Bringing Vector Search to OLTP
summary
Cloud SQL for MySQL adds built‑in vector search that generates embeddings by calling external models (ChatGPT, Anthropic, Vertex AI/Gemini), stores them as native vector/blob types, and builds an on‑disk index using the SCAN index — a K-means partition tree with quantization — pinned to memory but evictable. Search uses SQL extensions like approximate distance and a pushdown/cached hybrid loop for relational filters. Two‑level trees hold ~10 million vectors; tests show billions per instance with ~95% recall, full MVCC and crash safety.
AI-written summary · about
chapters (9)
- 0:00Vector search in Cloud SQL
- 1:09Embedding lifecycle in OLTP
- 2:32Embeddings from SQL
- 6:01SCAN index for vectors
- 14:34Vector search in SQL
- 22:04Monitoring vector indexes
- 24:32Vector search features
- 25:57Resources to try
- 27:11Q&A on hybrid search
related talks
27mUse AI Vector Search to Build GenAI Apps with Enterprise Data
27mUnified AI & Relational Power: Scalable Distributed Vector Search with
57mEnter The Brave New World of GenAI with Vector Search
22mVector Search Made Simple: Getting Started With OpenSearch for AI Applications
27mMachine Learning for Advanced AI Search
47mBuilding intelligent analytics at scale
30mVector Search on the Edge Powering Mobile and IoT with AI
30mVector Search Made Simple: Getting Started with OpenSearch for AI Applications