Please make sure your K8s cluster can access the corresponding `channel of conda <https://docs.conda.io/projects/conda/en/latest/user-guide/concepts/channels.html>`_ or `PyPi <https://pypi.org/>`_, when using ``pip`` and ``conda`` options.

doc/source/user_guide/storage_backend.rst

@@ -99,7 +99,15 @@ If you want to run tasks on GPUs, add the :code:`gpu=True` parameter to the data
 .. code-block:: python
 
     import xorbits.pandas as pd
-    pd.read_parquet(path, gpu=True)
+    import xorbits.numpy as np
+
+    df = pd.read_parquet(path, gpu=True)
+    ...
+
+    a = np.ones((1000, 1000), gpu=True)
+    b = np.ones((1000, 1000), gpu=True)
+    c = np.matmul(a, b)
+    ...
 
 
 All subsequent operations will run on GPUs.

python/xorbits/deploy/docker/Dockerfile.cpu.base

@@ -0,0 +1,46 @@
+FROM continuumio/miniconda3:24.9.2-0 AS base
+
+ARG PYTHON_VERSION=3.9
+SHELL ["/bin/bash", "-c"]
+
+RUN apt-get -y update \
+    && apt-get install -y \
+      curl \
+      procps \
+      gcc \
+      g++ \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN conda install python=${PYTHON_VERSION} \
+    nodejs=20.17 \
+    conda-forge::mkl \
+    conda-forge::libnuma \
+    && conda clean --all -f -y
+# TODO: UCXX is not mature enough for production, add it back when it's ready
+
+RUN pip install -U \
+    numpy \
+    scipy \
+    pandas \
+    numexpr \
+    psutil \
+    scikit-learn \
+    sqlalchemy \
+    tornado \
+    xoscar \
+    pyarrow \
+    cloudpickle \
+    azure-storage-blob \
+    adlfs \
+    fsspec \
+    s3fs \
+    pyopenssl \
+    datasets \
+    python-kubernetes \
+    jax \
+    uvloop \
+    Cython
+
+RUN if [ "$PYTHON_VERSION" == "3.9" ] ; then \
+    pip install -U 'setuptools<64' ; fi