Interesting. I've been on/off working on something, though it's an append-only hash index thing (based on Erlang's Bitcask): https://git.sr.ht/~tombert/feocask

I haven't done any benchmarks on it yet so I have no idea how well it performs, but I have a very strict "no explicit lock" policy, so everything is handled with separate files with a reconciliation service that uses a vector clock to determine ordering.

NOTE: A lot of this is AI-assisted, treat what I have written as more of a whiteboard proof of concept. I'm in the process of rewriting it by hand to do some more aggressive optimizations.