I'll warn you ahead of time and say this post isn't too much more than what you can find in the documentation, but I wanted to see it work for myself so I had to setup a test locally. Cloudflare Service bindings are a way for one Worker to connect to another. That seems simple enough, but while it defines a "connection", that connection is completely internal to the Cloudflare environment. I.e., incredibly fast with much lower latency. Let's consider a simple example.

The Receiver #

I began by creating a worker, named backworker, with just a simple message:

export default {
	async fetch(request, env, ctx) {
		return new Response('Hello from Backworker');
	},
};

The Front #

I struggled with what to call that header, "front end" felt like a loaded term as it implies HTML, etc. Anyway, I made a second worker named frontworker. In order to "connect" it to the back, you need to edit your wrangler.toml:

services = [
  { binding = "backlogic", service = "backworker" }
]

Two things to note here. The service value points to the name of the worker where the binding is how you will address it. I suppose normally you would make these the same. I chose a different name just so I could ensure it worked properly.

In order for this worker to communicate with the other, you use the env object and binding name in your code. Here's how it looks:

export default {
	async fetch(request, env, ctx) {
		const backResponse = await env.backlogic.fetch(request.clone());
		let resp = await backResponse.text();
		return new Response(`Hello from front, back said: ${resp}`);
	},
};

You use fetch to communicate, which is a network call, but remember this is going to be internal only. It does need a request object which can only be read once, hence the use of request.clone(). As I didn't bother changing my other service to return JSON, I just get the text response and include it in the response here.

Testing #

When working locally, you will need to have both workers running. While I wasn't sure it was required, I ensured I started backworker first, and then frontworker. The CLI noted the binding:

Terminal output showing that it recognized the binding to backworker.

Opening it up and running gives you what you expect:

Hello from front, back said: Hello from Backworker

That's mostly it, but there's one more cool aspect. If my intent is for backworker to never be used by itself, I can actually disable its route in the dashboard:

URL route disabled

Now the worker is no longer available publicly, but the front one works just fine: https://frontworker.raymondcamden.workers.dev/

If you would like to test this yourself, you can clone the two workers from my new demo repository here: https://github.com/cfjedimaster/cloudflareworkers-demos

Photo by Patrick Wittke on Unsplash