Breaking Database Lock-in: Agentic Regeneration of High Performance Storage Readers for Database Bypass

Jul 16, 2026

Analytical workloads accessing data through external database systems face a core bottleneck: JDBC/ODBC and driver layers force reads to pass through query execution and non-bulk parsing, limiting latency and throughput for downstream analytic engines. Jailbreak bypasses the database engine by directly reading storage files and regenerating operator-specific table readers. It uses LLM-assisted code synthesis from database format source code and documentation to decode storage into Apache Arrow in-memory columnar buffers. Evaluations on PostgreSQL and MySQL storage files with TPC-H validate correctness and show up to 27× speedups in end-to-end analytical throughput, supporting a generalizable approach to break database lock-in.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Abstract

Introduction

Bottleneck of engine-mediated access

Key observation: formats are specified

Motivation and barrier to custom readers

Breaking Database Lock-in: Agentic Regeneration of High Performance Storage Readers for Database Bypass Victor Giannakouris Cornell University Ithaca, NY, USA [vg292@cornell.edu](vg292@cornell.edu) Immanuel Trummer Cornell University Ithaca, NY, USA [itrummer@cornell.edu](itrummer@cornell.edu) arXiv :2607 .07696v 1 [ cs .DB] 8 Jul 2026 ABSTRACT Analytical workloads operating on data stored in external database systems face a fundamental bottleneck: data access is guarded entirely by the database driver, like JDBC or ODBC, forcing all reads through query execution and other driver layers that are not designed for bulk columnar analytics. Existing approaches, like JDBC and ODBC connectors, inherit this overhead, resulting in significant latency and throughput limitations for downstream external, analytic query engines. We present Jailbreak, an approach that bypasses the database engine entirely by reading storage files directly and materializing data as in-memory columnar buffers. Jailbreak’s key insight is that database file formats, while complex, are fully specified by their source code and documentation, artifacts that Large Language Models (LLMs) can ingest to regenerate operator-specific table reading components without human-engineered parsing logic. Jailbreak leverages LLM-assisted code synthesis for database storage decoding, turning a traditionally opaque format into a directly queryable artifact. We evaluate Jailbreak on PostgreSQL and MySQL storage files, targeting analytical snapshot scenarios common in read replicas and offline processing pipelines. The generated reader produces Apache Arrow buffers consumable directly by most of the widely known query engines, including DuckDB, Apache Spark, and GPU-accelerated frameworks such as cuDF and Spark RAPIDS. We validate correctness against JDBC/ODBC-based baselines using the TPC-H benchmark across all query results, and demonstrate significant performance improvements in end-to-end analytical throughput, achieving up to 27× speedups. Our results suggest that LLM-assisted storage reader synthesis is a viable and generalizable methodology for breaking data lock-in across database systems, with applications beyond PostgreSQL and MySQL for any system whose file format is available to the LLM from documentation or source code. VLDB Workshop Reference Format: Victor Giannakouris and Immanuel Trummer. Breaking Database Lock-in: Agentic Regeneration of High Performance Storage Readers for Database Bypass. VLDB 2026 Workshop: AIDB. VLDB Workshop Artifact Availability: The source code, data, and/or other artifacts have been made available at [https://github.com/gsvic/Jailbreak](https://github.com/gsvic/Jailbreak). This work is licensed under the Creative Commons BY-NC-ND 4.0 International License. Visit [https://creativecommons.org/licenses/by-nc-nd/4.0/ to view a copy of](https://creativecommons.org/licenses/by-nc-nd/4.0/ to view a copy of)[ ](https://creativecommons.org/licenses/by-nc-nd/4.0/ to view a copy of)[this license. For any use beyond those covered by this license](this license. For any use beyond those covered by this license), [obtain permission by](obtain permission by)[emailing info@vldb.org. Copyright](emailing info@vldb.org. Copyright) is held by the owner/author(s). Publication rights licensed to the VLDB Endowment. Proceedings of the VLDB Endowment. ISSN 2150-8097 . 1 INTRODUCTION The boundary between relational database systems (RDBMSes) and analytical query engines has long been guarded by the RDBMS engine itself. When an external system or application, like a GPUaccelerated framework, a vectorized query engine, or an ML feature store, needs data that resides in PostgreSQL or MySQL, it must issue SQL queries via a wire protocol, like JDBC or ODBC, wait for the server to traverse its storage layer, serialize rows through the wire protocol, and then deserialize those rows on the client side before any useful computation can begin. Th

FAQs

What problem does Jailbreak address in analytical workloads?

Jailbreak targets the latency and throughput limits caused by engine-mediated access paths where JDBC/ODBC and driver layers guard all reads and serialize/deserialize data before analytics can start.

How does Jailbreak bypass the database engine?

It directly reads database storage files and materializes data as in-memory columnar buffers, avoiding SQL wire-protocol and database-server traversal of storage for each request.

How are storage readers generated and validated for correctness and performance?

Jailbreak uses LLM-assisted code synthesis based on the database format’s source code and documentation to regenerate operator-specific table reading components, then validates results against JDBC/ODBC baselines using TPC-H across all query outputs while measuring end-to-end analytical throughput improvements.

Jul 16, 2026

Upload by Clementine