Memoji of Merlin smilingMemoji of Merlin winking

My blog.

All posts
Building a 'like button' with Astro & Val.Town

Building a 'like button' with Astro & Val.Town

6 mins read

There’s a lot of excitement currently around Val.Town so I thought I’d check it out and see what it’s all about. I thought it would be cool to add a like button to my Astro blog posts as this is suitably low risk and effort but significant enough that it throws up some real world challenges and allows us to test the Val.Town platform.

What is Val.Town?

Val.Town is a social platform that allows you to write and share snippets (either JSON or functions) of code with low friction. It’s sort of similar to a GitHub Gist but with several significant improvements:

  1. You can run the code you write in the browser.
  2. When you save a Val (snippet) it is instantly deployed as a serverless function that can be accessed via a URL.
  3. When you save a Val it is instantly published to an internal registry, meaning other Vals can use it as a dependency. This makes composability and sharing code in the community very easy.

They have great introduction to what Vals are and how to use them in the Val.Town docs.

What is Astro?

Astro is the framework this blog is built on, it’s a static site generator that implements the Islands Architecture. It’s entirely pre-generated at build time and has no backend.

What are we building?

A like button! You’ve probably seen one before. Let’s define some requirements:

  • The like button should appear on the bottom of every blog post
  • It should display the count of likes per post
  • It should enable a reader to like a post without having to authenticate
  • It should prevent a reader from liking a post more than once
  • We should take reasonable steps to prevent from spam and abuse

API design

From the requirements I think we’re going to need two API endpoints.

GET /likes/:slug to get the number of likes for a given blogpost slug. However there’s a limitation of Val.Town that we can’t use dynamic routes like this, so instead we’ll use a query parameter like this: GET /likes?postSlug=example. This feels slightly less RESTful, but it’s not a big deal.

POST /likes to increment the number of likes for a given blogpost slug. This time we’ll include the postSlug in the body of the request.

User identification

We need to identify users so we can prevent them from liking a post more than once. Let’s look at the options:

  1. ❌ User accounts - We could create user accounts and require authentication, this is the most robust solution but it’s also the most friction for the user and added implementation time for us.
  2. ❌ Cookies - We could use a cookie to identify users, this is quick to implement but less robust as users can clear their cookies or switch to incognito mode. It also means writing a cookie policy and having an annoying cookie banner, ain’t nobody got time for that.
  3. ✅ Fingerprinting - Use a device fingerprinting library, this is quick to implement and is reasonably reliable to uniquely identify devices. It works across incognito and normal browsing modes and clearing the cache doesn’t affect it. Additionally when using it for the purposes of adapting a user interface it’s exempt from requiring user consent or a privacy policy. This is the option we’ll go with. We can send the fingerprint as a header in the request to our API like so Authorization: Bearer <fingerprint>.

Persisting likes

Val.Town has several options for persisting data with third party services such as Upstash, Neon, Supabase, PlanetScale etc. All of these seem overkill for our requirements. They also offer a SQLite instance you can access via WASM which looks cool but really I think we can settle with the simplest option of persisting state in a Val itself as a simple Key/Value store. This currently has a limit of 100kb per Val which roughly gives me space for storing 3000 likes. This blog has very limited traffic so I think we’ll be fine, and will be a nice problem to have in the future if we ever get to that point.

    "example-post-slug": ["fingerprint1", "fingerprint2"]

Preventing spam and abuse

This is somewhat tricky in a world where we don’t have user accounts. We want to take a defence in depth approach, layering protections to give us the best chance of preventing abuse. Let’s look at the (albeit limited) options:

  1. ✅ Rate limiting - We can limit the number of requests per IP address in a given time frame. This is a good first line of defence but it’s not foolproof as attackers can use proxies to get around it.
  2. ✅ Restrict domains - In theory we could use CORS to limit the domains that can make requests to our API. Unfortunately while Val.Town uses Express.js, you cannot use the CORS middleware. We can however check the Origin header to ensure it matches this domain. It’s worth noting that while you cannot spoof the Origin header in a browser, you can spoof it using curl or other HTTP clients, this limitation is also true of CORS.
  3. ✅ Input validation - We can validate the input to our API to ensure it’s what we expect. We can validate the slug is in fact a valid slug from our blog and that the fingerprint is a 32 character string. Fingerprints could still be fabricated but it limits the damage of spam.
  4. ❌ Captcha - We could use a captcha to ensure the request is coming from a real user. Realistically this is overkill for a like button and would create too much friction for the user.

Val.Town implementation

POST /likes

Ok, let’s get started! First we’ll create a new Val, Val.Town automatically infers the name of the Val from your function. Here’s we’re creating an async function that accepts an Express.js request and response object.

You’ll also notice the @username.val syntax in the code, this is how we can reference other Vals. The @me is a shorthand reference to the current user, in our case it’s equivalent to @merlin.

Here you can see we call a Val we created called @me.auth which checks the rate limiting, Origin header and that a device fingerprint is provided in the expected format.

We also check that the provided slug actually exists in our blog and that the the req method is what we expect.

Right, now all that’s out of the way we can actually persist the data. We can just grab the state from our KV Val, modify it and re-save it. You can see we’re using new Set() and spreading back to an array to ensure our fingerprints are unique. Finally we can return a 200 success response.

GET /likes?postSlug=example

The GET endpoint is pretty simple, we just grab the state from our KV Val and return it as JSON.


Here’s a look at our auth Val. It also features a good example of how you can leverage other people in the communities Vals, in this case we’re using @stevekrouse.rateLimit to rate limit requests.

Validating a post slug is valid

Validating a post slug is valid is a little more involved. My initial plan was to create a webhook in Val.Town and call this everytime the site is deployed with a list of valid slugs. This would decrease latency when validating but digging into CI deploy tasks seemed like a potential rabbit hole so I decided to just fetch the list of slugs from the blog itself. This means we have to make a request to the blog every time we validate a slug, but I think that’s fine if we implement optimistic updates in the UI.

On the Astro side of things, this was delightfully easy to setup, I just created a file in the pages directory that returns a JSON response of all the slugs.

// /src/pages/posts.json.ts
import { getCollection } from "astro:content";

const allBlogPosts = await getCollection("blog", ({ data }) => !data.draft);

export async function get() {
    return {
        body: JSON.stringify({ slug }) => slug)),

Client side implementation

Finally we can implement the client side code. I won’t go into too much detail here as it’s pretty standard React code, but feel free to browse the source code if you’re interested.

<LikeButton slug={slug} client:visible />

We can include the LikeButton component in our blog posts as a dynamic island. Using Astro’s client:visible directive we ensure that the scripts for the island and the API requests are only made as the user scrolls to the island, reducing the initial page load time.


So, how was it? I think generally I was impressed with Val.Town, in terms of the friction of deploying a series of serverless functions it really can’t get much easier than this.

You could achieve the same thing with a similarly low amount of effort with Vercel’s KV store and NextJS, but if you’re using an entirely static site like Astro then Val.Town is a great option.

The editing experience in Val.Town is OK… They have syntax highlight and some level of autocomplete and linting but it’s not as good as VSCode and obviously lacks something like CoPilot. I did find sometimes when working on multiple Vals at once the linting would get out of sync and you’d have to refresh the page. The formatting currently removes line breaks which I find impacts the readability of the code. It’s also not really clear if you have unsaved changes or not.

The debugging experience is also a little lacking, you can see the logs of your Vals but you have to go to a different page to see them. It would be nice if you could see the logs inline with the code.

I also think a highly simplified testing solution would be really handy where you could provide a range of example inputs and expected outputs.

While there is a Explore page on their site it’s somewhat limited and the search isn’t great. I found I didn’t have that much opportunity to leverage other peoples Vals. At the moment you can’t really differentiate between Vals that are public so the community can reuse them and Vals that are public but are just for the author to use.

All this being said, it’s a new platform and I’m sure the team very busy working on all sorts of enhancements, so I’m confident it will improve over time. It’s a genuinely unique and potentially powerful idea for a platform.

There’s a growing list of integrations with other platforms too, which may prove useful. I think the fact you can schedule Vals to run might make it a useful tool for quick and dirty data gathering and aggregating between platforms, a problem nearly every startup has.

Amazing concept and I’m excited to see where it goes!

ps, I still love Astro ❤️

pps, Why not give the like button a go? 👇

Loading likes...