MeshXFS: AI Adapts XFS Into A Clustered File-System

MeshXFS, a clustered filesystem written by artificial intelligence, has emerged as a groundbreaking development in the world of open-source software. Born from an experiment, MeshXFS now stands as a fully functional, released, and GPL-2.0 licensed file system that caters to the needs of multiple nodes working in unison.

Adapting the upstream XFS code, AI took the helm in crafting key features that empower MeshXFS to excel. Two distributed lock managers were created, one relying on the network and the other utilizing SCSI COMPARE AND WRITE to manage lock state on the shared disk without a lock network. Cache coherency across nodes, fencing with SCSI persistent reservations, and survivor-led journal recovery were also part of AI's contributions.

As the AI took charge, it managed the on-disk envelope surrounding XFS, along with userspace tools like mkfs, fsck, and resize. Every modification to the forked XFS code since the fork saw the AI's touch, resulting in an XFS codebase that grew from approximately 226k lines to around 314k. Additionally, the AI designed the test harnesses, created the test rig automation, and was instrumental in release validation.

Claude Opus, running as an autonomous coding agent within Claude Code, played a significant role in writing a large portion of the code. The AI consulted various language models such as Claude Fable, GPT, Gemini, and Astra to tackle the most complex problems, including lock protocol design, recovery races, and fencing correctness.

While the AI took the lead in coding, the author of this story, Michael Larabel, provided the necessary guidance and ensured data integrity, prevention of host crashes, and adherence to time constraints. The goal was never to crash the host or exceed twice the time native XFS would take on the same storage.

Michael Larabel, the principal author of Phoronix.com and the founder of the site in 2004, has been a prominent figure in covering Linux hardware support, performance, graphics drivers, and other topics. With over 20,000 articles to his name, Michael is also the lead developer behind the Phoronix Test Suite, Phoromatic, and OpenBenchmarking.org automated benchmarking software.