As of today, with the Wrangler CLI, you can now deploy entire websites directly to Cloudflare Workers and Workers KV. If you can statically generate the assets for your site, think create-react-app, Jekyll, or even the WP2Static plugin, you can deploy it to our entire global network, which spans 194 cities in more than 90 countries.
While you could deploy an entire site directly to Workers before, it wasn’t the easiest process. So, the Workers Developer Experience Team came up with a solution to make deploying static assets a significantly better experience.
Using our Workers command-line tool Wrangler, we've made it possible to deploy any static site to Workers in three easy steps: run
wrangler init --site, configure the newly created
wrangler.toml file with your account and project details, and then publish it to Cloudflare's edge with
wrangler publish. If you want to explore how this works, check out our new Workers Sites tutorial for create-react-app, where we cover how this new functionality allows you to deploy without needing to write any additional code!
While in hindsight the path we took to get to this point might not seem the most straightforward, it really highlights the flexibility of the entire Workers platform to easily support use cases that we didn’t originally envision. With this in mind, I’ll walk you through the implementation and thinking we did to get to this point. I’ll also talk a bit about how the flexibility of the Workers platform has us excited, both for the ethos it represents, and the future it enables.
So, what went into building Workers Sites?
“Filesystem?! Where we’re going, we don’t need a filesystem!”
The Workers platform is built on v8 isolates, which, while awesome, lack a filesystem. If you’ve ever deployed a static site via FTP, uploaded it to object storage, or used a computer, you’d probably agree that filesystems are important. For many use cases, like building an API or routing, you don’t need a filesystem, but as the vision for Workers grew and our audience grew with it, it became clear to us that this was a limitation we needed to address for new features.
Welcome to the simulation
Without a filesystem, we decided to simulate one on top of Workers KV! Workers KV provides access to a secure key-value store that runs across Cloudflare’s Edge alongside Workers.
wrangler preview or
wrangler publish, we check your
wrangler.toml for the site key. The site key points to a bucket, which represents the KV namespace we’ll use to represent your static assets. We then upload each of your assets, where the path relative to the entry directory is the key, and the blob of the file is the value.
When a request from a user comes in, the Worker reads the request’s URI and looks up the asset that matches the segment requested. For example, if a user fetches “my-site.com/about.html”, the Worker looks up the “about.html” key in KV and returns the blob. Behind the scenes, we’ll also detect the mime-type of the requested asset and return the response with the correct content-type headers.
For folks who are used to building static sites or sites with a static asset serving component, this could feel deeply overengineered. Others may argue that, indeed, this is just how filesystems are built! The interesting thing for us is that we had to build one, there wasn’t just one there waiting for us.
It was great that we could put this together with Workers KV, but we still had a problem…
Cache rules everything around me
Workers KV is a database, and so it's set up for both read and write operations. However, it's primarily tuned for read-heavy workloads on entries that don’t generally have a long life span. This works well for applications where data is accessed frequently and often updated. But, for static websites, assets are generally written once, and then they are never (or infrequently) written to again. Static site content should be cached for a very long time, if not forever (long live Space Jam). This means we need to cache data much longer than KV is used to.
To fix this, on publish or preview, Wrangler walks the entry-point directory you’ve declared in your wrangler.toml and creates an asset manifest: a map of your filenames to a hash of their content. We use this asset manifest to map requests for a particular filename, say index.html, to the content hash of the most recently uploaded static asset.
You may be familiar with the concept of an asset manifest from using tools like
create-react-app. Asset manifests help maintain asset fingerprints for caching in the browser. We took this idea and implemented it in Workers Sites, so that we can leverage the edge cache as well!
This now allows us to, after first read per location, cache the static assets in the Cloudflare cache so that the assets can be stored on the edge indefinitely. This reduces reads to KV to almost nothing; we want to use KV for durability purposes, but we want to use a longer caching strategy for performance. Let’s dive in to exactly what this looks like:
How it works
When a new asset is created, Wrangler publish will push the new asset to KV as well as an asset manifest to the edge alongside your Worker.
When someone first accesses your page, the Cloudflare location closest to them will run your Worker. The Worker script will determine the content hash of the asset they’ve requested by looking up that asset in the asset manifest. It will use the filename and content hash as the key to fetch the asset’s contents from KV. At this time it will also insert the asset’s contents into Cloudflare’s edge cache, again keyed by filename and content hash. It will then respond to the request with the asset.
On subsequent requests, the Worker script will look up the content hash in the asset manifest, and check the cache to see if the asset is there. Since this is a subsequent request, it will find your asset in the cache on the edge and return a response containing the asset without having to fetch the asset contents from KV.
So what happens when you update your “index.html”- or any of your static assets? The process is very similar to what happens on the upload of a new asset. You’ll run
wrangler publish with your new asset on your local machine. Wrangler will walk your asset directory and upload them to KV. At the same time, it will create a new asset manifest containing the filename and a content hash representing the new contents of the asset. When a request comes into your Worker, your Worker will look into the asset manifest and retrieve the new content hash for that asset. The Worker will look in the cache now for the new hash! It will then fetch the new asset from KV, populate the cache, and return the new file to your end user.
Edge caching happens per location across 194 cities around the world, ensuring that the most frequently accessed content on your page is cached in a location closest to those requesting content, reducing latency. All of this happens in *addition* to the browser cache, which means that your assets are nearly always incredibly close to end users!
By being on the edge, a Worker is in a unique position to be able to cache not only static assets like JS, CSS and images, but also HTML assets! Traditional static site solutions utilize your site’s HTML an entry point to the static site generator’s asset manifest. With this method of caching your HTML, it would be impossible to bust that cache because there is no other entry point to manage your assets’ fingerprints other than the HTML itself. However, in a Worker the entry point is your *Worker*! We can then leverage our wrangler asset-manifest to look up and fetch the accurate and cacheable HTML, while at the same time cache bust on content hash.
Making the possible imaginable
“What we have is a crisis of imagination. Albert Einstein said that you cannot solve a problem with the same mind-set that created it.” - Peter Buffett
When building a brand new developer platform, there’s often a vast number of possible applications. However, the sheer number of possibilities often make each one difficult to imagine. That’s why we think the most important part of any platform is its flexibility to adapt to previously unimagined use cases. And, we don’t mean that just for us. It’s important that everyone has the ability to customize the platform to new and interesting use cases!
At face value, the work we did to implement this feature might seem like another solution for a previously solved problem. However, it’s a great example of how a group of dedicated developers can improve the platform experience for others.
We hope that by paving a way to include static assets in a Worker, developers can use the extra cognitive space to conceive of even more new ways to use Workers that may have been hard to imagine before.
Workers Sites isn’t the end goal, but a stepping stone to continue to think critically about what it means to build a Web Application. We're excited to give developers the space to explore how simple static applications can grow and evolve, when combined with the dynamic power of edge computing.
Go forth and build something awesome!
Have you built something interesting with Workers? Let us know @CloudflareDev!