HeavyDB is an open source SQL-based, relational, columnar database engine that leverages the full performance and parallelism of modern hardware (both CPUs and GPUs) to enable querying of multi-billion row datasets in milliseconds, without the need for indexing, pre-aggregation, or downsampling. HeavyDB can be run on hybrid CPU/GPU systems (Nvidia GPUs are currently supported), as well as on CPU-only systems featuring X86, Power, and ARM (experimental support) architectures. To achieve maximum performance, HeavyDB features multi-tiered caching of data between storage, CPU memory, and GPU memory, and an innovative Just-In-Time (JIT) query compilation framework.
For usage info, see the product documentation, and for more details about the system's internal architecture, check out the developer documentation. Further technical discussion can be found on the HEAVY.AI Community Forum.
The repository includes a number of third party packages provided under separate licenses. Details about these packages and their respective licenses is at ThirdParty/licenses/index.md.
HEAVY.AI provides pre-built binaries for Linux for stable releases of the project:
- Developer Documentation
- Doxygen-generated Documentation
- Product Documentation
- Release Notes
- Community Forum
- HEAVY.AI Homepage
- HEAVY.AI Blog
- HEAVY.AI Downloads
This project is licensed under the Apache License, Version 2.0.
The repository includes a number of third party packages provided under separate licenses. Details about these packages and their respective licenses is at ThirdParty/licenses/index.md.
In order to clarify the intellectual property license granted with Contributions from any person or entity, HEAVY.AI must have a Contributor License Agreement ("CLA") on file that has been signed by each Contributor, indicating agreement to the Contributor License Agreement. After making a pull request, a bot will notify you if a signed CLA is required and provide instructions for how to sign it. Please read the agreement carefully before signing and keep a copy for your records.
If this is your first time building HeavyDB, install the dependencies mentioned in the Dependencies section below.
HeavyDB uses CMake for its build system.
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=debug ..
make -j 4
The following cmake/ccmake options can enable/disable different features:
-DCMAKE_BUILD_TYPE=release- Build type and compiler options to use. Options areDebug,Release,RelWithDebInfo,MinSizeRel, and unset.-DENABLE_ASAN=off- Enable address sanitizer. Default isoff.-DENABLE_AWS_S3=on- Enable AWS S3 support, if available. Default ison.-DENABLE_CUDA=off- Disable CUDA. Default ison.-DENABLE_CUDA_KERNEL_DEBUG=off- Enable debugging symbols for CUDA kernels. Will dramatically reduce kernel performance. Default isoff.-DENABLE_DECODERS_BOUNDS_CHECKING=off- Enable bounds checking for column decoding. Default isoff.-DENABLE_FOLLY=on- Use Folly. Default ison.-DENABLE_IWYU=off- Enable include-what-you-use. Default isoff.-DENABLE_JIT_DEBUG=off- Enable debugging symbols for the JIT. Default isoff.-DENABLE_ONLY_ONE_ARCH=off- Compile GPU code only for the host machine's architecture, speeding up compilation. Default isoff.-DENABLE_PROFILER=off- Enable google perftools. Default isoff.-DENABLE_STANDALONE_CALCITE=off- Require standalone Calcite server. Default isoff.-DENABLE_TESTS=on- Build unit tests. Default ison.-DENABLE_TSAN=off- Enable thread sanitizer. Default isoff.-DENABLE_CODE_COVERAGE=off- Enable code coverage symbols (clang only). Default isoff.-DPREFER_STATIC_LIBS=off- Static link dependencies, if available. Default isoff. Only works on CentOS.
HeavyDB uses Google Test as its main testing framework. Tests reside under the Tests directory.
The sanity_tests target runs the most common tests. If using Makefiles to build, the tests may be run using:
make sanity_tests
AddressSanitizer can be activated by setting the ENABLE_ASAN CMake flag in a fresh build directory. At this time CUDA must also be disabled. In an empty build directory run CMake and compile:
mkdir build && cd build
cmake -DENABLE_ASAN=on -DENABLE_CUDA=off ..
make -j 4
Finally run the tests:
export ASAN_OPTIONS=alloc_dealloc_mismatch=0:handle_segv=0
make sanity_tests
ThreadSanitizer can be activated by setting the ENABLE_TSAN CMake flag in a fresh build directory. At this time CUDA must also be disabled. In an empty build directory run CMake and compile:
mkdir build && cd build
cmake -DENABLE_TSAN=on -DENABLE_CUDA=off ..
make -j 4
We use a TSAN suppressions file to ignore warnings in third party libraries. Source the suppressions file by adding it to your TSAN_OPTIONS env:
export TSAN_OPTIONS="suppressions=/path/to/heavydb/config/tsan.suppressions"
Finally run the tests:
make sanity_tests
HeavyDB uses CPack to generate packages for distribution. Packages generated on CentOS with static linking enabled can be used on most other recent Linux distributions.
To generate packages on CentOS (assuming starting from top level of the heavydb repository):
mkdir build-package && cd build-package
cmake -DPREFER_STATIC_LIBS=on -DCMAKE_BUILD_TYPE=release ..
make -j 4
cpack -G TGZ
The first command creates a fresh build directory, to ensure there is nothing left over from a previous build.
The second command configures the build to prefer linking to the dependencies' static libraries instead of the (default) shared libraries, and to build using CMake's release configuration (enables compiler optimizations). Linking to the static versions of the libraries libraries reduces the number of dependencies that must be installed on target systems.
The last command generates a .tar.gz package. The TGZ can be replaced with, for example, RPM or DEB to generate a .rpm or .deb, respectively.
The startheavy wrapper script may be used to start HeavyDB in a testing environment. This script performs the following tasks:
- initializes the
datastorage directory viainitdb, if required - starts the main HeavyDB server,
heavydb - offers to download and import a sample dataset, using the
insert_sample_datascript
Assuming you are in the build directory, and it is a subdirectory of the heavydb repository, startheavy may be run by:
../startheavy
It is assumed that the following commands are run from inside the build directory.
Initialize the data storage directory. This command only needs to be run once.
mkdir data && ./bin/initdb data
Start the HeavyDB server:
./bin/heavydb
If desired, insert a sample dataset by running the insert_sample_data script in a new terminal:
../insert_sample_data
You can now start using the database. The heavysql utility may be used to interact with the database from the command line:
./bin/heavysql -p HyperInteractive
where HyperInteractive is the default password. The default user admin is assumed if not provided.
Contributed code should compile without generating warnings by recent compilers on most Linux distributions. Changes to the code should follow the C++ Core Guidelines.
A .clang-format style configuration, based on the Chromium style guide, is provided at the top level of the repository. Please format your code using a recent version (8.0+ preferred) of ClangFormat before submitting.
To use:
clang-format -i File.cpp
A .clang-tidy configuration is provided at the top level of the repository. Please lint your code using a recent version (6.0+ preferred) of clang-tidy before submitting.
clang-tidy requires all generated files to exist before running. The easiest way to accomplish this is to simply run a full build before running clang-tidy. A build target which runs clang-tidy is provided. To use:
make run-clang-tidy
Note: clang-tidy may make invalid or overly verbose changes to the source code. It is recommended to first commit your changes, then run clang-tidy and review its recommended changes before amending them to your commit.
Note: the clang-tidy target uses the run-clang-tidy.py script provided with LLVM, which may depend on PyYAML. The target also depends on jq, which is used to filter portions of the compile_commands.json file.
HeavyDB has the following dependencies:
| Package | Min Version | Required |
|---|---|---|
| CMake | 3.16 | yes |
| LLVM | 9.0 | yes |
| GCC | 8.4.0 | no, if building with clang |
| Go | 1.12 | yes |
| Boost | 1.72.0 | yes |
| OpenJDK | 1.7 | yes |
| CUDA | 11.0 | yes, if compiling with GPU support |
| gperftools | yes | |
| gdal | 2.4.2 | yes |
| Arrow | 3.0.0 | yes |
HeavyDB requires a number of dependencies which are not provided in the common CentOS/RHEL package repositories. A prebuilt package containing all these dependencies is provided for CentOS 7 (x86_64).
Use the scripts/mapd-deps-prebuilt.sh build script to install prebuilt dependencies.
These dependencies will be installed to a directory under /usr/local/mapd-deps. The mapd-deps-prebuilt.sh script also installs Environment Modules in order to simplify managing the required environment variables. Log out and log back in after running the mapd-deps-prebuilt.sh script in order to active Environment Modules command, module.
The mapd-deps environment module is disabled by default. To activate for your current session, run:
module load mapd-deps
To disable the mapd-deps module:
module unload mapd-deps
WARNING: The mapd-deps package contains newer versions of packages such as GCC and ncurses which might not be compatible with the rest of your environment. Make sure to disable the mapd-deps module before compiling other packages.
Instructions for installing CUDA are below.
It is preferred, but not necessary, to install CUDA and the NVIDIA drivers using the .rpm using the instructions provided by NVIDIA. The rpm (network) method (preferred) will ensure you always have the latest stable drivers, while the rpm (local) method allows you to install does not require Internet access.
The .rpm method requires DKMS to be installed, which is available from the Extra Packages for Enterprise Linux repository:
sudo yum install epel-release
Be sure to reboot after installing in order to activate the NVIDIA drivers.
The mapd-deps-prebuilt.sh script includes two files with the appropriate environment variables: mapd-deps-<date>.sh (for sourcing from your shell config) and mapd-deps-<date>.modulefile (for use with Environment Modules, yum package environment-modules). These files are placed in mapd-deps install directory, usually /usr/local/mapd-deps/<date>. Either of these may be used to configure your environment: the .sh may be sourced in your shell config; the .modulefile needs to be moved to the modulespath.
The scripts/mapd-deps-centos.sh script is used to build the dependencies. Modify this script and run if you would like to change dependency versions or to build on alternative CPU architectures.
cd scripts
module unload mapd-deps
./mapd-deps-centos.sh --compress
scripts/mapd-deps-osx.sh is provided that will automatically install and/or update Homebrew and use that to install all dependencies. Please make sure macOS is completely up to date and Xcode is installed before running. Xcode can be installed from the App Store.
mapd-deps-osx.sh will automatically install CUDA via Homebrew and add the correct environment variables to ~/.bash_profile.
mapd-deps-osx.sh will automatically install Java and Maven via Homebrew and add the correct environment variables to ~/.bash_profile.
Most build dependencies required by HeavyDB are available via APT. Certain dependencies such as Thrift, Blosc, and Folly must be built as they either do not exist in the default repositories or have outdated versions. A prebuilt package containing all these dependencies is provided for Ubuntu 18.04 (x86_64). The dependencies will be installed to /usr/local/mapd-deps/ by default; see the Environment Variables section below for how to add these dependencies to your environment.
HeavyDB requires a newer version of Boost than the version which is provided by Ubuntu 16.04. The scripts/mapd-deps-ubuntu1604.sh build script will compile and install a newer version of Boost into the /usr/local/mapd-deps/ directory.
Use the scripts/mapd-deps-prebuilt.sh build script to install prebuilt dependencies.
These dependencies will be installed to a directory under /usr/local/mapd-deps. The mapd-deps-prebuilt.sh script above will generate a script named mapd-deps.sh containing the environment variables which need to be set. Simply source this file in your current session (or symlink it to /etc/profile.d/mapd-deps.sh) in order to activate it:
source /usr/local/mapd-deps/mapd-deps.sh
The CUDA and mapd-deps lib directories need to be added to LD_LIBRARY_PATH; the CUDA and mapd-deps bin directories need to be added to PATH. The mapd-deps-ubuntu.sh and mapd-deps-prebuilt.sh scripts will generate a script named mapd-deps.sh containing the environment variables which need to be set. Simply source this file in your current session (or symlink it to /etc/profile.d/mapd-deps.sh) in order to activate it:
source /usr/local/mapd-deps/mapd-deps.sh
Recent versions of Ubuntu provide the NVIDIA CUDA Toolkit and drivers in the standard repositories. To install:
sudo apt install -y \
nvidia-cuda-toolkit
Be sure to reboot after installing in order to activate the NVIDIA drivers.
The scripts/mapd-deps-ubuntu.sh and scripts/mapd-deps-ubuntu1604.sh scripts are used to build the dependencies for Ubuntu 18.04 and 16.04, respectively. The scripts will install all required dependencies (except CUDA) and build the dependencies which require it. Modify this script and run if you would like to change dependency versions or to build on alternative CPU architectures.
cd scripts
./mapd-deps-ubuntu.sh --compress
scripts/mapd-deps-arch.sh is provided that will use yay to install packages from the Arch User Repository and a custom PKGBUILD script for Apache Arrow. If you don't have yay yet, install it first: https://github.com/Jguer/yay#installation
Note: Apache Arrow, while available in the AUR, requires a few custom build flags in order to be used with Core. A custom PKGBUILD for it is included.
CUDA and the NVIDIA drivers may be installed using the following.
yay -S \
linux-headers \
cuda \
nvidia
Be sure to reboot after installing in order to activate the NVIDIA drivers.
The cuda package should set up the environment variables required to use CUDA. If you receive errors saying nvcc is not found, then CUDA bin directories need to be added to PATH: the easiest way to do so is by creating a new file named /etc/profile.d/mapd-deps.sh containing the following:
PATH=/opt/cuda/bin:$PATH
export PATH