Add latest cuGraph

.dockerignore

@@ -5,5 +5,6 @@ htmlcov
 dist
 **/__pycache__
 cpp/build
+cpp/memgraph/build
 Dockerfile*
-.dockerignore
+.dockerignore

.github/workflows/test.yml

@@ -9,7 +9,8 @@ env:
   NEO4J_CONTAINER: "neo4j_test"
   OFFICIAL: "true"
 
-on: [pull_request, workflow_dispatch]
+# TODO(gitbuda): Revert
+# on: [pull_request, workflow_dispatch]
 
 jobs:
   build:

.github/workflows/test_no_ml.yml

@@ -9,7 +9,8 @@ env:
   NEO4J_CONTAINER: "neo4j_test"
   OFFICIAL: "true"
 
-on: [pull_request, workflow_dispatch]
+# TODO(gitbuda): Revert
+# on: [pull_request, workflow_dispatch]
 
 jobs:
   build:

Dockerfile.cugraph

@@ -1,12 +1,13 @@
-ARG CUGRAPH_VERSION=22.02
-ARG CUDA_VERSION=11.5
-ARG CUDA_VERSION_MINOR=11.5.2
-ARG PY_VERSION=3.8
-ARG MG_VERSION=2.14.1
+ARG CUGRAPH_VERSION=24.06a
+ARG CUDA_VERSION=12.2
+ARG CUDA_VERSION_MINOR=12.2.2
+ARG PY_VERSION=3.11
+ARG MG_VERSION=2.15.1
 
-FROM rapidsai/rapidsai:${CUGRAPH_VERSION}-cuda${CUDA_VERSION}-runtime-ubuntu20.04-py${PY_VERSION} as cugraph-dev
+# TODO(gitbuda): Probably turn this into Dockerfile.only_cugraph
 
-FROM nvidia/cuda:${CUDA_VERSION_MINOR}-devel-ubuntu20.04 AS dev
+FROM rapidsai/base:${CUGRAPH_VERSION}-cuda${CUDA_VERSION}-py${PY_VERSION} as cugraph-dev
+FROM nvidia/cuda:${CUDA_VERSION_MINOR}-devel-ubuntu22.04 AS dev
 
 USER root
 
@@ -17,15 +18,12 @@ ENV MG_VERSION ${MG_VERSION}
 ENV PY_VERSION ${PY_VERSION}
 
 # Copy RAPIDS libraries
-COPY --from=cugraph-dev /opt/conda/envs/rapids/lib/libcugraph.so /opt/conda/envs/rapids/lib/libcugraph.so
-COPY --from=cugraph-dev /opt/conda/envs/rapids/include /opt/conda/envs/rapids/include
+COPY --from=cugraph-dev /opt/conda/lib/libcugraph.so /opt/conda/lib/libcugraph.so
+COPY --from=cugraph-dev /opt/conda/include /opt/conda/include
 
 # Prevent from linking the Conda environment
 ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/lib
 
-# NVIDIA key rotation
-RUN rm /etc/apt/sources.list.d/cuda.list
-
 # Essentials for production/dev
 RUN apt-get update && apt-get install -y \
     libcurl4        `memgraph` \
@@ -62,29 +60,30 @@ EXPOSE 7687
 # Copy and build MAGE
 WORKDIR /mage
 COPY . /mage
+# TODO(gitbuda): In the meantime, we moved the whole memgraph repo under cpp -> init submodules.
 
 RUN curl https://sh.rustup.rs -sSf | sh -s -- -y && \
     export PATH="/root/.cargo/bin:${PATH}" && \ 
     python3 -m pip install -r /mage/python/requirements.txt && \ 
     python3 -m  pip install -r /mage/python/tests/requirements.txt && \
-    python3 -m pip install dgl -f https://data.dgl.ai/wheels/repo.html && \ 
     python3 /mage/setup build \ 
     --gpu \
-    --cpp-build-flags MAGE_CUGRAPH_ROOT=/opt/conda/envs/rapids/ CMAKE_BUILD_TYPE=Release \
+    --cpp-build-flags MAGE_CUGRAPH_ROOT=/opt/conda/ CMAKE_BUILD_TYPE=Release \
     -p /usr/lib/memgraph/query_modules/ 
+    # python3 -m pip install dgl -f https://data.dgl.ai/wheels/repo.html && \  # TODO(gitbuda): Figure out DGL
 
-#DGL build from source
-RUN git clone --recurse-submodules -b 0.9.x https://github.com/dmlc/dgl.git   \
-    && cd dgl && mkdir build && cd build && cmake -DUSE_CUDA=ON .. \
-    && make -j4 && cd ../python && python3 setup.py install
+# # TODO(gitbuda): Figure out DGL
+# # Build DGL from source.
+# RUN git clone --recurse-submodules -b 2.1.x https://github.com/dmlc/dgl.git   \
+#     && cd dgl && mkdir build && cd build && cmake -DUSE_CUDA=ON .. \
+#     && make -j4 && cd ../python && python3 setup.py install
 
 USER memgraph
 ENTRYPOINT ["/usr/lib/memgraph/memgraph"]
 
-FROM nvidia/cuda:${CUDA_VERSION_MINOR}-runtime-ubuntu20.04 AS prod
+FROM nvidia/cuda:${CUDA_VERSION_MINOR}-runtime-ubuntu22.04 AS prod
 
 USER root
-
 ARG DEBIAN_FRONTEND=noninteractive
 ARG MG_VERSION
 ARG PY_VERSION
@@ -94,9 +93,9 @@ ENV PY_VERSION ${PY_VERSION}
 # Copy modules
 COPY --from=dev /usr/lib/memgraph/query_modules/ /usr/lib/memgraph/query_modules/
 # Copy cugraph library
-COPY --from=dev /opt/conda/envs/rapids/lib/libcugraph.so /opt/conda/envs/rapids/lib/libcugraph.so
+COPY --from=dev /opt/conda/lib/libcugraph.so /opt/conda/lib/libcugraph.so
 # Copy python build
-COPY --from=dev /usr/local/lib/python${PY_VERSION}/ /usr/local/lib/python${PY_VERSION}/
+COPY --from=dev /usr/lib/python${PY_VERSION}/ /usr/lib/python${PY_VERSION}/
 
 # NVIDIA key rotation
 RUN rm /etc/apt/sources.list.d/cuda.list
@@ -112,15 +111,15 @@ RUN apt-get update && apt-get install -y \
     libgomp1         `mage-memgraph` \
     python3         `mage-memgraph` \
     python3-setuptools     `mage-memgraph` \
-    && curl https://download.memgraph.com/memgraph/v${MG_VERSION}/ubuntu-20.04/memgraph_${MG_VERSION}-1_amd64.deb --output memgraph.deb \
+    && curl https://download.memgraph.com/memgraph/v${MG_VERSION}/ubuntu-22.04/memgraph_${MG_VERSION}-1_amd64.deb --output memgraph.deb \
     && dpkg -i memgraph.deb \
     && rm memgraph.deb \
     && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
 
-RUN export PATH="/usr/local/lib/python${PY_VERSION}:${PATH}"
+RUN export PATH="/usr/lib/python${PY_VERSION}:${PATH}"
 
 RUN rm -rf /mage \
-    && export PATH="/usr/local/lib/python${PY_VERSION}:${PATH}" \
+    && export PATH="/usr/lib/python${PY_VERSION}:${PATH}" \
     && apt-get -y --purge autoremove curl python3-dev \
     && apt-get clean
 

Dockerfile.experiment_full

@@ -0,0 +1,118 @@
+# It's important to be 3.10 on Ubuntu 22.04 because memgraph binary depends on exactly that version.
+ARG PY_VERSION=3.10
+ARG CUDA_VERSION=12.2
+ARG CUDA_VERSION_MINOR=12.2.2
+ARG RAPIDS_VERSION=24.04a
+ARG MG_VERSION=2.16.0
+# TODO(gitbuda): Try to build on ARM and adjust settings if needed.
+
+## COMPILATION
+FROM rapidsai/base:${RAPIDS_VERSION}-cuda${CUDA_VERSION}-py${PY_VERSION} as rapids-dev
+FROM nvidia/cuda:${CUDA_VERSION_MINOR}-devel-ubuntu22.04 AS cuda-dev
+USER root
+ARG DEBIAN_FRONTEND=noninteractive
+ARG PY_VERSION
+ENV PY_VERSION ${PY_VERSION}
+
+### SYSTEM SETUP
+COPY --from=rapids-dev /opt/conda/lib/libcugraph.so /opt/conda/lib/libcugraph.so
+COPY --from=rapids-dev /opt/conda/lib/libcugraph-ops++.so /opt/conda/lib/libcugraph-ops++.so
+COPY --from=rapids-dev /opt/conda/lib/libraft.so /opt/conda/lib/libraft.so
+COPY --from=rapids-dev /opt/conda/include /opt/conda/include
+# Prevent from linking the Conda environment
+ENV LD_LIBRARY_PATH=/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64:/usr/local/lib
+RUN apt-get update && apt-get install -y       \
+    vim                        `memgraph`      \
+    libcurl4                   `memgraph`      \
+    libpython${PY_VERSION}     `memgraph`      \
+    libssl3                    `memgraph`      \
+    libssl-dev                 `memgraph`      \
+    openssl                    `memgraph`      \
+    build-essential            `mage-memgraph` \
+    cmake                      `mage-memgraph` \
+    curl                       `mage-memgraph` \
+    g++                        `mage-memgraph` \
+    python3                    `mage-memgraph` \
+    python3-pip                `mage-memgraph` \
+    python3-setuptools         `mage-memgraph` \
+    python3-dev                `mage-memgraph` \
+    clang                      `mage-memgraph` \
+    git                        `mage-memgraph` \
+    unixodbc-dev               `mage-memgraph` \
+    software-properties-common `mage-cugraph`  \
+    lsb-release                `mage-cugraph`  \
+    wget                       `mage-cugraph`  \
+    --no-install-recommends                    \
+    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/* \
+    # Install newest CMake (cuGraph requires >= 20.01)
+    && wget -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc | apt-key add - && \
+    apt-add-repository "deb https://apt.kitware.com/ubuntu/ $(lsb_release -cs) main" && \
+    apt-get install -y \
+    cmake           `mage-memgraph` \
+    --no-install-recommends
+ENV PATH=/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/bin/cmake:/usr/lib/cmake
+# TODO(gitbuda): How to cleanup rust installation cache?
+RUN curl https://sh.rustup.rs -sSf | sh -s -- -y
+ENV PATH="${PATH}:/root/.cargo/bin"
+# # NOTE: miniconda will be installed under /root/miniconda3
+# RUN cd /root && \
+#     curl -o miniconda.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
+#     bash miniconda.sh -b
+WORKDIR /mage
+COPY . /mage
+RUN python3 -m pip install -r /mage/python/requirements.txt && \
+    python3 -m pip install -r /mage/python/tests/requirements.txt
+# TODO(gitbuda): Experiment with pipx, the below works on debian-12 (on newer systems)
+# RUN python3 -m pip install --break-system-packages -r /mage/python/requirements.txt && \
+#     python3 -m pip install --break-system-packages -r /mage/python/tests/requirements.txt
+
+### BUILD
+RUN python3 /mage/setup build \
+    --gpu \
+    --cpp-build-flags \
+      CMAKE_BUILD_TYPE=RelWithDebInfo \
+      MAGE_CUGRAPH_ROOT=/opt/conda/ \
+    -p /usr/lib/memgraph/query_modules/
+
+## RUNTIME
+# TODO(gitbuda): Add the ability to inject local RelWithDebInfo package of Memgraph
+FROM nvidia/cuda:${CUDA_VERSION_MINOR}-runtime-ubuntu22.04 AS cuda-prod
+USER root
+ARG DEBIAN_FRONTEND=noninteractive
+ARG MG_VERSION
+ENV MG_VERSION ${MG_VERSION}
+ARG PY_VERSION
+ENV PY_VERSION ${PY_VERSION}
+
+#### SYSTEM SETUP
+COPY --from=cuda-dev /usr/lib/memgraph/query_modules/ /usr/lib/memgraph/query_modules/
+COPY --from=cuda-dev /opt/conda/lib/libcugraph.so /opt/conda/lib/libcugraph.so
+COPY --from=cuda-dev /opt/conda/lib/libcugraph-ops++.so /opt/conda/lib/libcugraph-ops++.so
+COPY --from=cuda-dev /opt/conda/lib/libraft.so /opt/conda/lib/libraft.so
+COPY --from=cuda-dev /usr/lib/python${PY_VERSION}/ /usr/lib/python${PY_VERSION}/
+# TODO(gitbuda): Consider how to copy all the requirements from python -> because the above is not full, e.g. networkx is not working.
+RUN rm -rf /etc/apt/sources.list.d/cuda*
+RUN apt-get update && apt-get install -y \
+    libcurl4               `memgraph` \
+    libpython${PY_VERSION} `memgraph` \
+    libssl3                `memgraph` \
+    libssl-dev             `memgraph` \
+    openssl                `memgraph` \
+    curl                   `mage-memgraph` \
+    libgomp1               `mage-memgraph` \
+    python3                `mage-memgraph` \
+    python3-setuptools     `mage-memgraph` \
+    && curl https://download.memgraph.com/memgraph/v${MG_VERSION}/ubuntu-22.04/memgraph_${MG_VERSION}-1_amd64.deb --output memgraph.deb \
+    && dpkg -i memgraph.deb \
+    && rm memgraph.deb \
+    && rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*
+ENV PATH="/usr/lib/python${PY_VERSION}:${PATH}"
+# TODO(gitbuda): extra packages python3-setuptools, python3-pip, cmake -> make sure the dev step wasn't wrongly cached.
+# TODO(gitbuda): The rm part should be under the apt install RUN part (/mage is not present here)
+RUN rm -rf /mage \
+    && export PATH="/usr/lib/python${PY_VERSION}:${PATH}" \
+    && apt-get -y --purge autoremove curl python3-dev \
+    && apt-get clean
+
+USER memgraph
+ENTRYPOINT ["/usr/lib/memgraph/memgraph"]

Dockerfile.experiment_partial

@@ -0,0 +1,13 @@
+ARG PY_VERSION=3.11
+ARG RAPIDS_VERSION=24.06a
+ARG CUDA_VERSION=12.2
+ARG CUDA_VERSION_MINOR=12.2.2
+ARG MG_VERSION=2.15.1
+
+FROM rapidsai/base:${RAPIDS_VERSION}-cuda${CUDA_VERSION}-py${PY_VERSION} as rapids-dev
+RUN ls -alh /opt/ && ls -alh /opt/conda/ && ls -alh /opt/conda/lib/ | grep cugraph
+RUN nvcc --version
+#
+# FROM nvidia/cuda:${CUDA_VERSION_MINOR}-devel-ubuntu22.04 AS cuda-dev
+# RUN ls -alh /opt/ && ls -alh /opt/nvidia/
+# RUN nvcc --version

cpp/CMakeLists.txt

@@ -1,25 +1,40 @@
 # Memgraph Mage C++ Query Modules CMake configuration.
 cmake_minimum_required(VERSION 3.14)
 set(MEMGRAPH_MAGE_PROJECT_NAME "memgraph-mage")
+
+if (MAGE_CUGRAPH_ENABLE)
+  # TODO(gitbuda): Duplicated under cugraph.cmake -> FIX.
+  set(MAGE_CUGRAPH_TAG "v24.04.00" CACHE STRING "cuGraph GIT tag to checkout" )
+  # RAPIDS.cmake is here because rapids_cuda_init_architectures is required to
+  # properly set both CMAKE_CUDA_ARCHITECTURES and CUDA_ARCHITECTURES target
+  # property.
+  file(DOWNLOAD "https://raw.githubusercontent.com/rapidsai/rapids-cmake/${MAGE_CUGRAPH_TAG}a/RAPIDS.cmake"
+                ${CMAKE_BINARY_DIR}/RAPIDS.cmake)
+  include(${CMAKE_BINARY_DIR}/RAPIDS.cmake)
+  # NOTE: RAPIDS init is required to be called before project().
+  include(rapids-cuda)
+  rapids_cuda_init_architectures("${MEMGRAPH_MAGE_PROJECT_NAME}")
+  # rapids_cuda_patch_toolkit()
+endif()
 project("${MEMGRAPH_MAGE_PROJECT_NAME}" LANGUAGES C CXX)
-include(FetchContent)
+if (MAGE_CUGRAPH_ENABLE)
+  find_package(CUDAToolkit REQUIRED)
+  enable_language(CUDA)
+  add_definitions(-DLIBCUDACXX_ENABLE_EXPERIMENTAL_MEMORY_RESOURCE)
+endif()
 
+include(FetchContent)
 
 # setup CMake module path, defines path for include() and find_package()
 # https://cmake.org/cmake/help/latest/variable/CMAKE_MODULE_PATH.html
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${PROJECT_SOURCE_DIR}/cugraph_cmake)
-
 # Export the compile commands so that we can use clang-tidy. Additional benefit
 # is easier debugging of compilation and linker flags.
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
-
-# Required C++ standard.
 set(CMAKE_C_STANDARD 11)
 set(CMAKE_C_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
-
-
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 
 # Compile flags.
@@ -116,9 +131,10 @@ add_subdirectory(mg_utility)
 
 FetchContent_Declare(fmt
   GIT_REPOSITORY https://github.com/fmtlib/fmt.git
-  GIT_TAG 9.1.0
+  GIT_TAG 10.2.1
 )
 FetchContent_MakeAvailable(fmt)
+add_compile_definitions(SPDLOG_FMT_EXTERNAL)
 
 function(add_query_module target_name version src)
   add_library(${target_name} SHARED ${src})

cpp/cugraph_cmake/cugraph.cmake

@@ -27,37 +27,39 @@
 # Order of the languages matters because cmake pics different compilers.
 #
 # Compiling cugraph takes ages and it's complex.
-# TODO(gitbuda): Allow linking of an already compiled version of cugraph.
+#   -> Defining MAGE_CUGRAPH_ROOT will skip compilation.
 #
 # CUDA_ARCHITECTURES ->
 # https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
 # Rapids CMake add NATIVE + ALL options as CUDA_ARCHITECTURES which simplifies
 # build configuration.
+#
+# CUDA + CMake -> https://developer.download.nvidia.com/video/gputechconf/gtc/2019/presentation/s9444-build-systems-exploring-modern-cmake-cuda-v2.pdf
 
 option(MAGE_CUGRAPH_ENABLE "Enable cuGraph build" OFF)
 message(STATUS "MAGE cuGraph build enabled: ${MAGE_CUGRAPH_ENABLE}")
 
 if (MAGE_CUGRAPH_ENABLE)
   # Version of cuGraph for local build
-  set(MAGE_CUGRAPH_TAG "v22.02.00" CACHE STRING "cuGraph GIT tag to checkout" )
+  # set(MAGE_CUGRAPH_TAG "v24.04.00" CACHE STRING "cuGraph GIT tag to checkout" )
   set(MAGE_CUGRAPH_REPO "https://github.com/rapidsai/cugraph.git" CACHE STRING "cuGraph GIT repo URL")
   # Custom MAGE_CUGRAPH_BUILD_TYPE.
   set(MAGE_CUGRAPH_BUILD_TYPE "Release" CACHE STRING "Passed to cuGraph as CMAKE_BUILD_TYPE")
   # NATIVE | ALL -> possible because cugraph calls rapids_cuda_init_architectures
   set(MAGE_CUDA_ARCHITECTURES "NATIVE" CACHE STRING "Passed to cuGraph as CMAKE_CUDA_ARCHITECTURES")
 
-  # RAPIDS.cmake is here because rapids_cuda_init_architectures is required to
-  # properly set both CMAKE_CUDA_ARCHITECTURES and CUDA_ARCHITECTURES target
-  # property.
-  file(DOWNLOAD "https://raw.githubusercontent.com/rapidsai/rapids-cmake/${MAGE_CUGRAPH_TAG}a/RAPIDS.cmake"
-                ${CMAKE_BINARY_DIR}/RAPIDS.cmake)
-  include(${CMAKE_BINARY_DIR}/RAPIDS.cmake)
-
-  include(rapids-cuda)
-  rapids_cuda_init_architectures("${MEMGRAPH_MAGE_PROJECT_NAME}")
-  enable_language(CUDA)
+  # # RAPIDS.cmake is here because rapids_cuda_init_architectures is required to
+  # # properly set both CMAKE_CUDA_ARCHITECTURES and CUDA_ARCHITECTURES target
+  # # property.
+  # file(DOWNLOAD "https://raw.githubusercontent.com/rapidsai/rapids-cmake/${MAGE_CUGRAPH_TAG}a/RAPIDS.cmake"
+  #               ${CMAKE_BINARY_DIR}/RAPIDS.cmake)
+  # include(${CMAKE_BINARY_DIR}/RAPIDS.cmake)
+  #
+  # include(rapids-cuda)
+  # rapids_cuda_init_architectures("${MEMGRAPH_MAGE_PROJECT_NAME}")
+  # enable_language(CUDA)
 
-  set(CMAKE_CUDA_STANDARD 17)
+  set(CMAKE_CUDA_STANDARD 20)
   set(CMAKE_CUDA_STANDARD_REQUIRED ON)
 
   message(STATUS "MAGE cuGraph root: ${MAGE_CUGRAPH_ROOT}")
@@ -108,6 +110,6 @@ macro(target_mage_cugraph target_name)
       PRIVATE "$<$<COMPILE_LANGUAGE:CXX>:${MAGE_CXX_FLAGS}>"
               "$<$<COMPILE_LANGUAGE:CUDA>:${MAGE_CUDA_FLAGS}>"
     )
-    target_link_libraries("${target_name}" PRIVATE mage_cugraph)
+    target_link_libraries("${target_name}" PRIVATE mage_cugraph CUDA::toolkit)
   endif()
 endmacro()

cpp/cugraph_module/CMakeLists.txt

@@ -46,4 +46,4 @@ target_include_directories(cugraph.hits PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
 add_query_module(cugraph.generator 1 algorithms/graph_generator.cu)
 target_mage_cugraph(cugraph.generator)
 target_link_libraries(cugraph.generator PRIVATE mg_utility)
-target_include_directories(cugraph.generator PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
+target_include_directories(cugraph.generator PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})