A lightweight, GPU accelerated, SQL engine built on the RAPIDS.ai ecosystem.
Get Started on app.blazingsql.com
Getting Started | Documentation | Examples | Contributing | License | Blog | Try Now
BlazingSQL is a GPU accelerated SQL engine built on top of the RAPIDS ecosystem. RAPIDS is based on the Apache Arrow columnar memory format, and cuDF is a GPU DataFrame library for loading, joining, aggregating, filtering, and otherwise manipulating data.
BlazingSQL is a SQL interface for cuDF, with various features to support large scale data science workflows and enterprise datasets.
- Query Data Stored Externally - a single line of code can register remote storage solutions, such as Amazon S3.
- Simple SQL - incredibly easy to use, run a SQL query and the results are GPU DataFrames (GDFs).
- Interoperable - GDFs are immediately accessible to any RAPIDS library for data science workloads.
Try our 5-min Welcome Notebook to start using BlazingSQL and RAPIDS AI.
Here's two copy + paste reproducable BlazingSQL snippets, keep scrolling to find example Notebooks below.
Create and query a table from a cudf.DataFrame
with progress bar:
import cudf
df = cudf.DataFrame()
df['key'] = ['a', 'b', 'c', 'd', 'e']
df['val'] = [7.6, 2.9, 7.1, 1.6, 2.2]
from blazingsql import BlazingContext
bc = BlazingContext(enable_progress_bar=True)
bc.create_table('game_1', df)
bc.sql('SELECT * FROM game_1 WHERE val > 4') # the query progress will be shown
Key | Value | |
---|---|---|
0 | a | 7.6 |
1 | b | 7.1 |
Create and query a table from a AWS S3 bucket:
from blazingsql import BlazingContext
bc = BlazingContext()
bc.s3('blazingsql-colab', bucket_name='blazingsql-colab')
bc.create_table('taxi', 's3://blazingsql-colab/yellow_taxi/taxi_data.parquet')
bc.sql('SELECT passenger_count, trip_distance FROM taxi LIMIT 2')
passenger_count | fare_amount | |
---|---|---|
0 | 1.0 | 1.1 |
1 | 1.0 | 0.7 |
You can find our full documentation at docs.blazingdb.com.
- Anaconda or Miniconda installed
- OS Support
- Ubuntu 16.04/18.04 LTS
- CentOS 7
- GPU Support
- Pascal or Better
- Compute Capability >= 6.0
- CUDA Support
- 10.1.2
- 10.2
- Python Support
- 3.7
- 3.8
BlazingSQL can be installed with conda (miniconda, or the full Anaconda distribution) from the blazingsql channel:
conda install -c blazingsql -c rapidsai -c nvidia -c conda-forge -c defaults blazingsql python=$PYTHON_VERSION cudatoolkit=$CUDA_VERSION
Where $CUDA_VERSION is 10.1, 10.2 or 11.0 and $PYTHON_VERSION is 3.7 or 3.8 For example for CUDA 10.1 and Python 3.7:
conda install -c blazingsql -c rapidsai -c nvidia -c conda-forge -c defaults blazingsql python=3.7 cudatoolkit=10.1
For nightly version cuda 11+ are only supported, see https://github.com/rapidsai/cudf#cudagpu-requirements
conda install -c blazingsql-nightly -c rapidsai-nightly -c nvidia -c conda-forge -c defaults blazingsql python=$PYTHON_VERSION cudatoolkit=$CUDA_VERSION
Where $CUDA_VERSION is 11.0 or 11.2 and $PYTHON_VERSION is 3.7 or 3.8 For example for CUDA 11.2 and Python 3.8:
conda install -c blazingsql-nightly -c rapidsai-nightly -c nvidia -c conda-forge -c defaults blazingsql python=3.8 cudatoolkit=11.2
This is the recommended way of building all of the BlazingSQL components and dependencies from source. It ensures that all the dependencies are available to the build process.
conda create -n bsql python=$PYTHON_VERSION
conda activate bsql
./dependencies.sh 21.06 $CUDA_VERSION
Where $CUDA_VERSION is is 10.1, 10.2 or 11.0 and $PYTHON_VERSION is 3.7 or 3.8 For example for CUDA 10.1 and Python 3.7:
conda create -n bsql python=3.7
conda activate bsql
./dependencies.sh 21.06 10.1
The build process will checkout the BlazingSQL repository and will build and install into the conda environment.
cd $CONDA_PREFIX
git clone https://github.com/BlazingDB/blazingsql.git
cd blazingsql
git checkout main
export CUDACXX=/usr/local/cuda/bin/nvcc
./build.sh
NOTE: You can do ./build.sh -h
to see more build options.
$CONDA_PREFIX now has a folder for the blazingsql repository.
For nightly version cuda 11+ are only supported, see https://github.com/rapidsai/cudf#cudagpu-requirements
conda create -n bsql python=$PYTHON_VERSION
conda activate bsql
./dependencies.sh 21.08 $CUDA_VERSION nightly
Where $CUDA_VERSION is 11.0 or 11.2 and $PYTHON_VERSION is 3.7 or 3.8 For example for CUDA 11.2 and Python 3.8:
conda create -n bsql python=3.8
conda activate bsql
./dependencies.sh 21.08 11.2 nightly
The build process will checkout the BlazingSQL repository and will build and install into the conda environment.
cd $CONDA_PREFIX
git clone https://github.com/BlazingDB/blazingsql.git
cd blazingsql
export CUDACXX=/usr/local/cuda/bin/nvcc
./build.sh
NOTE: You can do ./build.sh -h
to see more build options.
NOTE: You can perform static analysis with cppcheck with the command cppcheck --project=compile_commands.json
in any of the cpp project build directories.
$CONDA_PREFIX now has a folder for the blazingsql repository.
To build without the storage plugins (AWS S3, Google Cloud Storage) use the next arguments:
# Disable all storage plugins
./build.sh disable-aws-s3 disable-google-gs
# Disable AWS S3 storage plugin
./build.sh disable-aws-s3
# Disable Google Cloud Storage plugin
./build.sh disable-google-gs
NOTE: By disabling the storage plugins you don't need to install previously AWS SDK C++ or Google Cloud Storage (neither any of its dependencies).
To build without the SQL providers (MySQL, PostgreSQL, SQLite) use the next arguments:
# Disable all SQL providers
./build.sh disable-mysql disable-sqlite disable-postgresql
# Disable MySQL provider
./build.sh disable-mysql
...
NOTES:
- By disabling the storage plugins you don't need to install mysql-connector-cpp=8.0.23 libpq=13 sqlite=3 (neither any of its dependencies).
- Currenlty we support only MySQL. but PostgreSQL and SQLite will be ready for the next version!
User guides and public APIs documentation can be found at here
Our internal code architecture can be built using Spinx.
conda install -c conda-forge doxygen
cd $CONDA_PREFIX
cd blazingsql/docsrc
pip install -r requirements.txt
make doxygen
make html
The generated documentation can be viewed in a browser at blazingsql/docsrc/build/html/index.html
Have questions or feedback? Post a new github issue.
Please see our guide for contributing to BlazingSQL.
Feel free to join our channel (#blazingsql) in the RAPIDS-GoAi Slack: .
You can also email us at info@blazingsql.com or find out more details on BlazingSQL.com.
The RAPIDS suite of open source software libraries aim to enable execution of end-to-end data science and analytics pipelines entirely on GPUs. It relies on NVIDIA® CUDA® primitives for low-level compute optimization, but exposing that GPU parallelism and high-bandwidth memory speed through user-friendly Python interfaces.
The GPU version of Apache Arrow is a common API that enables efficient interchange of tabular data between processes running on the GPU. End-to-end computation on the GPU avoids unnecessary copying and converting of data off the GPU, reducing compute time and cost for high-performance analytics common in artificial intelligence workloads. As the name implies, cuDF uses the Apache Arrow columnar data format on the GPU. Currently, a subset of the features in Apache Arrow are supported.