Pure-Go Race Detector - Race Detection Without CGO

#go #concurrency #testing

Hi!
I've been working on a pure-Go race detector that works without CGO. Just released v0.3.0 with major performance improvements.

The Problem

Go's built-in race detector requires CGO. This means:

No race detection in Docker scratch images
Cross-compilation becomes complicated
Cloud functions and embedded systems are out of luck

The Solution

A pure-Go implementation of race detection using the FastTrack algorithm. Works with CGO_ENABLED=0.

What's New in v0.3.0

99% overhead reduction - from ~10,000x down to ~100x slowdown.

Key improvements:

Adaptive Sampling - configurable 1-100% sample rate for production use
Sparse VectorClocks - O(active goroutines) instead of O(total)
Address Compression - 8x memory reduction
4 Inline Slots - zero allocations for common cases

Installation

go install github.com/kolkov/racedetector/cmd/racedetector@v0.3.0

Usage

# Build with race detection
racedetector build -o myapp main.go

# Run with race detection
racedetector run main.go

# Production with 10% sampling
RACEDETECTOR_SAMPLE_RATE=10 racedetector run main.go

Looking for Feedback

I'd appreciate if you could try it on your projects and report any bugs or false positives. Contributions are welcome.

Also curious: should something like this be proposed for Go toolchain integration, or is it better as a standalone tool?

Thanks!

DEV Community