The Value of Data Engineering

The who, what and why of pipelines, persistence and platforms.

2020/02/16


A few years ago a colleague of mine explained how, when asked about his role, he would glibly respond, "I do things with computers". At the time this sounded fairly reasonable. I too have fun doing things with computers and will always consider myself something of a generalist technologist.

However, having been involved in a variety of projects, I've begun to realise it's the data-intensive ones which particularly hold my interest. So I'm now embracing this and taking on data engineering as a full time role.

Whilst it's not the first time I have held this job title, working in tech consulting has broadened my horizons on the various challenges folks have getting value from their data and I find myself returning to the role with newfound enthusiasm.

Distilling Data & Enabling Decision Making

The goal of this post was to share with you why I think data engineering is a pretty exciting discipline. And that remains the case. But, truth be told, as I started writing I had a minor identity crisis as to what exactly my role as a data engineer should entail. So I had a go at exploring some definitions.

Data engineers are variously described as being responsible for:

Is this a fair characterisation? Well, sort of. The above responsibilities do describe activities that both myself and colleagues have carried out while working on data projects. But, this explanation doesn't really touch on why we do this work. Why do we build pipelines, optimise queries and scale out clusters? What end are we working towards? Here is my attempt at defining data engineering in terms of what it enables.

The practice of building software that can reliably distill data into a meaningful form, such that a human or machine can easily make a decision and/or take an action.

Ultimately, it's a lot easier to demonstrate value when you empower people to confidently make a decision, or at the very least, reduce uncertainty. I wouldn't even suggest we necessarily need super sophisticated techniques to do so. Never underestimate the power in having the right data in the right place, with some basic analysis.

This definition could also perhaps apply to data science, but I think that's OK given the highly complementary nature of these two roles. In fact, as a colleague of mine helpfully pointed out, if the act of "distilling data" is orthogonal (i.e not exclusive to engineers or scientists), then the role of the data engineer is about doing so with a focus on reliability, data quality and data availability. I agree with this and think a key enabler for doing so is to bring all the principles we know and love about making high quality software to the realm of data. This includes practices like test driven development, domain modeling, observability, etc. I would also make the case that in certain circumstances it also entails building platforms to enable economy of scale. (The Data Mesh architecture is a great example of how to do this elegantly.)

Reasons to like Data Engineering

So now we have a bit more of an understanding as to what data engineers are responsible for, let's explore some of reasons you might consider pursuing it as a skill. Here are some things that motivated me.

Close to the domain

Unlike my previous role as an infrastructure developer*, data engineering fully immerses you within a particular domain. Calculating sales forecast for a supermarket? You are probably going to learn a fair bit about the economics of avocados. Building a routing service to plan journeys? You run the risk of eventually becoming something of a train geek.

Every dataset has its own quirks. For example, while working within automotive retail I uncovered some quite interesting 'rare colours' of car (colours that only appear once in a dataset of several million cars). I was amused to discover that car manufacturers would do things like describe a completely unremarkable SUV as being "Mystic Beige" - definitely an oxymoron.

I love finding a weird outlier in the dataset and trying to uncover the story behind it. "What on earth happened here, did a single customer really order 187 mangos?" Working in data is rewarding for those who are fundamentally very curious.

Also, it's gratifying to see users glean insights from data you have delivered. This is great. It's a valuable form of feedback and it's easy to see the impact of your work.

* Big respect for infra-devs. This work is super valuable, sometimes under appreciated and so so vital to get right.

But still close to the metal

Ultimately, all data, whether the colour of a car, or a count of mangos sold, is stored as a series of bits and bytes in one or more computers. Whilst I enjoy indulging my data-curiosity, I also find a lot of satisfaction in learning about the lower-level mechanics of how these values move about the computer. Not always, but often enough, having some appreciation of this can also be particularly important when designing and implementing data solutions.

This applies to all sorts of cases. Maybe the data is big, maybe the data is small but computationally expensive. Perhaps the data needs to be fetched from a distant machine running a strange and arcane legacy system. In these situations, where performance starts to matter, we need to pay attention as to how the shape of the problem intersects with the strengths and weaknesses of the hardware we are building on top of.

This concept of designing software to take into account the design of the underlying hardware is sometimes expressed as "mechanical sympathy". This term was popularised by Martin Thompson and attributed to the Formula 1 racing champion Jackie Stewart. He "believed the best drivers had enough understanding of how a machine worked so they could work in harmony with it".

We can use this principle to guide many of our design choices when coming up with solutions. For instance:

Even when performance isn't as much of a concern, there are still compelling economical and ecological arguments to be made for being at least mindful of this and writing software that uses less resources.

Room for innovation

The other main reason I want to invest in data engineering is because the entire "data" field continues to have lots of interesting developments.

There are focused but useful tools such as DBT that help compose data transformations. There are projects like Apache Arrow which aim to establish solid foundations for the interplay of data between different technologies in the open source ecosystem. We have programming languages such as Rust that demonstrate a lot of potential for solving data intensive problems.

Exciting happenings in the ops space could be beneficial to experiment with. For example, instrumenting data pipelines for observability. We are also starting to see examples of how to do continuous delivery for machine learning.

Beyond technology, the shape of collaboration is also evolving. Instead of a central team managing pipelines, many are now embedding data engineers in product teams.

Being excited about data

If you've read this far, you can see I'm enthusiastic about diving into all things data. My plan is to keep writing data-related articles on this blog, and I will have some interesting topics to share in due course.

Let's see what the next decade brings for data engineering and software development. Exciting times!