Don’t Let Your Love For Code Make You a Bad Engineer

If someone offers me $10k to help them sell their hand knit mittens online, I’m not going to build a custom e-commerce website from scratch. Instead, I’ll shoot them a link to Shopify and advise them to keep their coin. What do you do, however, if you’ve never heard of Shopify or the plethora of other WYSIWYG e-commerce solutions?

At Spraoi, we provide an “insurance as a service” platform called Kwikcover. Kwikcover is built such that we can add, remove and reuse features on a per-client basis. Oh, and it scales—not just in the number of users that can buy insurance at the same time, but also in the number of engineering teams that can work on the codebase simultaneously without stepping on each others toes.

To achieve this, we launched domain specific APIs (read: “microservices”) for users, products, policies and payments (among others), but it wasn’t immediately obvious how we should tie these services together.

When one of our UIs makes a request with a user id, product id, and some payment info (i.e someone is buying insurance), which API should handle it? If you’ve never worked with microservices, your initial thought might be payments. The payments service would make a call to the products API to fetch the product info, then, after successfully executing the transaction locally, would create a new policy by making a call to the policies API.

That would technically work, but then our payments service ends up with logic that isn’t directly related to payments. A better solution is to have a separate service specifically built to orchestrate requests. This is one of the first problems I was tasked to solve when I joined the Spraoi team.

After doing some research, I was ultimately convinced that there weren’t any tools out there that met our specific needs. As such, I happily started writing a custom solution from scratch. After a few days of hacking together some Javascript and throwing it on AWS Lambda, we had the original version of what we now call the “request composer”.

Each new endpoint exposed to our UIs has to declare a “request sequence” that determines how that request should be orchestrated. This sequence is defined in a domain specific language (DSL) that only the request composer can parse, and its syntax was conjured by yours truly. This boils down to the fact that, if you need to create a public endpoint at Spraoi, you have to write, or at least copy/paste, this DSL.

To make it possible for others to write these sequences, I’ve written tests and documentation, and attempted to keep both up-to-date. However, when things go wrong, or another requirement comes through, I’m still on the hook. The kicker? It’s possible that the majority of this work could have been avoided.

Well after the creation of the request composer, I stumbled across another AWS service: Step Functions. The first paragraph on their homepage states:

AWS Step Functions makes it easy to coordinate the components of distributed applications and microservices using visual workflows.

It was almost painful to read. Is that not exactly the purpose of our request composer? Also, “visual workflows” sounds user friendly… Hindsight is 20/20, but I feel pretty silly for not finding this and trying it out before attempting to build an in-house solution.

A chef probably prefers a self-prepared meal over takeout. Similarly, writing code to solve a tasty problem is where most developers feel at home. More code, though, means more time spent testing, debugging and documenting code.

This might have been a long-winded way of saying “reinventing the wheel is expensive”. However, keeping this story in mind when you’re tasked with solving a “new” problem could be the difference between coming three feet from gold and striking it.