Learned Indexes in PostgreSQL: Theory & Implementation
summary
Claims a learned-index prototype for PostgreSQL that trains models externally (using TensorFlow compiled to C++), then deploys them alongside B‑trees as failback. Recreated RMI/Alex-style multi‑stage models, benchmarks show much smaller storage footprints and up to 2–3× speedup on read-heavy, skewed datasets in prior papers, but the prototype highlights problems with updates, outliers, last‑mile accuracy and planning. Integration touches the index access method API, requires sampling, external training, runtime monitoring and periodic retraining rather than in‑server on‑the‑fly learning.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 5:00What are learned indexes
- 12:24Studies and models overview
- 17:43Why PostgreSQL for learned indexes
- 21:32Implementation approach
- 25:47Q&A and closing remarks
related talks
37mLearned Indexes: a New Idea for Efficient Data Access #bbuzz
50mExploring the benefits of learned indexes in PostgreSQL: Theory and Implementation
01h 00mPostgreSQL and Artificial intelligence
37mHow Postgres could index itself
53mAdvanced Indexing Techniques in PostgreSQL
48mB-tree Indexes in Databases Illustrated by IndexedDB and PostgreSQL
45mA Deep Dive into PostgreSQL Indexing
37mRound Table Discussion