Fluctlight is a playground for testing random ideas for speed improvements on a chat server with the Matrix protocol.

The goal is to implement a server with extreme performance (i.e. the Nginx of Matrix servers), using as many tricks as I can possibly think of, including:

• As few allocations/copying as possible
• Intern all state strings
• Group allocations in cache-friendly arenas
• Allow bits of JSON objects to be pre-serialized
• Keep all state in memory
• Allow room servers to be on separate hosts, with a router proxy in front

This server will only ever implement the federation APIs of the Matrix protocol, with no support for clients, and will never become a real server.

For a real server, use Conduit instead. Note that in the world of Matrix servers, performance is less important than memory usage, which is not the goal of Fluctlight, but is the goal of Conduit.

The server is currently not usable for anything.

Incoming APIs:

• Prototype /key/server and /key/query

Outgoing APIs:

• Prototype /make_join and /send_join

Currently implemented:

• All non-networking logic is bundled in a module, which is automatically reloaded at runtime whenever cargo builds a new library
• Requests use a per-request memory pool to store ephemeral strings and lists
• Deserialized requests and response structures use borrowed data wherever possible
• Requests use canned (pre-rendered) JSON snippets as part of the response
• Canonical JSON using a borrowing and sorting version of serde_json::Value
• Hashes and signatures computed without storing canonical JSONs, by piping via serde straight to a sha256 writer sink
• An admin HTML page to show information using compiled Askama templates

For planned features, see DESIGN.md.

The code is dual-licensed under either Apache-2.0 or MIT.