Skip to content
<open-source>
Oeiuwq Faith Blog OpenSource Porfolio

kalbasit/ncps

Nix binary cache proxy service -- with local caching and signing.
Source code at GitHub https://ncps.dev
kalbasit/ncps.json
{
  "createdAt": "2024-11-28T06:11:23Z",
  "defaultBranch": "main",
  "description": "Nix binary cache proxy service -- with local caching and signing.",
  "fullName": "kalbasit/ncps",
  "homepage": "https://ncps.dev",
  "language": "Go",
  "name": "ncps",
  "pushedAt": "2026-07-12T12:09:01Z",
  "stargazersCount": 320,
  "topics": [],
  "updatedAt": "2026-07-12T02:39:57Z",
  "url": "https://github.com/kalbasit/ncps"
}

ncps logo

ncps: Nix Cache Proxy Server

Section titled “ncps: Nix Cache Proxy Server”

A high-performance proxy server that accelerates Nix dependency retrieval across your local network

Go Report Card License: MIT Sponsor

What is ncps?

Section titled “What is ncps?”

ncps acts as a local binary cache for Nix, fetching store paths from upstream caches (like cache.nixos.org) and storing them locally. This reduces download times and bandwidth usage, especially beneficial when multiple machines share the same dependencies.

Key Features

Section titled “Key Features”
  • Multi-upstream cache with automatic failover
  • Flexible storage: local filesystem or S3-compatible (AWS S3, Garage, etc.)
  • Database support: SQLite, PostgreSQL, or MySQL/MariaDB
  • High availability with Redis distributed locking for zero-downtime deployments
  • Smart caching: LRU management with configurable size limits
  • Secure signing: Signs cached paths with private keys for integrity
  • Observability: OpenTelemetry and Prometheus metrics support
  • Easy setup: Simple configuration and deployment

⚠️ Development Status & Data Safety

Section titled “⚠️ Development Status & Data Safety”

[!important] Production Warning: The main branch and pre-release versions are under active development and should never be used in production. Your data may be lost or corrupted! Always use the latest release for production environments.

Early Stage Notice: ncps is in early development and data consistency/recovery is not guaranteed. Please maintain regular backups of your data, especially when updating ncps versions.

Quick Start

Section titled “Quick Start”

Get ncps running in minutes with Docker:

Terminal window
# Pull images and create storage
docker pull alpine && docker pull ghcr.io/kalbasit/ncps
docker volume create ncps-storage
docker run --rm -v ncps-storage:/storage alpine /bin/sh -c \
  "mkdir -m 0755 -p /storage/var && mkdir -m 0700 -p /storage/var/ncps && mkdir -m 0700 -p /storage/var/ncps/db"


# Initialize database (ncps embeds the migrations and applies them via goose)
docker run --rm -v ncps-storage:/storage ghcr.io/kalbasit/ncps \
  /bin/ncps migrate up --cache-database-url=sqlite:/storage/var/ncps/db/db.sqlite


# Start the server
docker run -d --name ncps -p 8501:8501 -v ncps-storage:/storage ghcr.io/kalbasit/ncps \
  /bin/ncps serve \
  --cache-hostname=your-ncps-hostname \
  --cache-storage-local=/storage \
  --cache-database-url=sqlite:/storage/var/ncps/db/db.sqlite \
  --cache-upstream-url=https://cache.nixos.org \
  --cache-upstream-public-key=cache.nixos.org-1:6NCHdD59X431o0gWypbMrAURkbJ16ZPMQFGspcDShjY=

Your cache will be available at http://localhost:8501. See the Quick Start Guide for more options including S3 storage.

Documentation

Section titled “Documentation”
  • Getting Started - Quick start guide, core concepts, and architecture
  • Installation - Docker, Docker Compose, Kubernetes, Helm, NixOS
  • Configuration - Complete configuration reference, storage and database options
  • Deployment - Single-instance and high-availability deployment guides
  • Usage - Client setup and cache management
  • Operations - Monitoring, troubleshooting, backup and upgrades
  • Architecture - System architecture and design details
  • Development - Contributing, development setup, and testing

Installation Methods

Section titled “Installation Methods”
MethodBest ForDocumentation
DockerQuick setup, single-instanceDocker Guide
Docker ComposeAutomated setup with dependenciesDocker Compose Guide
KubernetesProduction, manual K8s deploymentKubernetes Guide
Helm ChartProduction, simplified K8s managementHelm Guide
NixOSNixOS systems with native integrationNixOS Guide

Deployment Modes

Section titled “Deployment Modes”
  • Single-instance: Simple deployment with local or S3 storage, SQLite or shared database
  • High Availability: Multiple instances with S3 storage, PostgreSQL/MySQL, and Redis for zero-downtime operation. Note: Must enable CDC.

See the Deployment Guide for detailed setup instructions.

Choosing a storage backend

Section titled “Choosing a storage backend”

The local filesystem backend assumes single-writer POSIX semantics: one process owning the directory, with read-after-write consistency. Use it for single-instance deployments.

For multiple replicas, use the S3-compatible backend. Do not point the local backend at a shared network filesystem (NFS/SMB) written by more than one replica: network filesystems provide only close-to-open consistency with client-side attribute caching, so one replica can read stale metadata for a file another replica just wrote — which presents to ncps as a NAR that is in the database but “missing” from storage, and to clients as spurious cache misses or truncated transfers. An object store gives every replica a single, read-after-write-consistent view.

Storage latency also drives the CDC decision (see cache.cdc in [config.example.yaml]!(config.example.yaml)): CDC’s many small chunk reads suit low-latency backends, while high-latency or spinning-disk storage is better served whole-file with CDC disabled. To move an existing deployment between the two, ncps migrate-nar-to-chunks chunks whole files and ncps migrate-chunks-to-nar reconstructs whole files from chunks (verified against each narinfo’s NarHash) so you can safely exit CDC.

Architecture

Section titled “Architecture”

ncps is written in Go. The notable internal dependencies:

  • Ent — type-safe ORM. Schemas under ent/schema/ are the only source of truth for the database DDL; the generated Ent client lives under ent/ (committed; produced by go generate ./ent/...).
  • Atlas — schema-diff engine, used as a Go library (no atlas CLI binary). cmd/generate-migrations diffs the Ent schemas against the on-disk migration history and emits one Goose-formatted .sql per dialect under migrations/<dialect>/, sealed by per-dialect atlas.sum integrity files.
  • Goose — runtime migration runner. ncps migrate up embeds the per-dialect migrations and applies pending ones. Migrations are forward-only; column changes follow an expand-contract policy.
  • Chi — HTTP router for the cache server.
  • Go-based storage backends — local filesystem and S3 (MinIO-compatible).

See Architecture for the full design and Development for the contributor workflow.

Support the Project

Section titled “Support the Project”

If you find ncps useful, please consider supporting its development! Sponsoring helps maintain the project, fund new features, and ensure long-term sustainability.

Sponsor this project

Contributing

Section titled “Contributing”

Contributions are welcome! We appreciate bug reports, feature requests, documentation improvements, and code contributions.

See the Developer Guide for:

  • Development setup and workflow
  • Code quality standards and testing procedures
  • How to submit pull requests

License

Section titled “License”

This project is licensed under the MIT License - see the [LICENSE]!(LICENSE) file for details.

Report Bug • Request Feature • Discussions • Sponsor