SingleStore 8.5 was released on January 9th, 2024. Below you can find the release highlights. In addition, you can see the complete feature list on our Release Notes page and in our announcement Forum Post.
Vector Search Enhancements
Two important new features have been added to improve vector data processing and the performance of vector search.
Indexed ANN vector search facilitates the creation of large-scale semantic search and generative AI applications. Supported index types include inverted file (IVF), hierarchical navigable small world (HNSW), and variants of both based on product quantization (PQ) – a vector compression method. The VECTOR type makes creating, testing, and debugging vector-based applications easier. New infix operators are available for DOT_PRODUCT (
<*>) and EUCLIDEAN_DISTANCE (
<->) to help shorten queries and make them more readable.
- INSERT … ON DUPLICATE KEY… now supports a DELETE option , enabling streaming maintenance of aggregates, e.g. the total number of people currently viewing an online video stream.
- Name length limits for tables, views, and columns have increased to 256 characters (from 64) for easier porting of apps from other database products such as Oracle and Microsoft SQL Server.
- Extensions are supported. These are packages of PSQL and Wasm-based functions, which can now be installed as a unit with a single command rather than one at a time; this eases the management of logical extension packages and improves the sharing of extension code.
- New table%rowtype and tablecolumn%type notation is supported to allow easier creation and evolution of stored procedures and anonymous code blocks that need to declare variables with types that track table objects. This also makes it easier to port code from Oracle to SingleStore.
- Query history is now supported. This allows you to trace every occurrence of every query (above some runtime threshold you can set), including the query text, who ran it, when it ran, how long it took, and much more. This greatly simplifies application performance monitoring and tuning.
- ORDER BY/LIMIT queries now perform better under flexible parallelism.
- More than a dozen query optimization improvements, including correlation statistics, LEFT JOIN optimizations, and multiple new query transformations.
JSON and Kai
- An internal BSON type has been added for use by the Kai MongoDB® -compatible API. This dramatically improves compatibility with MongoDB when using the Kai interface.
- Improved JSON computed column matching, statistics, query optimizer rewrites, and query execution memory management
- Shallow table copy (CREATE TABLE … WITH SHALLOW COPY) copies a table using a copy-on-write approach, copying the metadata for a columnstore table, not the blobs. Blobs of the new table are shared with the original copy. This enables near-instant copy of large tables with minimal storage overhead; benefits include easier development and the ability to back up whole tables in place quickly.
Expect several major improvements to come that will be made available in patch releases of 8.5, including Projections (secondary sort and shard keys), CDC out, Smart DR for the cloud, and Database Branching for the cloud.