Jul 5, 2022

The Infrastructure Is The Framework

#programming

Last week at the San Francisco Vercel meetup, Malte Ubl was talking about exciting things in the future of frontend engineering. Next.js 12.2 had just launched the same day with Layouts and Nested Layouts. Coming from an Ember.js background that has had routing and nested routing forever (albeit with longstanding, thorny issues and usability problems), I had skimmed past this announcement, but a particular piece of Malte's answer stuck out to me as one of the most innovative things I'd heard in a long time.

One of Vercel's value propositions is to easily deploy websites to "The Edge". Generically, that means deploying to datacenters around the world so that requests from end users around the world can be delivered with the lowest possible latency. Latency is a function of geographic distance, so the closer the datacenter is to the website visitor's location, the faster the website can be delivered.

But most websites need data. And data lives in databases that don't live on The Edge (because replicating data is a hard problem). If a website is delivered from The Edge network, but it needs to fetch data from some centralized location, it adds latency back into equation. For example, let's say I visit a web page from India, and this web page requests some data during a server-side render, but the data is in a US data center. The latency avoided by routing the request to a datacenter at the Edge is hamstrung by the latency added by the data request.

Drawn with tldraw.com

So how do we solve this? Generically speaking, there isn't a good answer. Web authors need to carefully measure the tradeoff between keeping the user close to the web server, or keeping the web server close to the database server.

Malte talked about how Next.js Layouts could enable Vercel to intelligently split a website such that the outer layers (the "skin") could be deployed to the Edge, but the "meat" of the website could live close to the data. In other words, a Next.js website deployed on Vercel could deliver the static part of the website instantly from the Edge, improving First Paint timings, but then deliver specific data-heavy parts of the page from datacenters close to the database servers. 🤯🤯🤯

In response, Theo pointed out that this seemed like a very complicated solution to avoid talking about delivering data to the edge. (And when can we have that, please?) There aren't any clear answers to that yet, sadly, but the idea of using templating layouts to draw boundaries between datacenters felt Cutting Edge™ to me.


I've recently accepted a job at Vercel to work on Turborepo (consider this a disclosure for this post?). While talking to folks at the company about how Vercel can differentiate themselves from other platforms, I heard several times that tightly integrating Next.js the framework with Vercel the platform can create some pretty unique products for end users. At the time of these conversations, I didn't really understand this integration. Incremental Static Regeneration was brought up as an example, but ISR didn't particularly seem like a Next.js specific feature. I didn't understand how this integration was unique to Next.js and Vercel. This comment (perhaps just an idle thought from Malte) brought the philosophy home for me, and I think that's the real story here.

The idea that the infrastructure is part of the web framework's design is really exciting. My tiny, feebly brain can't think of any other integrations yet, but I can grasp the paradigm shift here. We've come a long way since developer teams handed their code over to an Ops team to deploy their code, to when "devops" became a thing, to "infrastructure expressed alongside code", and now to this world where infrastructure is an integral part of how you think about your code.

If you like this post, please share it on Twitter and/or subscribe to my RSS feed. Or don't, that's also ok.