Hashiverse uses fjall for its distributed post metadata subsystem at scale!

[jimmejardine]

Hashiverse is an open-source decentralized Twitter replacement running on thousands of volunteer nodes. On the servers, while the post themselves are persisted ecrypted to disk, their metadata lives in fjall for rapid, repeat access.

It was picked over the alternatives for a handful of reasons that matter to a hashiverse server: it's actively maintained (sled stalled before its v1.0 rewrite shipped, and redb's single-writer model serialises everything behind its compactor); it auto-compacts in the background without blocking writes, where redb compacts synchronously and stalls the write path while it runs; it doesn't need free disk space equal to the live dataset to compact, so a server running close to its disk budget stays healthy; and being pure Rust it drops straight into the workspace with no C/C++ build dependency, no FFI surface, and no surprises in our cross-platform CI. Under our load tests it sustains the write rates we throw at it (bundle-metadata inserts, feedback batches, range scans during decimation) without latency spikes, which is the bar we actually care about.

Epic work, Marvin and contributors!