AlloyDB ScaNN four-level tree indexes GA, supporting up to 10 billion vector rows

The following AlloyDB AI and vector search features and improvements are generally available (GA):

• The alloydb_scann extension supports four-level tree (4LT) ScaNN vector indexes on PostgreSQL 17 and later, providing support for tables with up to 10 billion vector rows. For more information, see Four-level ScaNN tree indexes.

• 4LT ScaNN vector indexes support automatic index maintenance. For more information, see Create a ScaNN index and Maintain vector indexes automatically.

• You can enable streaming functionality for ScaNN vector indexes by setting the scann.satisfy_limit parameter to relaxed_order. When enabled, the vector scan continues searching through additional leaf partitions until it finds enough results to satisfy the LIMIT clause of your query, improving recall on filtered searches. Additionally, you can enable this feature for parallel process scans by setting the scann.enable_parallel_satisfy_limit database flag to on. For more information, see ScaNN index reference and Best practices for tuning ScaNN indexes.

• Vector assist simplifies the deployment and management of your AlloyDB vector workloads by helping you set up production-ready vector search capabilities, including embedding generation, query optimization, and index creation for vector types such as Hierarchical Navigable Small World (HNSW). For more information, see Vector assist overview.

• For AlloyDB for PostgreSQL instances running PostgreSQL 17 and later, the maximum supported value for num_leaves when building ScaNN vector indexes is increased from 30,000,000 to 104,857,600. For more information, see Create a ScaNN index and ScaNN index reference.

• You can use the external_search_fdw extension to connect to external search clusters and query their data directly from your database using the following search sources:

  • Elasticsearch

  • Solr

  • OpenSearch

• AlloyDB supports the rum extension for high-performance full-text search operations on PostgreSQL 17 and earlier. The rum extension extends standard GIN indexes by storing positional information directly in the index, letting you perform faster phrase searches and relevance ranking without accessing the table data. For more information, see Create and manage a RUM index.