• PlanetScale has made vector support generally available (https://planetscale.com/blog/... It's a fork of MySQL that allows vector data to be stored with relational MySQL data, eliminating the need for a separate vector database.
• PostgreSQL was the default open-source choice for vector search, but the company behind Vitess database announced in 2023 to fork MySQL and add vector search capabilities. After a public beta in late 2024, performance has improved. Patrick Reynolds, a software engineer at PlanetScale, said query performance has doubled, memory efficiency has improved eight times, and focus is on robustness.
• The new vector capabilities enable direct support for recommendation systems, semantic search, and RAG workloads on a MySQL-compatible engine. Advanced vector-index features like ranking vectors by different distances, storing vectors up to 16,383 dimensions, and supporting both fixed and product quantization are built.
• A key differentiator of PlanetScale's vector support is its ability to use indexes larger than RAM, based on two Microsoft Research papers: SPANN and SPFresh. SPANN enables scaling to larger-than-RAM indexes, and SPFresh defines background operations.
• PlanetScale has designed the SPANN and SPFresh operations to be transactional and integrated them into MySQL's default storage engine. However, there is little hope that Oracle will merge the change into the MySQL Community Edition.
• Writes and queries for the new feature work like a normal RDBMS, using ALTER or CREATE VECTOR INDEX statements and SELECT statements with JOIN and WHERE clauses. The tight integration ensures immediate reflection of data changes in the index and full coverage by the MySQL binlog, with scalability to terabytes of data and integration with the query planner.
• PlanetScale is built on top of Vitess, an open source database clustering system for horizontal scaling of MySQL. A list of compatibility limitations is available online.