Awesome SQL Query Rewrite

A curated list of papers, systems, benchmarks, and best practices for SQL query rewriting.

This repository focuses on semantically equivalent SQL transformations that improve execution efficiency and reliability, including rule-based rewriting, learned rewriting, and LLM-assisted rewriting.

Contents

• Papers
• Recommended Taxonomy
• Systems and Tools
• Benchmarks and Workloads
• Evaluation and Methodology
• Contribution Guide

Papers

2026

Year	Title	Venue	Paper	Code	Category	Notes
2026	Efficient Query Rewrite Rule Discovery via Standardized Enumeration and Learning-to-Rank (extend)	arXiv preprint (cs.DB)	arXiv	-	Automatic rule discovery	SLER combines template enumeration with learning-to-rank and reports a million-scale verified rule library.
2026	LASER: A Data-Centric Method for Low-Cost and Efficient SQL Rewriting based on SQL-GRPO	arXiv preprint (cs.DB)	arXiv	GitHub	LLM + RL rewriting	Introduces SQL-MCTS slow-query corpus and SQL-GRPO alignment for low-cost rewrite generation.
2026	Leveraging Query Optimizers to Verify the Soundness of LLM-based Query Rewrites for Real-World Workloads, and More!	CIDR 2026	PDF	-	Soundness and verification	Proposes QO-Verify to check rewrite soundness using optimizer internals on benchmark and real enterprise workloads.

2025

Year	Title	Venue	Paper	Code	Category	Notes
2025	GRewriter: Practical Query Rewriting with Automatic Rule Set Expansion in GaussDB	PVLDB 18(12):4991-5003	PDF / DOI	-	Production rule engine	A bolt-on extensible rewriter integrated into GaussDB with automated rule generation and production deployment results.
2025	QUITE: A Query Rewrite System Beyond Rules with LLM Agents	arXiv preprint (cs.DB/cs.AI)	arXiv	GitHub	Agentic LLM rewriting	Uses multi-agent LLM workflow to go beyond fixed rule sets and reduce regressions from rigid heuristics.
2025	Query Rewriting via LLMs	arXiv preprint (cs.DB)	arXiv	-	LLM-assisted rewriting	Studies prompt ensembles, database-sensitive hints, and token-probability guidance for performant and correct rewrites.

2024

Year	Title	Venue	Paper	Code	Category	Notes
2024	Learned Graph Rewriting with Equality Saturation: A New Paradigm in Relational Query Rewrite and Beyond	arXiv preprint (cs.DB/cs.LG)	arXiv	-	Learned graph rewriting	Combines equality saturation with graph RL for relational rewrite search.
2024	LLM-R2: A Large Language Model Enhanced Rule-based Rewrite System for Boosting Query Efficiency	arXiv preprint (cs.DB/cs.CL)	arXiv	GitHub	LLM-enhanced rule system	Uses LLMs to propose rewrite rules and contrastive selection to improve rule recommendation quality.

2023

Year	Title	Venue	Paper	Code	Category	Notes
2023	A Learned Query Rewrite System	PVLDB 16(12):4110-4113	PDF / DOI	GitHub	Learned rewrite system demo	Demonstrates a practical LearnedRewrite pipeline over Calcite with MCTS and hybrid estimation.
2023	QueryBooster: Improving SQL Performance Using Middleware Services for Human-Centered Query Rewriting	arXiv preprint (cs.DB)	arXiv	GitHub	Human-in-the-loop rewriting	Presents middleware-driven, user-oriented SQL rewrite assistance when DB/application internals are black-boxed.

2022

Year	Title	Venue	Paper	Code	Category	Notes
2022	A Learned Query Rewrite System using Monte Carlo Tree Search	PVLDB 15(1):46-58	PDF / DOI	GitHub	Learned search	Models rewrite orders as a policy tree and uses MCTS plus learned estimators to find high-benefit rewrite sequences.
2022	WeTune: Automatic Discovery and Verification of Query Rewrite Rules	SIGMOD 2022	DOI	-	Automatic rule discovery	Automatically discovers and verifies rewrite rules from real SQL workloads and known anti-patterns.

Recommended Taxonomy

To keep this repository future-proof, we recommend organizing query rewrite research using five categories instead of only "rule-based vs LLM":

1. Rule-based / Heuristic Rewriting: Manual rules and fixed exploration orders inside query optimizers.
2. Automatic Rule Discovery & Learned Search: Automatically mined rules and learned strategies for rewrite-order exploration.
3. LLM-Enhanced Rule Systems: LLMs assist with rule proposal, ranking, or selection while execution stays rule-engine driven.
4. LLM-Direct / Agentic Rewriting: LLMs or agents directly generate rewritten SQL with runtime feedback loops.
5. Soundness & Verification: Methods focused on semantic-equivalence checks, guardrails, and regression prevention.

Systems and Tools

• Apache Calcite - Foundational framework for SQL parsing, relational algebra, and rule-based optimization.
• SQLGlot - SQL parser/transpiler with optimizer utilities useful for rewrite experimentation.
• LearnedRewrite - Research implementation for learned SQL rewrite search.

Benchmarks and Workloads

• TPC-H - Widely used analytic benchmark for SQL performance studies.
• TPC-DS - Decision-support benchmark with complex SQL templates.
• Join Order Benchmark (JOB) - Realistic multi-join workload often used in optimizer and rewrite studies.
• SQL-MCTS (introduced in LASER) - A slow-query corpus designed for LLM-oriented rewriting research.

Evaluation and Methodology

Recommended minimum reporting dimensions for each paper/system:

• Semantic correctness checks: result-equivalence rate and failure taxonomy.
• Performance gains: latency speedup distribution (median, p90/p99) and tail regressions.
• Rewrite overhead: rewrite-time/optimization-time overhead and search cost.
• Generalization: cross-workload and cross-engine robustness.
• Safety: rate of harmful rewrites, rollback strategy, and verification coverage.
• Practicality: deployment constraints (privacy, token cost, integration complexity).

Contribution Guide

Please read CONTRIBUTING.md before submitting.

Quick rules:

• Keep entries concise and source-verifiable.
• Add year and venue for every paper.
• Prefer primary links (official venue or arXiv).
• Keep year sections in descending order and entries alphabetically sorted within each year.