zig-redis-server

Zig Redis Server

A Redis-compatible server implementation in Zig using the RESP protocol.

This implementation has been featured in the official solutions list for the "Build Your Own Redis" coding challenge.

Features

• RESP Protocol Implementation

  • Supports Simple String, Bulk String, Integer, Error and Nil types
  • Full serialization/deserialization

• Core Redis Commands

  • PING - Server test command
  • ECHO - Echo back input
  • GET/SET - Basic key-value operations

• Architecture

  • Non-blocking I/O using Facil.io
  • Thread pool for concurrent request handling
  • In-memory key-value store with read-write locking
  • Arena allocator for efficient memory management

Usage

Build from source

# Build an optimized binary (ReleaseFast mode)
make

# Start the server (default port: 6377)
./zig-out/bin/redis-server

Connect with redis-cli

# Basic commands
redis-cli -p 6377 PING
> PONG

redis-cli -p 6377 ECHO "Hello World"
> "Hello World"

redis-cli -p 6377 SET mykey "Hello"
> OK

redis-cli -p 6377 GET mykey
> "Hello"

Optimizations

Memory I/O Optimisation

Before Optimisation

Moved from RESP parsing and connection read and write operations using Heap allocated buffer to Fixed Buffer allocator to allocate on Stack,

❯ redis-benchmark -p 6377 -t SET,GET -q
WARNING: Could not fetch server CONFIG
SET: 17513.13 requests per second, p50=2.647 msec
GET: 22867.60 requests per second, p50=2.007 msec

After Optimisation

fn on_data(uuid: isize, _: *fio.fio_protocol_s) callconv(.C) void {
    var buf: [4096]u8 = undefined;
    var fba = std.heap.FixedBufferAllocator.init(&buf);
    const allocator = fba.allocator();
    // rest of the code...

❯ redis-benchmark -p 6377 -t SET,GET -q
WARNING: Could not fetch server CONFIG
SET: 107526.88 requests per second, p50=0.239 msec
GET: 90171.33 requests per second, p50=0.303 msec

Lock Free Single Threaded

Moved from Multi Threaded Non Blocking Evented I/O with Locks on global dictionary to Lock Free Single Threaded design

Before Optimisation

❯ redis-benchmark -p 6377 -t SET,GET -q
WARNING: Could not fetch server CONFIG
SET: 107526.88 requests per second, p50=0.239 msec
GET: 90171.33 requests per second, p50=0.303 msec

After Optimisation

❯ redis-benchmark -p 6377 -t SET,GET -q
WARNING: Could not fetch server CONFIG
SET: 140449.44 requests per second, p50=0.183 msec
GET: 142450.14 requests per second, p50=0.183 msec

Performance Benchmarks

Benchmark Comparison: Zig Redis Server vs Original Redis

This comparison measures the performance of this Zig Redis implementation against the official Redis server using redis-benchmark.

Latest Benchmark Results

Server	Command	Requests/sec	Latency (p50)
Zig Redis (6377)	SET	140,449.44	0.183 ms
Original Redis (6379)	SET	146,198.83	0.175 ms
Zig Redis (6377)	GET	142,450.14	0.183 ms
Original Redis (6379)	GET	140,646.97	0.183 ms

Performance Evolution

This Zig implementation has shown significant improvements through optimization:

Initial Implementation

SET: 17,513.13 requests per second, p50=2.647 msec
GET: 22,867.60 requests per second, p50=2.007 msec

After Stack Allocation Optimization

SET: 107,526.88 requests per second, p50=0.239 msec
GET: 90,171.33 requests per second, p50=0.303 msec

Current Implementation

SET: 140,449.44 requests per second, p50=0.183 msec
GET: 142,450.14 requests per second, p50=0.183 msec

This represents an 8x improvement for SET operations and a 6.2x improvement for GET operations from the initial implementation.

Analysis

• The Zig implementation achieves approximately 96% of Redis's performance for SET operations
• For GET operations, the Zig implementation slightly outperforms Redis by about 1.3%
• The current implementation is effectively at performance parity with the official Redis server

These benchmarks were conducted on the same hardware using identical benchmark parameters.

Development

# Build for development,
# then run the server from the executable directory
zig build -freference-trace
./zig-out/bin/redis-server

# Build and run for development
zig build run

License

MIT