Opengauss AI capability upgrade to create a new AI native database

Database and AI What kind of sparks will burst out when combined ？ Different database vendors 、 The open source community 、 Teachers and students in Colleges and universities don't understand the same . Although it is difficult to form a unified standard in terms of precise concept , However, the overall evolution idea is the same . about openGauss Come on , Since the first version of open source in the community ,openGauss They continue to evolve and contribute code in this field , For this open source openGauss 3.0.0 Version is no exception .

stay 3.0.0 In the version ,openGauss Of AI The field will evolve in many directions ：

1. Integrate existing AI4DB function , Open source openGauss Database autonomy platform ;

2. Refactoring existing AI4DB Ability , Plug in implementation 、 Support service-oriented operation mode ;

3. Support Prometheus ecology ;

4. New slow SQL Root cause analysis 、 New features such as time series prediction , Optimize the existing AI Ability ;

5. DB4AI Function supports more algorithms .

DB4AI The native engine is further upgraded

stay openGauss 3.0.0 in ,DB4AI The native engine supports more machine learning algorithms , For example, support SVM Nonlinear kernel function of , Support XGBoost wait . meanwhile ,openGauss It also provides Explain Interface , The information of the model can be observed .

AI4DB Support service-oriented 、 pluggable

The original openGauss AI4DB Ability is an offline tool form , The database cannot be fully monitored in the background , Nor can you regularly find problems in the database . The latest version ,openGauss Realize the background monitoring service , And regularly check the status of the database system in the background , Thus, an autonomous database platform is formed DBMind. In the form of off-line calculation , Save the diagnosis results , The user can go through Grafana And other software for visualization , So as to find the problem at the first time and know the root cause of the problem .

Due to the need to monitor regularly in the background openGauss The running state of the database system , therefore , It is necessary to connect with the monitoring platform to collect database monitoring indicators and conduct offline calculation . so , stay 3.0.0 In the version ,openGauss Realized two exporter Used with Prometheus Dock with the platform , Its architecture is ：

among ,openGauss-exporter It is used to obtain the monitoring indicators of the database system （metric）,reprocessing-exporter Used to store in Prometheus The data in is processed again . The two above exporter The data of , Can pass Prometheus Collect data regularly .DBMind The system periodically starts from Prometheus Obtain timing data from , And in DBMind Parallel computing on the deployment machine . After the calculation is completed , Store the calculation results in the metabase （meta-database） in . after , The user can get the diagnosis results from the metabase , Further , Can be configured by Grafana And so on .

As shown in the figure above , It is based on the data in the metabase , use Grafana Examples of Visualization .

meanwhile ,openGauss It also fully integrates the existing AI Ability , And redesigned a plug-in mode . for example , The user wants to call the parameter tuning function , Debugging database parameters based on reinforcement learning , It can be realized by the following command ：

gs_dbmind component xtuner tune …

Through the above gs_dbmind command , You can call all AI function , adopt component Sons command , You can call specific AI function . The user can view the help details through the following commands ：

gs_dbmind component --help

Through the above design ,openGauss If community developers want to contribute some kind of database AI function , Then you only need to ensure that the interface can be gs_dbmind Just get it . meanwhile , The developed plug-in can also call DBMind All that's available API, For example, from Prometheus To get data dai （data access interface） Interface , To metabase （meta database） To insert data into dao （database access object） Interfaces, etc. .

AI4DB existing AI Comprehensive improvement of ability

In this release 3.0.0 In the version ,openGauss Recommend for existing indexes 、 Time series prediction and other functions have also been comprehensively upgraded . Added omission scenarios in previous versions . meanwhile , Slow delivery SQL Root cause analysis and recommendation function , help DBA Quickly identify slow SQL, And according to the monitored database operation indicators , adopt AI The feature library recognition algorithm innovatively gives slow recognition SQL Cause and confidence , Give optimization suggestions at the same time .

Develop a comprehensive database for the next step AI Lay the foundation for an autonomous platform

As mentioned above , stay openGauss 3.0.0 In the version , Innovatively completed the study of history AI Integration of capabilities , Discard the burden left by the process of historical research and development , go to the battle-front without any burden , Innovatively realize the serviceability 、 Offline type 、 Plug in 、 Freely assembled DBMind platform , And release it together with the database installation package . For the results after diagnosis , Users can use Grafana And other tools for custom Visualization （ Of course , We will also provide Grafana Example ）.

This will help us further in the future DBMind Lay the foundation for platform upgrading , It is expected that this year openGauss Will be more AI Rich functions into the platform , At the same time, the platform is independent from the existing code warehouse , And provide native Web Front and rear display platforms , At the same time, it supports self repair function , Let users really experience the one click 、 Database autopilot out of the box .

- END -