Date: 2025-05-14
Time: 14:00–14:50
Room: Breakout Room 3
Level: Intermediate
There are many impressive automated tools for optimizing PostgreSQL deployments (e.g., knob tuning, picking indexes). These existing tools specialize in optimizing one aspect of a database and do not look at all aspects of a system holistically. The inability to tune everything in a database leads to sub-optimal performance and efficiency. But breakthroughs in machine learning allow for semantic reasoning (think LLMs) about a database, making it possible for a single tool to consider all the facets of optimizing Postgres simultaneously.
This talk presents our research on a new generation of autonomous tuning agents for PostgreSQL that optimize more parts of a database in less time and at lower costs than previous methods. We will describe our method for tuning every possible option available in PostgreSQL: system knobs, table knobs, index knobs, index selection, query plan hints, and more. As part of this talk, we will also describe the changes we made to PostgreSQL to facilitate control by autonomous agents better.