Introduction

HybridSE (Hybrid SQL Engine) is a LLVM-based, hybrid-execution and high-performance SQL engine. It can provide fast and consistent execution on heterogeneous SQL data systems, e.g., OLAD database, HTAP system, SparkSQL, and Flink Stream SQL.

HybridSE is a modularized SQL compiler and executor, with following features:

SQL syntax validation
Logical plan generation and optimization
Expression optimization
Online/Offline physical plan generation
Native code generation
Cluster/Standalone runner

By leveraging the power of HybridSE, developer can implement high performance SQL database with ease, or improve performance over existing offline SQL execution engine. Unlike MySQL and SparkSQL, which have buildin SQL engine, HybridSE offers better performance. What's more, it's designed for AI scenarios, did grammar expansion and optimization, making HybridSE more like a modern SQL engine.

HybridSE has following characteristic:

High Performance

Leveraging the power of LLVM JIT, HybridSE can generate binary code dynamicly based on different hardware environment. It also have dozens of plan pass buildin, a more flexible memory management, which ensure High-Performance for HybridSE.
Great Scalability

Thanks to the modularized design, HybridSE can generate logical and physical plan for different stages. With sdk for multiple languages, HybridSE can be used on SQL optimization, regardless the system is realtime OLAD database, distributed OLAP or stream SQL.
Machine Learning Aimed Optimization

Offer the special table join operation and customized UDF/UDAF, which fullfill the requirements of feature extraction and deployment from machine learning applications.
Online-Offline Consistency

Same SQL and CodeGen logical, is guaranteed equivalent meaning and exact same result. Consistency also apply to UDF/UDAF in multiple programing languages.

Quick Start

Requirements

git
docker

Prepare Code & Docker

git clone --recursive https://github.com/4paradigm/HybridSE.git
cd HybridSE
docker run -v `pwd`:/HybridSE -it ghcr.io/4paradigm/hybridsql:0.3.0
cd /HybridSE
# init enviroment before build
./tools/init_env.profile.sh

It is recommended using the docker image listed above for faster start and avoid dependency hole. You may checkout HybridSQL-docker for complete dependency.

Build

cd /HybridSE
mkdir -p build && cd build
cmake ..
# compile the core library
make -j$(nproc) hybridse_core # install coreutils if nproc not found in mac

Install

cd /HybridSE
mkdir -p build && cd build
cmake ..  -DCMAKE_INSTALL_PREFIX="CONFIG_YOUR_HYRBIDSE_INSTALL_DIR"
make -j$(nproc) install

checkout HybridSE Quick start for more information

Run tests

cd /HybridSE
mkdir -p build & cd buid
cmake .. -DTESTING_ENABLE=ON
export SQL_CASE_BASE_DIR=/HybridSE
make -j$(nproc) && make -j$(nproc) test

Run simple engine demo

cd /HybridSE
mkdir build
cd build
cmake ..
make -j$(nproc) hybridse_proto && make -j$(nproc) simple_engine_demo
./src/simple_engine_demo

simple_engine_demo is a in-memory SQL engine implemented on HybridSE。For more information see How to create a simple SQL engine

Run ToyDB

Build ToyDB

cd /HybridSE
mkdir build
cmake .. -DEXAMPLES_ENABLE=ON
make -j$(nproc) hybridse_proto && make toydb -j$(nproc)

Start ToyDB

cd /HybridSE/examples/toydb/onebox
sh start_all.sh
sh start_cli.sh

ToyDB is a simple in-memory database powered by HybridSE, supporting basic CRUD operations. see more information at ToyDB quick start

Related Projects

Project	Status	Description
OpenMLDB	Open Source	NewSQL database optimised for realtime inference and decisioning applications
NativeFlink	Under Develop	High-performance, Batch-Stream-in-onebox FlinkSQL execution engine

Roadmap

ANSI SQL compatibility

HybridSE is compatible with mainstream DDL, DML already，and will support ANSI SQL progressively, which will reduce the cost of migration from other SQL engine.

[2021H1&H2] Enrich standard syntax of Window，support Where, Group By, Join etc
[2021H1&H2] Extend AI-domain specific grammar and UDAF functions

Performance

HybridSE offer dozens of SQL expression and logical plan optimization, standardized optimization pass interface, and will implement more SQL optimization.

[2021H1] Logical and physical plan optimization for batch mode and request mode data processing
[2021H1] High-performance, distributed execution plan generation and codegen
[2021H2] Compilation and codegen optimization for LLVM-based expression
[2021H2] More classic SQL expression pass support

Ecosystem Integration

HybridSE has been integrated by NativeSpark and FEDB already. It can been integrated into NoSQL, OLAP, OLTP system, and will support more open source system in the future.

[2021H2] Adapt to open source SQL compute framework like FlinkSQL
[2021H2] Adapt to various encoding format in row and column, be compatible with Apache Arrow
[2021H2] Support popular programing languages，including C++, Java, Python, Go, Rust etc

Contribution

Report issues in Github Issue
Ask and discuss more in Discussions and slack

License

Apache License 2.0

jingchen2222 / HybridSE