SDKs

The dvarapala engine in process (RFC 0008): the gateway's per-client rate limit, flood detection and connection limits for services that cannot sit behind the proxy, plus routes, manual blocks, egress shaping and reload (RFC 0028). Each package is a thin layer over the C ABI in crates/ffi (crates/ffi/dvarapala.h). Settings take the CONFIG.md names (rate_limit.*, detection.*, trusted_proxies, the connection keys of [l4], routes and forward.destinations as JSON arrays, admin.max_blocks, admin.max_block_ttl) with the gateway's defaults. One guard per process: behind several workers, divide the rates by their number.

On macOS the shared libraries end in .dylib. Install from the registries (RFC 0026, RFC 0027), with the library prebuilt for Linux (glibc 2.28 or later) and macOS on x86_64 and arm64:

go get github.com/franzramadhan/dvarapala-go@v<version>
pip install dvarapala
npm install dvarapala

The Go module is generated from sdk/go by scripts/go-module.sh; the wheels by scripts/wheel.py; the npm addon ships in the platform packages under js/npm/. scripts/sdk-packages-check.sh builds and installs all three for the host platform. Releases also attach the libraries as dvarapala-sdk-<tag>-<target>.tar.gz.

g, err := dvarapala.New(dvarapala.Config{"rate_limit.rate": "20"})
http.ListenAndServe(":8080", g.Middleware(mux))

guard = Guard({"rate_limit.rate": 20, "detection.mode": "enforce"})
app = ASGIMiddleware(app, guard)

const guard = new Guard({ 'rate_limit.rate': 20 });
const limit = guard.middleware();
guard.attach(http.createServer((req, res) => limit(req, res, () => res.end('ok'))));

ASGI and WSGI expose no accepted connections, so in Python the [l4] limits need the server's own hook: guard.conn_open(ip) on accept, guard.conn_close(handle) on close.