When a filesystem like bcachefs steps out of its experimental phase, it’s more than a version number shift. Release 1.38.6, dubbed by Kent Overstreet as "the performance release," marks a status change for a project that has weathered years of development and some notable turbulence within the kernel community. Overstreet used his Patreon blog for the belated yet official announcement: "Consider this the belated official announcement :)".

More devices, fewer bugs, and numbers that matter

The changelog isn't just noise. This version bumps the maximum number of devices per filesystem to 255, fixes half a dozen bugs, and introduces six performance optimizations. The Reconcile operation—previously called rebalance—is now faster and more parallel. But the numbers on real hardware give substance to the maturity leap.

On an AMD EPYC 9454 with 48 Zen4 cores, dbench throughput with 48 clients hit 16.5 GB/s, compared to 16 GB/s for XFS. In 4k random write tests with fio, bcachefs reached 700,000 IOPS, while XFS topped one million. Overstreet acknowledges that bcachefs is doing much more work here—erasure coding, checksums, and other data integrity features. Reporting lower numbers under a heavier workload is a sign of intellectual honesty and a project that isn’t chasing cheap comparisons.

The Rust bet and AI skepticism

One detail that will please software security advocates: Overstreet has already converted the userspace code to Rust. The next release will bring the safe interfaces into the DKMS module, with a target of roughly 50% Rust by year’s end. "Converting the journal to safe Rust will be… interesting," he notes, but the path is set. This intersects with the growing adoption of Rust in the Linux kernel, and bcachefs could become the first mainstream memory-safe filesystem.

Overstreet’s skepticism toward AI-generated code is another notable stance. He expressed irritation at lazy patches produced by LLMs without human review—a position that resonates for those managing critical infrastructure. Knowing that the maintainer guards against low-quality automated contributions signals seriousness.

Why it matters for on-premise (and AI)

Bcachefs is a copy-on-write filesystem with snapshots, compression, erasure coding, and multi-tier caching: features that align well with large-scale on-premise workloads, such as training datasets, checkpoint storage, and data pipelines. NASty, an experimental NAS OS built around bcachefs, offers a low-friction way to test it without manual compilation. For those evaluating self-hosted solutions for growing data volumes—perhaps in air-gapped or sovereignty-sensitive environments—a mature filesystem with advanced integrity checks is a valuable weapon.

Of course, bcachefs’s journey hasn’t been smooth: the personal clash between Overstreet and Linus Torvalds led to the filesystem’s removal from the mainline kernel. That remains a risk factor for production adoption. But the technical direction—improving performance, data integrity, and now Rust—sends a clear message: bcachefs is alive, and for anyone wanting to test it on their own hardware, the time has never been more mature.