Croit rewrites Ceph's storage daemon in Rust and lets old and new servers share the same cluster
The first parts of ceph-rust have been released as a preview; in its own tests croit reaches write speeds on NVMe up to more than twice as fast.
Ceph specialist croit has released the first components of its ceph-rust project as a public preview. They consist of FlashStore OSD, a storage daemon for NVMe drives written from scratch in Rust without reusing Ceph's C++ code, and Rust libraries for the RADOS object store and for RBD block devices, heise reports. According to croit itself, the bottleneck for Ceph on flash storage is no longer the drive but the processor time needed per operation.
The new daemon speaks the same network protocol as the classic Ceph components. As a result, old and new storage daemons can run in the same pool and administrators can switch over gradually, croit writes on its blog. Monitors, managers and command-line tools remain unchanged. In its own tests, croit achieved 17.9 instead of 8.5 gigabytes per second for RBD writes with 4 MiB blocks, although heise stresses that such figures do not apply to every workload.
It is not yet a complete rewrite of Ceph: CephFS, deduplication and an S3 gateway in Rust are still missing. Croit advises against using the preview for production data and, according to heise, is aiming for general availability in the first quarter of 2027. The daemon is licensed under LGPL-3.0, the libraries under MPL-2.0.