The paper “Is Your Learned Query Optimizer Behaving As You Expect: A Machine Learning Perspective” by Claude Lehmann, Pavel Sulimov and Kurt Stockinger has been accepted at VLDB 2024 (International Conference on Very Large Databases), which is considered among the most prestigious international database research conferences. The paper analyzes various learned query optimizers (LQOs), i.e. query optimizers that use machine learning approaches to improve or replace the classical optimizer in database engines. On top of it, the paper provides a framework for a fair evaluation of LQOs and advises on how to enhance the stability of LQOs via machine learning techniques. The paper shows that LQO evaluation is non-trivial, and even the best machine learning approaches currently do not systematically outperform the classical optimizer in PostgreSQL - one of the most commonly used open-source database systems.

This publication is the output of the GraphQueryML project, funded by the Swiss National Science Foundation (SNSF) and the Deutsche Forschungsgemeinschaft (DFG). A preprint of the paper can be found at https://arxiv.org/abs/2309.01551