A few months ago, I wrote about our adventures building a brand new project using React. Our existing SparkPost app was built many moons ago in the ancient framework known as AngularJS. In that earlier post, I said (and I quote), “So what about that 50,000-line Angular app? We won’t be migrating it to React, at least not directly.” Well, we’re migrating it to React. Directly.
It wasn’t that I lied, but let’s say I was a victim of my own lofty idealism. Shortly after that article was published, I started in on my plan to improve the Angular app. After almost two weeks of struggling to get Bower out of the project, I was immediately rethinking all of my ideals and life choices. I’d tried one thing and it was hard. It was time to leave ideals behind.
In all seriousness, we considered a couple of other wild ideas like writing React inside of Angular or migrating everything one page at a time. Ultimately we realized that the fastest and safest way to rebuild our entire app in a brand new framework was to do exactly what I’d publicly ridiculed: direct migration, i.e. a parallel rebuild. The big question was would SparkPost get behind this big idea?
Before I go on, I should explain something very clearly. The chance to completely rewrite your existing app from scratch is a rare occurrence in the world of engineering. If you come away from this article thinking, “I want to totally rewrite OUR app!”, you should know that you’re going to need a Very Good Pitch (and some flexible decision makers willing to carve you a good chunk of time).
That said, there’s tremendous value in the chance to rethink and rebuild your entire user interface as a singular, consistent unit instead of continuing to stack more cards on top of a shakier and shakier house. SparkPost got on board with the rebuild and we started in on a plan.
So how do you rebuild an entire web application?
Step 1: Catalog and build a solid foundation of reusable components, design tools, and critical features like authentication.
Step 2: Draw the rest of the [censored] owl.
To be honest, our original plan was only slightly more fleshed out than that, and we’ve shifted and adapted it a lot along the way. Here are a few of the pieces we’ve been most happy about and that have helped us prepare for maximum owl-drawing.
- Focus: Our entire application team has always been “full-stack”, bouncing from back end to front end and back. As we’ve grown we’ve recognized a need for a core group of folks who can completely focus on a project like this front-end rebuild. That kind of focus helps us maximize efficiency and create the best foundation possible. It also allows us to be a bridge to the rest of the team about how this new app is built and, eventually, how they can contribute.
- Modular design: Jon Ambas, our front-end engineer and design lead (and my main partner in the early days of dreaming up this plan), has been all-in on creating a fully-featured component library to make our work in React a lot simpler and more visually consistent. We now use Matchbox throughout the app and rarely have to think about CSS for anything except custom corner cases.
- Openness: Client-side code is already public when it runs in your browser. Therefore, I saw no reason to keep the code private in a repository. We don’t expect anyone will use this code but us (please don’t). It’s already been nice to be able to point to examples of things we’re doing and link directly to the code in GitHub.
- Testing and Monitoring: This has been a big goal of the rebuild work we’re doing. Our Selenium-driven functional tests in Angular were taking hours to run and constantly failing until we finally turned them off completely. In the new app, we’re writing lots of tests (? snapshots) but we decided early on that we’ll never be able to predict every bug with test cases. For that reason, we’re also investing in real-time error logging and monitoring. No more waiting for a customer to report an error introduced in a deploy that went out 8 hours ago. Fast stable tests supplemented by error monitoring is our new way forward.
- “Preserve and improve” as a redesign strategy: In other words, we’ve preserved the architecture of the app. Leaving most things generally in the places you’d expect to find them. Avoid incredibly drastic changes that would require lots of approvals, user testing, etc. We’ve also taken every opportunity we’ve had to make easy, obvious improvements as we rebuild each part of the app. The result so far feels like an incredibly fresh coat of paint on a familiar base. Which will hopefully help us when we roll the new app out to our existing customers.
- Rollout: Incremental rollout, to be specific, is the last big part of our very big plan and probably the least-well-defined at the moment. We know we want to incrementally roll the app out to users (randomly and/or by allowing users to opt-in) so we can test out how it works for real use-cases before we unleash it on everyone. We’re still looking at options for this, but getting the rollout right will be one of the most important parts of our plan overall.
The takeaway here is that you should be skeptical of anything I say, especially if it sounds good or idealistic. When it comes to something like application development, being practical is almost always better than being pure.