A Redis-compatible database server in Rust that runs commands on different keys in parallel and builds in JSON, time-series, Bloom filters, schemas and SQL queries.
Redis is the default in-memory store, but it executes commands on a single thread, and JSON, time-series and probabilistic types need separate modules. The goal was a wire-compatible alternative in Rust that uses every core and ships those types built in.
A Rust server speaks the Redis protocol, runs a task per connection and keeps every value in one sharded map, so commands on different keys run in parallel. Writes go to a background log writer, so a slow disk never blocks a client.
A hand-written RESP parser handles partial and pipelined frames, so redis-cli, redis-benchmark and standard client libraries work unchanged.
Set, get and update parts of JSON documents by path, including array appends and numeric increments.
Time series with retention, labels and range queries, and Bloom filters for membership checks, with no modules to install.
Define a schema, and typed writes are rejected when the value does not match it.
Select, count and delete keys by name pattern, type, memory or TTL with a SQL statement.
A Prometheus endpoint with 13 metric families, a health endpoint, a slow log for commands over 1 ms and hot-key tracking.
Developers start one binary and point any Redis client at it. Familiar commands work as before, and the same connection can store JSON documents, time series and Bloom filters, validate values against schemas, and query the keyspace with SQL. Prometheus metrics, a slow log and hot-key tracking are built in.
Architecture · Architecture
One write, end to end · Sequence
Whiteboard on the database · Sequence
Infrastructure · Infrastructure
Data model · Data Model
Authentication · Auth Flow
Chose DashMap plus a Tokio task per connection over The single-threaded, lock-free event loop Redis uses
Commands on different keys run in parallel. The project's benchmark shows the gain on non-pipelined commands and the cost under heavy pipelining, where per-command locking dominates.
Chose A bounded channel feeding a background append-only-file writer with a 64 KB buffer over Writing and syncing inside the command path
A slow disk never blocks a client. The trade-off is written in the code: when the channel is full, entries are dropped in favour of speed.
Chose A RESP2 parser instead of a custom protocol over A new protocol with its own client library
Every demo uses a stock Redis client, and benchmarking uses Redis's own tool, so there is nothing new for developers to install.
Chose JSON, time-series, Bloom filters and schemas as variants of one value type over A plug-in module system
One binary, one memory estimate per value, and one keyspace that the SQL query command can scan by type.
Chose Socket.IO broadcasting in the demo, with state stored in the database over Database-level publish and subscribe
The server has no pub/sub or transactions yet, so the demo apps handle broadcasting and order serialisation themselves.
Phase 1
Phase 2
Phase 3
Phase 4
103
Commands routed
Redis-style commands plus JSON, time-series, Bloom, schema and SQL commands, counted in the command router.
+42.8%
Non-pipelined INCR vs Redis
Single machine, redis-benchmark, as reported in the project paper. Pipelined writes are slower than Redis.
78
Automated tests passing
32 library and 46 command tests, run for this case study. 7 integration tests need a live server.
Tell us what you're building and we'll show you how we'd approach it.
Consumer apps and creator economyA studio promised its creator client an AI social app in a month with one developer. We scoped the real need and had four roles working eight days after the first call.
Gaming and PC optimizationA full-stack, design-first e-commerce build for a PC optimization brand, with cart, checkout and key delivery, that reached a client-reported 25% CTR in a low-conversion niche.
Gaming and PC optimizationA Discord bot that pulls the best gaming deals from Amazon through its Product Advertising API and posts them to the community every four hours.