Have you read the awesome essay “Death by a thousand microservices”? I agree with everything said in it. However, I’d like to emphasise that microservices aren’t a problem in itself. Also, I don’t think monoliths are necessarily a better fit most of the time even for smaller products. In fact, personally, I prefer services (intentionally avoiding the “micro”) to a huge monolith in the work setting.
The real problem, I think, is the lack of engineering competence and the lack of “giving a shit” in companies. That’s what creates complex systems nobody can efficiently work on.
The software engineering world is incredibly hype-driven. Microservices are the current hot thing (although I think people are now ready for the next trend).
I, actually, love the fact that we’re so quick to adopt new tech. It keeps the profession interesting. If I didn’t like exploring stuff, I wouldn’t have had so much experience with completely different technology and would’ve been a much worse programmer (probably working with the same shitty russian PHP framework I started my professional path with).
However, adopting a new (groundbreaking) technology means that we turn from seniors to juniors. Yes, we apply the knowledge and techniques we’ve learned so far but sometimes it’s not very transferable. Remember when Scala had gained some popularity and people noticed that the code in it is either overly “clever” math gibberish or simply Java with a different syntax?
Companies often assume that if they hire “smart people” they’ll simply figure stuff out. That’s not how it works (also, see the “smart-ass complex” below). Being smart doesn’t guarantee that you can solve any problem. Experience is probably the most valuable asset (and the experience itself needs to be of the high quality which is a rare gem too) and we give it up when adopting new technology. We willingly make ourselves incompetent.
Of course, some people are quite knowledgeable in that stuff. But do they manage to educate and drive other people into leveraging the tech properly? Nope.
Example: Apache Kafka and event-driven systems
Nowadays there’s a surge in Google searches for “apache kafka”. Lots of companies, especially startups are using it (just search for it on LinkedIn or some job boards). Many people I’ve met get outraged (and have that condescending pity look on their faces) when I call it an event bus. “Nooo, it’s not an event bus, it’s an event streaming and storage platform”, “Sooo it’s an event bus that doesn’t get rid of the messages?”, “Nooo, you don’t understand, let me explain again”. The truth is, many companies don’t even use it for event streaming but instead use it as a message queue that they can replay if someone fucks up. In my current company it’s even used with Avro schemas literally as an alternative to GRPC with protobuff. What’s the point? Might as well use HTTP.
Now, people who introduce it often (but not always) have a pretty compelling pitch and do sound like they understand why it’s useful and how to use it efficiently. The problem is, even if that’s true, nobody they hire or manage do. Kafka wasn’t really a thing 10 years ago so where would I get the experience with it?
If someone’s used hammers and nails all their life and you give them screws and a screwdriver, they’ll simply say “oh that’s a pretty crappy hammer but the job is a job” and proceed to smashing the screw into the wood with the handle. If you want them to use it properly, invest in their education or at least handhold them until proficient. Do people do that? Not really unless it’s a very small engineering team where everyone works closely together (and even then it’s not necessarily a thing). And don’t forget teams where the tech leads have no idea what they’re doing either.
I actually love kafka-based architectures. To me it’s like a better AWS SNS (if I’m not the one setting it up and maintaining it). However, if I see it in a job description, the chances are, the company is using it wrong so it’d be a nuisance in your day-to-day job rather than a tool that helps you.
Not giving a shit
Following my example with a hammer and a screwdriver, you may ask, why not find some time yourself to learn more? After all, you can spend an hour or two a day reading something and still get paid (and it’s still useful to the company).
Well, because most people don’t give enough shit to improve. Yes, programmers are supposed to learn all the time blah-blah and all that crap people say to the newbies on Reddit. The truth is, most don’t.
There’s also another category of people. Good (sometimes brilliant, even) engineers who know exactly what they’re doing but who simply don’t care enough to do the right thing, to guide others into doing right thing. As far as they’re concerned, they have a task, they solve it, they get paid. No more extras. The company may burn in hell for all they care. I’ve met a few of those. They’re incredibly insightful when having a casual conversation but very quiet and bored in meetings.
Having a “no fucks given” culture is a disease. It poisons everything around it. At some point everyone stops trying to improve things. They simply go with the flow. Just like shit in the sewers (as my teacher used to say). That’s one of the reasons why I always recommend junior people to change jobs at least once a year. It’s better to learn bits and pieces from different places rather than getting stuck at one.
This also translates into code reviews. And decision making. And hiring. Incompetent people hire other incompetent people who then get involved in hiring. It’s exponential. I sometimes feel like we need to enforce a lockdown and develop a vaccine for this.
The “smart-ass complex”
Let’s face it, engineers, most of us aren’t smart. We’ve just been led to
believe we are. Knowing how to code or open a terminal and use
doesn’t make you smart nor educated. At best, it makes you tech-savvy adjacent.
In quite a few countries people in STEM are even considered more smarter than people studying social studies (get it? more smarter haha I sometimes crack myself up). The truth is, people are idiots no matter what their interests, hobbies, and work are. Being an engineer (even an experienced one) doesn’t make you nor your decisions smart.
I’m bringing this up because I think that having the, let’s call it “smart-ass complex”, oftentimes will make us to:
- Overcomplicate things. Let’s build a slack bot with 10 different Haskell container Lambdas on AWS talking to each other through SNS continuously deployed through Circle CI with the usage data sinking to Athena. Now we don’t have to pick the standup facilitator ourselves, it’ll be done automatically every morning at 8am!
- Attach different names to the same thing. Adapter? Translator? Facade? Delegator? Decorator? Bridge? Ah, you mean the thing we can call simply a wrapper? Yes, you could argue the important part here is the context but how about you simply explain the problem you’re trying to solve first instead of sounding like a douchebag nerd with your fancy-ass terminology? Also think about this next time you call your kafka-based system “event-driven” (is it really?).
- Seeing problems that don’t exist. Yeah, we need to scale a lot, let’s follow Twitter’s lead and use Java and Scala because we’ll definitely have the same scaling issues in the future in our niche-ass application only a few thousands professionals use. And when that future comes, we’ll be like “psych! we’re ready for you”. Actually, Java is too slow and old, let’s use Elixir. We’ve got a bunch of Ruby devs, they’ll definitely know how to build systems in it, it’s pretty much Ruby anyway just more faster and more functional-programmier.
- And so on and on. I’m getting off topic here
Not being smart is totally fine (I certainly am not one). We just need to avoid being stupid. And we need to be humble and not let our egos blind us. If we recognise we aren’t smart, we can start going back to basics by keeping it simple. If we aren’t smart enough to predict future problems, we shouldn’t even try. Instead, let’s solve the problems at hand as they come and do our best at keeping things clean and flexible (see two-way door decisions). That would be a smart decision.
P.S. Coincidentally, when writing this section I saw a question on Reddit about toxicity in the software engineering culture (unfortunately, the context was deleted by the mods). One of the points mentioned was the culture of trying to be the smartest guy in the room. That happens a lot (I have sinned so myself).
So how do we navigate this hype-driven incompetent world of software engineering? I wish I knew the answer. I’ll try to provide some thoughts tho.
We’ve all heard of the SOLID principles. However, I notice that people almost always associate it with OOP specifically and completely dismiss it when talked about in the context of other languages or system design. In fact, you wouldn’t believe how many times I was corrected (i.e. put in my place lol) at meetings whenever I mentioned SOLID outside object-oriented context. The reality is, if you think really hard about it, every principle applies as much to the whole systems as to a particular codebase.
As we often say, programming is all about splitting a problem into a smaller problems and solving them one by one. However, it’s not only about problems. The same goes for systems and organisations (see Conway’s law). System at every level (including organisational) needs to be sanely split into smaller units. And at every level a well split system units will have the characteristics defined by SOLID principles.
- Single responsibility principle. A unit should have a clear responsibility
and keep on top of it at all times.
- At the code level it means your library/module/class/function/etc should only have one purpose.
- At the system level it means the same thing for microservices. However, in practice it doesn’t matter if a service have multiple responsibilities as long as they’re provided as separate independent APIs. Basically, a service is simply an infrastructural unit/junction for APIs which are above in abstraction level.
- At organisational level that would mean the team should have a clear role. In tech teams that would mean looking after specific projects, ideally, in a specific domain (this one’s hard to achieve).
- Open-closed principle. The unit must be open to extension but closed to
modification. Personally, I also include “composition over modification” here.
- Your data structures should be backward-compatible (at both code and system level).
- Your services/APIs should allow to build on top of but closed to any modifications. This is related to the single responsibility principle. Example: a payment service should only provide money movement features. And particular product-related business logic should exist as a separate loosely-coupled API utilising the original payment service.
- At the organisational level the role of the team should not change over time. If at some point the role is deprecated, the team needs to be disbanded and a new one created with a new clear purpose. At tech teams it’s also closely related to code ownership.
- Liskov substitution principle. All subclasses should be usable instead of
their parents. This one sounds very specific to OOP. But no.
- At the code level it means that when you have two units implementing the same interface, you need to make sure both are usable in any situations the interface is required. Simply providing the same structure/methods/etc is not always enough, keep that in mind when designing, implementing, and using interfaces.
- At the system level it’s exactly the same. I mean API literally stands for “application programming interface”
- At the org level teams should not be relying on other teams but rather the work they do. Sounds pretty obvious but the communication processes are very inefficient at times. How do you normally receive work from other teams? Do they send a JIRA ticket? Do they talk to your manager or tech lead first? How does it get prioritised? How are deadlines/estimates/updates communicated?
- Interface segregation principle. Units should not depend on other units they
- At the code level it basically means that A using B should not have to use C directly just because B requires it. Simple example: I should not have to access database credentials and establish connection to it simply because I want to fetch a specific user. This should be done somewhere else (where and how exactly will depend on the exact codebase).
- At the system level, it’s pretty much the same. Example: TODO
- At the org level it’s again all about communication. I shouldn’t have to deal with irrelevant people and jump hoops to get something from another team.
- Dependency inversion principle. Units shouldn’t depend on each other,
instead they should depend on abstractions (interfaces, contracts)
- At the code level you basically don’t want to use other units directly. It’s not always practical tho e.g. in Ruby it’s usually more convenient and doesn’t really have any downsides. For instance, in a language like Java using other units directly may lead to awkward/bad tests. I guess I would reword the principle as “interface first, implementation second”; it’s hard to establish specific coding rules on this one.
- At the system level, again, the communication between services should rely on contracts because internally services tend to change. Sometimes services even get replaced.
- At the org level it’s exactly the same. Contracts over specific people.
Wow this section turned out to be much larger than I intended and still doesn’t feel thorough. Oh well.
Monolith vs services vs code
What’s the difference between having a monolith and multiple services? Communication. Consider these two related units:
module Authorisation def allowed?(user, action) check_database end end module Users def freeze(user) if !can_freeze_users?(current_user) return auth_error end do_it_mf end def can_freeze_users?(user) check_permission(user, :freeze_users) end end
Pretty normal monolithic code, right? What if I told you the units aren’t from the same codebase? The first module is actually a separate service. What changed?
The real difference between monolithic codebases and service-oriented ones is how you call the APIs.
# Monolithic Authorisation.allowed?(user, :freeze_users) # Service-oriented # makes HTTP/GRPC/etc request AuthorsationApiClient.allowed?(user.id, :freeze_users)
In an async service-oriented system the communication will be less direct and the exact implementation will depend on the context. You’ll probably need to keep a copy of user permissions locally by subscribing to some changelog.
Coming back to incompetence, one of the reasons I prefer service-oriented architectures is the fact that I’d rather own a few small services nobody outside my own team can touch so we’ll have complete control over them. I don’t trust other people so I’d rather not let them touch my stuff. When everyone owns everything (and that’s what often happens in monolithic codebases), nothing gets done.
Also, monolith supporters often say that a properly written monolith can always have stuff taken out of it easily. That is true. However, what are the chances your organisation was competent enough to write such monolith?
Think about Developer Experience
We always talk about UI/UX, customer obsession, working backwards - many companies steal those principles from Amazon. What people don’t always think about is that the customer is not always the same as business customer.
When you write code or create a project, your colleagues (and yourself) become your users too. Because they’ll work on the code you just wrote. They’ll read it; they’ll modify it. Think of ergonomics for your users.
Will it be easy to test your system/service? Is it easy to run locally? Can we spot and solve production problems easily? Do we need documentation? Is there a clear way to solve specific problems (existing code should encourage writing good code).
Keep it simple. Just because the rest of your organisation overcomplicates things, doesn’t mean you have to. When it affects system level, you should probably talk to your tech lead so they can deal with any push back. They are your best friend (or at least should be) when dealing with outsiders. sdaf
Exercise common sense at all times. And do learn to recognise and act accordingly on the one/two-way door decisions, please.
Doubling down and going all-in is bad
Just because services or monolith seem like a good idea doesn’t mean you need to go to extremities. You can always mix and much. I think people often assume “all or nothing” stance. But that’s impractical, often dangerous. We need to always exercise common sense. Also, you may have noticed how many startups have suddenly decided to switch to services. When they do so drastically, it gets out of control. It’s better to do so gradually and the initiative should be overseen by someone competent.
There’s no shame in adding extra features to your “microservice”. Just make sure the features don’t get entangled with the existing stuff. You can always extract them later if you ever need to do so (spoiler alert: YAGNI).
Single responsibility and open-closed principles - take ownership of a few services and don’t let anyone touch them. Ideally, don’t even touch them yourself. All the change/feature requests must come through a specific communication channel into your backlog. If there’re multiple urgencies, well, tough break, tell them to fuck off. No, we don’t care about your KPIs. We have our own.
If you always let them dump “urgent” stuff on you, they’ll keep doing that. Take a stand from time to time. Making compromises all the time will simply slow down the work in the future and the company won’t be able to scale (seen plenty examples of that).
Ownership is my biggest gripe in my current workplace. There’re 3 different teams working on my repos. What the actual fuck? Recently I took a look at the code and didn’t recognise some parts of it. Some changes even screwed our team over (not too much, luckily).
This creates a lot of different problems on different levels.
- Motivation. You’re more likely to give shits about stuff only your team works on. You want to keep it neat and clean. It’s yours. This goes away when the ownership is shared. An no, calling a service yours but letting other people working on it is not ownership. As I said above (although pretty sure Aristotle said it first), if everyone owns everything, nothing gets taken care of.
- Lack of communication. When there’s no proper ownership, relevant people are often not involved in the decision making enough. Their attention span is also reduced due them being involved in lots of things (sometimes irrelevant things).
- Excess of communication. Yep. Some relevant people don’t get involved enough but some completely irrelevant people get involved more often. This reduces the quality of meetings and enforces the need for more. Hence, lack of communication also leads to having to communicate too much.
- Knowledge deprecation speed. If things people work on change too often by different people, their overall understanding of the system gets affected. All the knowledge you accumulate deteriorates faster. What you know today is not true tomorrow. This isn’t true when you only need to know your parts of the system and all the changes go through you and your team.
- Inability to estimate and tight deadlines. Because everyone can make changes everywhere, the management puts deadlines and expects estimates in the context of a single team. And if the estimate date is too far in the future, they just throw more people into the problem (2 women can deliver a baby in 4.5 months). The problem here is, by working on the same stuff, people get on each others toes so it creates extra blocks and unnecessarily conversations. Also, due to the deprecated knowledge, the team can’t even make a proper estimate how much time work in a specific service would take. When there’s a clear ownership, you contact the relevant teams and they give you more or less precise estimates on their own work and potential blockers.
- Shit code snowballs. Related to motivation. Have you ever found yourself having to add more tests or an extra function to a file full of shitty code but you can’t really write yours properly because you have to follow the same pattern? Yep. Not owning stuff creates bigger snowballs. People don’t care and sometimes people are unable to confirm crappy code isn’t shitty for a reason. This, obviously, slows down future product features delivery. Also it makes it even less possible to transition to your precious “proper” microservices so you end up in the half-assed state where you tried to go all in but failed. Crappy code leads to more crappy code. Exponentially.
- Human resource costs. All that stuff above leads to having to hire more people, and more people leaving (and having to replace these). Properly organised teams will work much more efficient. It won’t matter if you hire brilliant developers or not, if you can’t manage them properly.
I now realise that because it has a butterfly/domino effect on the whole organisation and products, I could write this list for a long time. So I’ll stop.
I’ve mentioned previously that you may have brilliant architects working for you but they’re gonna be absolutely worthless if there’s nobody to bring their plans to life.
It’s not enough to tell people “we’re gonna be event-driven now”. You need to make sure they understand what that means and actually follow that. How do we accomplish this? I don’t know.
I think maybe a stricter hierarchy could help: architects govern team leads and team leads govern their own teams and report back to the architects. Engineering managers should be driving the hierarchy enforcement. Yes, sounds restrictive. But we are an organisation not a bunch of individuals, we need to work efficiently together not by ourselves.
If you don’t care about the company, care about yourself. And speak up
Honestly, I understand the “no fucks given” attitude. After all, it’s not your business at stake. But surely you don’t want to just get through the day. Surely, you don’t want to work on shitty code and doing shitty tasks for some random idiot manager who wants to hit some KPIs.
Don’t worry about the company’s success. Think about your own comfort. What could we change so you felt a bit better? Identify those problems and speak up.
Yes, it’s not easy to speak up. Sometimes you want to avoid confrontation. In such cases, find a person you can talk to about this. The person, who doesn’t avoid confrontation. Who can try and make this happen. It may be your manager. Or the team lead. Or even someone irrelevant to you or your team. But you need to talk to someone about this. If you don’t, everything will stay the same.
As they say on the London underground: see it, say it, sorted.
I understand it’s a harsh title and at times a harsh essay. Such is life. All opinions are my own.
I also feel like I made it sound like all the problems come from stupid engineers. Not at all. On average engineers are, well, average. So all those people on LinkedIn quitting their jobs and saying how smart the people they had worked with are make me cringe.
No, by incompetence I meant incompetence across organisations. And the biggest damage is done when incompetent people are leading them. Inefficient management leads to inefficient workers. Software engineering is easy (tell all the elitist gatekeepers to fuck right off). It’s a simple craft not much different from locksmithing. There’s no romance in it. You don’t need to be a genius to be proficient in it. Hell, you don’t even need education. But if your organisation is fucked up, your work will be too. A fish rots from the head. So we need to do fix problems from the top not the bottom.