Skip to main content

Deploying

Kraken is a normal long running server: it opens a port and waits for connections. Anywhere you can run a Linux binary that listens on a port, you can run Kraken. This page shows how the live demo is hosted for free on AWS Lambda, and a few other options.

All hosting platforms tell your server which port to listen on with the PORT environment variable, so read it in main:

char *port_env = getenv("PORT");
int port = port_env ? atoi(port_env) : 8000;

http_server_t *server = http_server_init(port, BACKLOG);
caution

Kraken doesn't speak HTTPS. On every option below, TLS is handled by the platform or a proxy in front of Kraken, which then talks plain HTTP to it.

Serverless on AWS Lambda (what the demo uses)​

A small demo gets very little traffic, so paying for a server that runs all day is wasteful. Lambda only runs your code while a request is being handled and costs nothing while idle. The free tier (1 million requests a month) covers a personal project.

Lambda normally calls a function with an event object, not with an HTTP request on a socket. The Lambda Web Adapter, published by AWS, bridges the two. It runs next to your server, receives each Lambda event, sends it to your server as a plain HTTP request on 127.0.0.1:$PORT, and returns your response. Kraken runs without any changes.

browser → Lambda Function URL (HTTPS) → Web Adapter → 127.0.0.1:8000 → kraken accept()

The repository has scripts for this in deploy/lambda/:

deploy/lambda/build.sh   # builds kraken for Amazon Linux (arm64) inside Docker
deploy/lambda/deploy.sh  # creates or updates the function and prints its public URL

build.sh compiles Kraken inside an Amazon Linux 2023 container, so the binary matches the OS Lambda runs on, and zips it with the templates, static files and this bootstrap script:

#!/bin/sh
cd "$LAMBDA_TASK_ROOT" || exit 1
exec ./main

deploy.sh needs the AWS CLI with credentials, then:

  1. Creates an IAM role that only lets the function write logs.
  2. Creates the function on the provided.al2023 runtime (arm64, 128 MB) with the Web Adapter layer and PORT=8000.
  3. Sets AWS_LWA_READINESS_CHECK_PATH=/healthz. The adapter calls this path until the server answers, then starts sending traffic.
  4. Creates a public Function URL.

Running it again after a change rebuilds and updates the code.

Cold starts​

When nobody has visited for a few minutes, Lambda stops the process. The next request starts it again (a "cold start"). Kraken is a 200 KB binary with no runtime to load, so this takes well under a second. The live demo shows whether your request was a cold start, and how many requests that process has served since it started.

Custom domain with Cloudflare​

Function URLs look like https://<id>.lambda-url.<region>.on.aws. To serve the demo on kraken.rahulgpt.com, a small Cloudflare Worker forwards requests to the Function URL. A plain CNAME doesn't work, because Lambda rejects requests whose Host header isn't its own URL.

export default {
  async fetch(request, env) {
    const url = new URL(request.url);
    const upstream = new URL(url.pathname + url.search, env.KRAKEN_ORIGIN);
    return fetch(new Request(upstream, request), { redirect: "manual" });
  },
};

The worker and its config are in deploy/cloudflare/. With your domain on Cloudflare, set your subdomain and Function URL in wrangler.jsonc and run:

cd deploy/cloudflare
npx wrangler deploy

Wrangler creates the DNS record and the HTTPS certificate for the custom domain. Workers are free for up to 100,000 requests a day.

Other options​

  • Container platforms like Google Cloud Run, Fly.io or Render run a Docker image and pass PORT. Kraken's Dockerfile builds and runs the demo. Cloud Run and Fly.io can also scale to zero.
  • A VPS (any Linux server): build with make, run ./main under systemd, and put Caddy or nginx in front for HTTPS.
  • Your own machine: run ./main and expose it with a tunnel such as cloudflared tunnel.