Exploring the benefits of learned indexes in PostgreSQL: Theory and Implementation
summary
Built an extension that wraps the ALEX learned index and calls its C++ API via the PostgreSQL AM hooks, bulk‑loading tuples from heap into an in‑memory ALEX and measuring fallback ratio and gap density for planner costing. Experiments show ALEX’s in‑memory design clashes with PostgreSQL’s disk‑first durability and shared buffers so that serializing ALEX to disk is slower than B‑tree, making generic replacement infeasible; focus shifts to temporary in‑memory indexes for joins, ingest, FDWs and ML pipelines as plausible niches.
AI-written summary · about
chapters (6)
- 0:00Introduction and motivation
- 10:04What are learned indexes
- 21:24Learned index models Alex and RMI
- 31:58Integration challenges in PostgreSQL
- 40:52Cost modeling and statistics
- 47:52Conclusion and future work
related talks
30mLearned Indexes in PostgreSQL: Theory & Implementation
37mLearned Indexes: a New Idea for Efficient Data Access #bbuzz
45mA Deep Dive into PostgreSQL Indexing
37mHow Postgres could index itself
53mAdvanced Indexing Techniques in PostgreSQL
48mTricks from in-memory databases
01h 00mFlexible Indexing With Postgres
39mAn Introduction to B-trees and LSM-trees for DEVs and DevOps