∼ The Bitwise Hero ∽

Posted on 11/6/2019 by Randall Dolifka

Reliability is a User Experience

This post is mainly intended for developers and service operators, but hopefully there is something in it for anyone who cares about reducing mistakes.

As reliability-minded developers, we are responsible for maintaining a reliable service that our customers can depend on. This usually means writing safe code, creating dependable tests, and making sensible decisions when planning our next big feature release. But sometimes, prioritizing reliability means putting on our designer hats and looking at our problems from a different perspective.

Today, let's talk about how the reliability of your service starts and ends with a user experience, intentionally or not.

For the context of this article, let's borrow this (admittedly, rather broad) definition of User Experience (UX) from Caglar Araz and the UX Collective...

User experience refers to the singular and accumulated experiences that occur for users as a consequence of them interacting with an object in a given context.

So what does that have to do with operating reliable services? Well, a lot actually.

Let's start with a bit more background. Many software developers tend to associate the terms user experience with user interface, and for good reason: most of the people with UX in their job description are working directly with user interfaces. Using our handy UX definition from above, let's imagine a common scenario...

A customer (the user) is interacting with a search component (the object) because they want to find and delete a document they had previously created (the context).

What might a good user experience look like here?

The customer wastes no time and is able to quickly and easily locate, verify, and delete their desired document using an easy-to-locate search bar, a memorable search phrase, and two clicks.

What might a bad user experience look like here?

The customer struggles to find the search component and has to ask a friend for help. Then every search they attempt brings up 15 documents that look nothing like theirs. Finally, they locate one with a title that looks familiar and accidentally delete another person's work.

So what controls the actual outcome of the user experience? Since we can't control the user or the goal they are trying to accomplish, the best way to persuade the experience in a positive direction is through design improvements to the user interface. Painting in broad strokes, this is why companies hire designers.

Back-end developers should keep in mind that this concept doesn't only apply to graphical interfaces. The lens of user experience maps directly to command line applications and API consumption as well. Like really, when was the last time you created a symbolic link without having to look up the correct ln syntax?

So then how exactly does this affect reliability? I promise we're getting there!

Let's take our example above and apply it a bit more liberally. What if we replaced the concrete artifact of a UI with a more abstract idea, such as a READ-DO checklist? This might sound a bit strange, but hear me out.

A developer (the user) is following a deployment checklist (the object) because they want to release a new feature into production (the context).

Hopefully this is starting to feel a bit more real. So what might a good user experience look like here?

The developer is able to follow a short, easy checklist that allows them to safely deploy the new feature without having to worry about silly mistakes.

You probably see where I'm going with this. Now for a bad user experience.

The developer struggles to follow a long, partially outdated checklist that leaves production in a state of disrepair while reporting a successful deployment to the developer, who is now leaving for a month-long trip to Antarctica.

Okay, I'll admit it, I am taking a bit of poetic license here; but, you see what I'm saying right? In the same way that a customer has a user experience when using a UI or command-line tool, a developer has a UX while following a checklist. And exactly the same as before, if we want to sway the user experience in a positive direction, our best bet is to improve the object of interaction: the checklist.

"What if I don't have a checklist like that?" you might ask.

"Well, you're wrong," I would reply.

All teams have checklist-like processes, steps of sequential actions that they perform when trying to accomplish a specific goal. Yours might not look like a written deployment checklist. In fact, it might not be a checklist at all, but maybe just a loose ordering of steps. It might be a short list with a few steps, or a long complex one with branching paths. It might be explicitly written out in a shared document, or it might live entirely in your head. Regardless of the form it takes, any regular process is, in its most basic form, just a checklist.

Therefore, every process used by your team has a user experience to consider.

For any one of your team's processes, consider what a good user experience might look like. What might a bad user experience look like? What aspects of your process might sway it one way or the other? What could you change about the process to tip the odds in a positive direction?

You will be amazed what you find broken when you start examining the user experience element of your team's various processes. Sometimes UX problems are purely technical, like a flaky test script or an outdated dependency causing frustration. But in many cases, the problems can have a much more human component, such as a ridiculously long command that everyone runs incorrectly, or an emergency runbook that is hidden somewhere deep in the bowels of some Confluence space. Imagine if the runbook author were camping far from civilization and the incident-response team couldn't find the document.

For illustration's sake, let's look at one UX scenario my team ran into recently. For context, we make reliability-minded tools and services that are used internally by the rest of our company.

My team kept seeing bugs appearing in our production deployments, and they were usually fairly stupid ones. The whole user experience of releasing a change was miserable because we often had to waste precious development time performing rollbacks and investigations that could have been avoided. Clearly, things needed to improve.

Our first response was to add some extra scrutiny to the code reviews that were already a part of our release process. More eyes and more time probably did improve overall code quality, but the added review didn't seem to decrease the number of bugs that crept into production.

The next idea was to add a manual testing step to our release process. One of my teammates created a runbook that guided the change developer through a comprehensive testing procedure on our staging environment. We all agreed that the runbook should help us catch most bugs before they reached production, but practice found us routinely ignoring the runbook out of impatience or frustration, rendering any of the expected benefits null and void.

"Well," we wondered, "If none of us want to run the full test, could we just have a script do the hard work?" Yes, actually. Using a cron-like scheduler, we began performing regular tests of our staging environment which exercised the "happy path" of our application. We could simply deploy our changes to the staging environment and wait for the test to pass a few times without raising alarms. If nothing broke, we could be reasonably confident that the core functionality of our service was working correctly. Would that finally fix our bug problem?

Jackpot! With alarms sounding for any core defects appearing in our staging environment, we started catching most of our bugs before production. Because we had an automated test that performed the bulk of our testing workload, we felt more comfortable digging deeper into specific focus areas and let the automation handle all of the tedious toil. And now, after our long journey, we finally had a reliable and satisfying release experience, and all was right and happy in the world!

Well, almost. Our release experience isn't quite perfect yet, as evident by the changes we tend to leave "soaking overnight" in staging. (Then, of course, we forget them in the morning like that one university roommate we don't talk to anymore). We are still designing some potential solutions that discourage this bad habit, but overall, our team is much happier with the reliability of our service today than we were three months ago.

Now it's your turn. Can you think of any tools or processes that your team regularly struggles to use effectively? Looking at them as a user experience might help you get your reliability SLIs trending in the right direction.

Not sure of a UX to start with? Here's a few questions to get your creativity flowing.

Do we have high-confidence in the changes we release to customers? Do we have any vulnerabilities or outdated dependencies from weeks ago that still haven't been fixed? Do we keep writing the same code over and over to fix the same problems again and again? Do we make it easy for new teammates to start working on our services? Do we have any sections of code that we are afraid to touch?

Changing any UX starts with a conversation. Sit down with a few teammates and ask, "How do we do X?" Try to talk through each of the steps you would perform when interacting with a tool or process. As you go, pay close attention to steps that seem confusing, prove time-consuming, or spur strong memories. At the end, review all of the steps you performed as well as any notes you may have taken, then ask, "Are we happy with how we accomplished our goal?" If not, you probably have one or more steps that need to be added, changed, or even removed.

While you are walking through a user experience, it can be helpful to keep these questions in mind...

Can anyone on the team use this tool or process, or is the knowledge limited to only a few people? Is it obvious how to use this tool or process, or will the user need some kind of outside help? Is it easy to make mistakes, or are there safeguards that help prevent them from happening? Is it straightforward to do things "the right way", or is it often easier to employ risky shortcuts? Is it obvious what your actions are doing, or do you only see the end result of a hidden sequence of events? In what kinds of situations or emotional states would users be performing these actions? How do we know that we completed the process successfully? Could a computer be doing this for us, or should it be done "by hand"?

By intentionally considering the interactions you make with a tool or process, and by asking yourself the right questions for your context, you will find ways to improve your user experience. This might make the difference between a release process people hate or one they don't have to think about at all. It could be the difference between a debugging tool that raises service quality across the company or yet another tool that nobody knows/cares about.

Return to Top