no ai. just docs, articles, code, and time.

RedrumDB is an in-memory datastore implemented in Rust.

The project focuses on building core components from scratch to better understand the design and performance characteristics of systems like Redis.

Features

• key-value store (strings, lists, streams)
• RESP protocol support
• pub/sub
• blocking operations (BLPOP)
• key expiries
• non-blocking I/O

Architecture

The server runs on a single-threaded, event-driven loop using mio, with slab-based connection management and non-blocking I/O.

Requests are parsed using a low-copy RESP parser, decoding command names and arguments directly from the connection buffer before dispatch.

Core components:

• event loop + connections

  • mio-based loop with explicit readable/writable interest switching
  • slab + token model for stable connection indexing
  • backpressure-aware writes with per-connection output buffers

• command execution

  • command dispatch table for normalized O(1)-style handler lookup
  • centralized handling of pub/sub mode constraints

• memory + data model

  • in-memory keyspace with typed values (string, list, stream)
  • shared key storage using Arc<[u8]> to reduce duplication

• scheduling + time

  • expiry engine using a min-heap (BinaryHeap<Reverse<Instant>>)
  • incremental cleanup bounded per loop to avoid latency spikes

• blocking + async behavior

  • blocking list operations (BLPOP) with per-key wait queues
  • timeout scheduling integrated into the main event loop

• streams

  • custom listpack implementation for compact storage
  • radix tree with prefix compression for ordered indexing
  • Redis-like stream ID semantics (*, ms-*, explicit IDs)

Design prioritizes predictable latency, minimal allocations, and explicit control over execution.

running

git clone https://github.com/thedevyashsaini/redrumdb
cd redrumdb
cargo run

Connect using:

redis-cli -p 6379