Arc Gateway

Warning

Documentation is still in progress and test coverage is still expanding. Please be patient while updates are being finalized.

A high-performance HTTP reverse proxy built on Linux io_uring, designed for modern cloud-native environments.

Arc delivers 2x the throughput of Nginx in HTTP/1.1 proxy scenarios while maintaining zero-error guarantees under load. It combines a thread-per-core architecture with lock-free algorithms to minimize latency and maximize resource efficiency.

Performance

Benchmarked on WSL2 Debian (2 vCPU):

Scenario	Arc	Nginx	Ratio
HTTP/1.1 proxy (c=256)	102,203 RPS	49,930 RPS	2.05x
HTTP/1.1 proxy (c=512)	96,143 RPS	41,740 RPS	2.30x
HTTP/2 high concurrency	100% success	91.5% success	-
Proxy latency overhead	+0.82ms	+1.35ms	39% lower

Quick Start

# Build (requires Linux with kernel >= 5.10)
cargo build -p arc-gateway --release

# Run with example config
./target/release/arc-gateway --config arc.yaml

Arc supports .yaml, .yml, .json, and .toml config files.

Minimal configuration (arc.yaml):

listen: "0.0.0.0:8080"
admin_listen: "127.0.0.1:9900"

upstreams:
  - name: backend
    addr: "127.0.0.1:3000"

routes:
  - path: /
    upstream: backend

Verify it works:

curl http://localhost:8080/
curl http://localhost:9900/healthz
curl http://localhost:9900/metrics

Features

• io_uring data plane - Zero-copy I/O with fixed buffers, multishot accept, kernel-bypass polling
• Thread-per-core - Each worker binds to a CPU core with independent resources, no cross-thread contention
• Radix tree routing - O(log n) path matching with prefix wildcards, named parameters, and glob patterns
• Lock-free rate limiting - GCRA algorithm with atomic CAS, no mutex in hot path
• WASM plugins - Wasmtime-based plugin system with instance pooling and epoch-based timeout isolation
• Hot reload - ArcSwap-based config updates without dropping connections
• TLS termination - Rustls with SNI-based certificate selection and ACME automation
• HTTP/2 support - Full h2 multiplexing with configurable stream limits
• Distributed rate limiting - Redis-backed global limits with circuit breaker fallback
• Custom error pages - Route-level and global error pages with template variables and upstream fallback
• Multi-format config - JSON, TOML, and YAML configuration files with hot reload
• Observability - Prometheus metrics, phase-level latency tracking, connection lifecycle counters
• Access logging - Node-local NDJSON logs with arc logs tail/query for local inspection

Documentation

• Getting Started - Installation, first config, verification
• Configuration Reference - Complete field reference with examples
• Architecture - Design decisions, module boundaries, io_uring strategy
• Components - Custom vs. borrowed code breakdown
• Security - Request smuggling, slowloris, HPACK bomb defenses
• Benchmarks - Methodology, environment setup, detailed results
• Migration from Nginx - Config translation guide

Requirements

• Linux kernel >= 5.10 (6.1+ recommended for best io_uring support)
• Rust 1.75+
• For distributed rate limiting: Redis 6+

License

Apache 2.0