ArchBench

ArchBench is a lightweight benchmark and test orchestration tool for native cross-architecture workflows. It is aimed at developers who work on ARM64 laptops but need reproducible results from AMD64 Linux machines, homelab boxes, or other SSH-accessible benchmark hosts.

• local, SSH, Docker, and GitHub Actions targets
• bench and test modes
• Go output normalization through the built-in go-test parser
• JSON result artifacts
• terminal and Markdown reports
• local and remote build-cache wiring through $ARCHBENCH_CACHE
• GitHub Actions workflow generation with archbench generate

Install

Homebrew (macOS)

brew install --cask sirrobot01/tap/archbench

ArchBench ships as a Homebrew cask, which is macOS-only. On Linux, use Go or a prebuilt binary below.

Go

go install github.com/sirrobot01/archbench@latest

Prebuilt binaries

Download a tar.gz for your OS/arch from the releases page.

From source

go build -o archbench .
./archbench version

Note: Windows is intentionally excluded until the runner layer supports non-POSIX process handling.

Quick Start

Create a spec:

archbench init

Run all configured targets:

archbench run

Run one target:

archbench run --target local

Run multiple targets concurrently:

archbench run --concurrency 2

The default concurrency is 1, so runs remain sequential unless you opt in.

Render saved results:

archbench report --format md

Compare two result artifacts:

archbench compare baseline.json candidate.json

Add --threshold to fail (non-zero exit) when any benchmark's ns/op regresses past that percentage — useful as a CI gate:

archbench compare baseline.json candidate.json --threshold 50

Examples

The examples/ directory has self-contained suites you can run locally — a basic benchmark, one with meaningful memory metrics, and a test-mode suite. Each is its own module with its own spec:

archbench run \
  --spec examples/go-basic/archbench.yaml \
  --dir examples/go-basic \
  --target local \
  --no-cache

Spec Shape

name: my-suite
mode: bench

targets:
  - name: local
    type: local

  - name: amd64-box
    type: ssh
    host: bench-box

  - name: amd64-container
    type: docker
    image: golang:1.26
    platform: linux/amd64   # pin a non-native arch via emulation

  - name: ci
    type: github-actions
    runsOn: ubuntu-latest

runs:
  - name: parser
    command: go test ./internal/parser/... -run '^$' -bench=. -benchmem -count=10

  - name: stream
    command: go test ./pkg/stream/... -run '^$' -bench='BenchmarkRead|BenchmarkWrite' -benchmem -count=10

parser: go-test

Each selected target executes every entry in runs in order, writing one archbench-results/<target>.json artifact with a runs array so reports and comparisons keep benchmark groups separate. Per-target setup/env, exec mode, Docker, GitHub Actions, caching, and test mode are covered in the docs below.

Documentation

• Getting Started — full spec reference: local, SSH, Docker, and GitHub Actions targets, setup/env, exec mode, PATH setup, caching, and test mode.
• Security Model — SSH and Docker trust, project sync, and how environment secrets are handled.