Changes in HeatWave 9.3.0 (2025-04-15, General Availability)

Note

These release notes were created with the assistance of HeatWave GenAI.

Important

As of MySQL 9.3.0, to help you generate better quality embeddings, HeatWave GenAI uses multilingual-e5-small as the default embedding model for encoding documents in all supported languages including English. This means that minilm is no longer used as the default embedding model for encoding English documents.

This default value change impacts the following processes:

• New vector store tables are created using multilingual-e5-small by default.

• Retrieval augmented generation (RAG) searches tables created using the default embedding model, multilingual-e5-small, unless you explicitly specify the embedding model to use.

• HeatWave GenAI

• HeatWave Lakehouse

• HeatWave MySQL

HeatWave GenAI

• HeatWave GenAI now supports partial failures with better error reporting in batch queries that are run using the ML_GENERATE_TABLE, ML_RAG_TABLE, and ML_EMBED_TABLE routines. This enhancement allows queries on certain rows to fail in case of errors without discarding successful work on other rows. (WL #16749)

HeatWave Lakehouse

• HeatWave now supports automatic recovery of loaded Lakehouse tables from the HeatWave Storage Layer after a planned or unplanned restart of the primary DB system. When a DB system restarts, data for Lakehouse tables is loaded from the HeatWave Storage Layer in Object Storage. After a successful load, the data is current as of the last refresh of the table. If this operation fails, data for these Lakehouse tables is then loaded from the Object Storage bucket.

  This feature extends the existing functionality of automatic recovery from Storage Layer for Standalone DB systems to High Availability DB systems, thus reducing the recovery time, increasing the availability, and improving the overall system reliability.

  For more information, see Scale-Out Data Management. (WL #16758)

HeatWave MySQL

• HeatWave now supports the creation of temporary tables, which are stored in a hybrid columnar format within the HeatWave Cluster. If you use temporary tables to store intermediate results during data transformation, aggregation or consolidation for data analysis or reporting purposes, you can now accelerate the processes with HeatWave.

  For more information, see Creating HeatWave Temporary Tables. (WL #16541)

• Bulk Load, which is used for ingesting data into the DB System, now supports tables that include the VECTOR data type. This enables you to easily and quickly import existing or pre-generated embeddings to HeatWave for your HeatWave GenAI workloads.

  For more information, see Bulk Ingest Data Type Support (WL #16510)

• HeatWave now supports the use of the ANALYZE TABLE statement to update statistics for tables loaded into the HeatWave Cluster. This enhancement improves query performance by refreshing critical statistics used during query processing. In mixed or Lakehouse workloads, where changing data characteristics can degrade performance, this feature helps maintain optimal query execution by keeping statistics up-to-date.

  For more information, see Analyzing Tables. (WL #16641)

• HeatWave now supports acceleration of queries with general quantified comparison predicates.

  For more information, see Optimizer Notes and Optimizing ANY and ALL Subqueries. (WL #13052)