As Artsy Engineering grows in 2018, we have so many newcomers looking for context: they want to understand the systems they'll be working in day-to-day. Awesome! But it's not enough to understand the systems themselves, it's often helpful to understand the history of how we ended up where we are.
Frontend web development has changed a lot during Artsy's existence, and it continues to advance at a blistering pace. It's easy to get caught up in the churn of frameworks and languages and tools, so I want to use this post as an opportunity to contextualize each transition that Artsy's web presence has made over the past seven years. We've changed technologies, but we've tried to do so with care and attention. Documenting these decisions is important (and is ideally done contemporaneously), but even with the best documentation, sometimes our own documentation is unclear to us.
In an effort to help contextualize our web frontend (which is open source), this blog post will document the major transitions that Artsy's web presence has made over the past seven years. Let's begin!
Backbone + CoffeeScript
Artsy as you know it today began as a standard Rails application. We ran
git init in January 2011, which coupled
our backend API to our web frontend, but since our frontend was just a fancy user interface for our API, this
worked for over two years. The web app itself was a kind of simplified MVC – controller logic lived inside the
views and models dealt with backend communication and notifying the view of state changes. For CSS, we used the
SASS CSS preprocessor. The Rails backend served initial pages that were then populated with follow-up API calls
made on the client-side. At a very high level, this isn't that different from what we do today with React.
Our site was built with a framework called Backbone, which was really well-suited for our needs at the time. From their documentation:
As an outsider to the web at that time, I can't comment too heavily on Backbone. It seems like the freedom (emphasis theirs) that they describe is a freedom from tangled jQuery code everywhere. I think our definition of freedom on the web frontend has evolved since then, but that's just my feeling.
Our goal as a (very small!) engineering team at the time was to keep our moving parts to a minimum. Rails+SASS+CoffeeScript+Backbone helped us achieve that goal, and we couldn't have gotten this far without the help of those projects.
Ezel & Friends
In November 2013, we split our web frontend from the API backend. You can read all the details in this blog post, but the story is summarized nicely as "moving from a single monolithic application to modular Backbone apps that run in Node and the browser and consume our external API." This move from monolith to modular systems continues to influence day-to-day work on the Artsy Engineering team.
We had already started moving away from a typical Rails app by moving our API to Grape in order to support an iOS application. The monolith also had some clear drawbacks including severe page load times, maintaining duplicated backend and frontend UI templates, slow test suites, and poor developer productivity. We took the project of building our mobile web frontend, m.artsy.net (still known as "martsy" internally) as an opportunity to address these problems.
We built our new site with Node.js since it allowed us to share and consolidate our server/client rendering code. We split out areas of concern into separate "apps", with their own bundled CSS/JS to help page load times. We server-side rendered above-the-fold content and used client-side JS to load the rest, which helped SEO and user experience. We took a BEM-like approach to our CSS, which helped developer productivity. Our technical decisions were driven primarily by our desire to create great user experiences.
And because we are an open source by default organization, we collected these approaches into an open source project called Ezel. While our main app used this Ezel approach, other new web apps – CMS systems for our partners, auction-management systems for our admins, etc – were built on new internal tools to share assets and code across the apps. We experimented a lot; we got pretty good at sharing resources across codebases. Most of our web projects started on Heroku before moving to heavier-duty deployments as needed. Our frontend mindset at the time (2015) was focused on getting to a stable, predictable stack. However... we started experimenting with React around the same time.
CoffeeScript and Backbone were still working for us, and we still use them in production in many systems. However, the state of the art in web development moved on. When I joined the auctions team and helped maintain one of our CoffeeScript+Backbone apps, I was very confused about how data flowed from one part of the app to another, across languages, with a lot of magic happening. I think that's typical in these kinds of apps – "convention over configuration" is a good mantra if you can expect that incoming engineers are familiar with the conventions. That's just not the case anymore.
By 2016, we had experimented with React and followed up with another app built with the technology. React (and Redux) were very well-suited for our realtime auction bidding UI, and would later prove helpful in our editorial CMS. These experiences helped prove the technology was ready for production use and convinced us that React was great at reducing the complexities of building user interfaces (the realtime nature of our auctions product was particularly well-suited for Redux's state management; it was our first from-scratch React app).
When the Artsy business require us to make changes to how we build software, like splitting up our monolith, we try to take full advantage of those changes to improve how we work, which means evaluating new tools. Adopting Node.js and Ezel wouldn't make sense today, but at the time, they helped us scale up Artsy's business without the same scaling up of our engineering resources. Ezel helped us do more with less, which is still an important criteria we use for evaluating new tools.
By 2017, the divisions between our mobile frontend and web frontend teams had been totally dissolved (as they should – the division between mobile and web developers is a false dichotomy). Our 2017 tech stack post discusses this in depth, but our goal was really to unify the paradigm that frontend engineers at Artsy use to build user interfaces, whether that's on mobile or web. React and React Native were our answer to that challenge.
On the web side of things, however, Artsy had another challenge. Sure, React is great, and sure, it's how we want to build user interfaces, but how do we get there? We're not fans of large rewriting projects, so we opted for what we call an "incremental revolution" approach. We built a library called Stitch that would let us mount React components inside our existing app. Using this approach, we could migrate to React component-by-component. We've been using Stitch in production for over a year and have been very happy with its approach; you can read more details of integrating it into our main frontend app in this blog post.
Today, principal React work takes place in a shared components repo. We share these components across several of our web apps using Stitch. We have been pretty pleased with the results! But our dive into React is only just beginning. The community is moving quickly to figure out what best practices make sense in the React paradigm, and we're a part of that. We are evaluating technologies like styled-components and styled-system to create a universal design system within Artsy. The area is under very active development, so I'll save details for a future blog post.
I can't go too much further without talking about GraphQL. v1 of our API (REST) is still in use around much of Artsy and, despite the best efforts of some of our engineers, v2 of our API (HAL) hasn't gained significant internal use yet. Instead, we found ourselves building a GraphQL server to orchestrate API calls to our existing APIs. This confers many benefits, which I describe from a mobile perspective in some detail here. The key thing to understand about our GraphQL server, which is open source, is that it is under the stewardship of our frontend engineers, not our platform engineers. That's not to say that our platform team isn't involved with its development – in fact, they've been key to scaling it up – but Artsy frontend engineers created the server to help us build better UIs, and while the technology is still very new, we continue to see it pay dividends.
Since I joined Artsy, I've seen us continually investing in tooling that helps us build better software. The results of this culture-of-continuous-improvement speak for themselves: with fewer than 30 engineers total, we support a growing company with a suite of software built for many canvasses and we have an outsized impact on the software industry relative to our size. Our frontend web stack is just one facet of our evolving technology – there's lots of exciting stuff on the backend, too (for example...). Through my research for this blog post, I learned a lot about what drives technological decisions on our team. From humble beginnings as a Rails app, to CoffeeScript and Bootstrap, to React and GraphQL, Artsy Engineering has evolved our frontend software to achieve a quality worthy of art – both from the user's perspective and from the developer's. I'm very excited about what's coming next, and I can't wait to share it with you. Have a great day!