⚠️ Warning
SingleStore 9.1 allows you to preview, evaluate, and provide feedback on new and upcoming features before their general availability. In the meantime, SingleStore 9.0 is recommended for production workloads, which can later be upgraded to SingleStore 9.1.
SingleStore Self-Managed RC Announcement – The performance and functionality you need for enterprise data & AI applications
We’re thrilled to announce the latest SingleStore database engine features, available as a release candidate (RC). These features strengthen SingleStore’s support for enterprise data & AI application workloads.
Vector search is cheaper and faster, new observability tools simplify monitoring, and new query execution capabilities reduce application complexity and improve performance.
F16 Vectors - Cheaper, Faster, Better
The new float16 (F16) vectors use half the storage space of our existing float32 (F32) vectors, while improving KNN search performance and maintaining ANN search performance and recall for machine learning use cases.
Exact KNN queries on F16 vectors are up to 38% faster than queries using F32 vectors for DOT_PRODUCT and EUCLIDEAN_DISTANCE. ANN index search on F16 vectors shows modest gains of 4–8% due to better cache utilization from the smaller memory footprint. Recall is statistically equivalent.
For the majority of machine learning vector search use cases, F16 delivers the same results at half the storage cost and with faster query performance.
Full-Text Search - Performance and Observability
A new full-text index merger (preview feature) improves full-text search performance. The new ANALYZE FULLTEXT command lets you inspect tokens generated by a Lucene analyzer to understand how your data is being indexed.
New observability tools, the SHOW FULLTEXT INDEX STATUS command and the MV_FULLTEXT_INDEX_STATUS information schema view, let you monitor index build status.
Overloaded Functions and Stored Procedures (Preview)
User-defined functions and stored procedures now support overloading. Multiple functions with the same name but different arguments can be created, and SingleStore chooses the correct variant to execute based on argument types.
Developers can use the same function name for the same logical operation across different types, eliminating the need for naming workarounds like PrettyPrint_Int, PrettyPrint_2Ints, and PrettyPrint_Varchar.
Observability
New observability features simplify memory monitoring, DDL auditing, and full-text index build monitoring.
MV_MEMORY_USAGE allows you to identify the most significant memory consumers. MV_TRACE_EVENTS, also viewable in Query History, traces DDL. SHOW FULLTEXT INDEX STATUS and MV_FULLTEXT_INDEX_STATUS display full-text index build status.
Data Integration
Kafka Connect Pipelines leverage the Kafka Connect ecosystem to introduce an extensibility framework for additional queuing systems. Amazon Kinesis is the first supported data source.
Query Execution
New UPDATE … RETURNING and DELETE … RETURNING statements return a result set of affected rows in a single statement, eliminating the need for a separate SELECT and reducing table scans to improve performance.
The Distributed Plancache (DPC) has been enhanced with a Plancache Manager and asynchronous lookup, improving performance for Fast Scale and Suspend & Resume scenarios.
Additional improvements include support for correlated EXISTS subqueries, writable view enhancements, the ability to cancel running columnstore alter operations, and feature and performance enhancements to JSON processing.
Try this Release Candidate Build
Self-managed customers can explore these new features by deploying a new, non-production SingleStore 9.1 cluster.
Refer to the release notes for additional information about the features.