MySQL HeatWave has added several new features, improved operational efficiency, and improved query performance in 8.0.30.
We have enhanced HeatWave ML to provide user options to customize various aspects of the machine learning training pipeline including algorithm selection, feature selection and hyperparameter optimization. Users can indicate specific algorithms to be included or excluded from the automated training process, a particular set of columns can be excluded, or the model can be optimized for a specific scoring metric. For more information, go to here
You can now train HeatWave ML model on tables containing date, time, year, datetime and timestamp datatypes.
We have added support for Support Vector Machine (SVM) algorithms as a candidate while selecting the best algorithm for both classification and regression tasks. For the complete list of algorithms supported, go to here
HeatWave ML now supports a more interpretable explanations, providing user with a textual explanation for their prediction. The user is also informed of the quality of the trained model, allowing them to revisit their data to further improve the model.
HeatWave now supports built-in server-side data masking and de-identification to help you protect sensitive data from unauthorized uses by hiding and replacing real values with substitutes. This masking, data substitution and blurring that obfuscate of sensitive data is done in the server and queries involving these functions can be accelerated by HeatWave. For more information, go to here
We have extended Auto Error Recovery to cover the case of MySQL node failure. Previously, when MySQL node failed for some reason and restarted, the user had to restart the HeatWave cluster and reload data manually. Now, when MySQL fails and restarts, HeatWave cluster automatically restarts, identifies the tables which were loaded prior to the failure and reloads those tables automatically from MySQL. This reduces intervention on part of the user and also improves service up time.
HeatWave automatically reloads data from MySQL InnoDB after MySQL node restarts due to maintenance upgrade or planned restarts. Previously, the user had to identify the tables which were loaded into HeatWave and then load those tables. With auto reload capability, you no longer need to take manual steps after maintenance or restart operation - this reduces the operational overhead and improve service availability.
Data sizes and the complexity of schema which customers are using MySQL HeatWave for is increasing steadily. In order to keep up with the demand, MySQL HeatWave has increased the maximum number of tables HeatWave can handle to 1.6M tables (400K table for cluster with MySQL node in MySQL.HeatWave.VM.E3.Standard shape, and 1.6M tables for cluster with MySQL node in MySQL.HeatWave.BM.E3.Standard shape). This is 4X more than previously allowed.
MySQL AutoPilot Auto Encoding has been augmented to include query performance in addition to memory usage to predict optimal encoding. It also provides prediction of the expected performance improvement with the suggested encoding change. With this enhancement, for a 30TB TPCH workload, Auto Encoding recommendation improves the query runtime by 49%. For more information, go to here
Logical optimizer enhancement
Enhancements have been made to the MySQL and HeatWave logical optimizer to reduce query compilation and execution time. Queries with semi-joins and queries where filter can be pushed down will see improvement in query performance. More classes of queries, such as queries with complex predicates and post join filters will now get offloaded to HeatWave.
Greatest/Least, FROM_DAYS Support
Queries with GREATEST(), LEAST() and/or FROM_DAYS() functions are now accelerated in HeatWave.
Join and Group by queries enhancement
Partitioning operator in HeatWave facilitates massively parallel query processing. However, in presence of large number of projection columns, the partitioning operation for join and group by queries could become a bottleneck. Now the partition operation has been optimized to improve query performance of joins and group queries with large numbers of projection columns.
Addition Resources
MySQL HeatWave Product Manager