GOTO - Today, Tomorrow and the Future

Modern Software Practices in a Legacy System • Audun Strand & Kevlin Henney

June 10, 2022 Audun Strand, Kevlin Henney & GOTO Season 2 Episode 22
GOTO - Today, Tomorrow and the Future
Modern Software Practices in a Legacy System • Audun Strand & Kevlin Henney
Show Notes Transcript Chapter Markers

This interview was recorded for GOTO Unscripted at CodeNode in London.
gotopia.tech

Read the full transcription of this interview here

Audun Fauchald Strand - Principal Engineer at NAV
Kevlin Henney - Consultant, Programmer, Keynote Speaker, Technologist, Trainer & Writer

DESCRIPTION
Audun Fauchald Strand, principal engineer at NAV (Norwegian Labour and Welfare Administration), shares the story of how they modernized their long-running open source platform NAIS.
In their conversation, Audun and Kevlin Henney touch on topics such as how to organize massive amounts of data and if legacy code can be considered as something good. You’ll also find out how open source fits in a large governmental software team that focuses on transparency.

RECOMMENDED BOOKS
Kevlin Henney & Trisha Gee • 97 Things Every Java Programmer Should Know
Kevlin Henney • 97 Things Every Programmer Should Know
Henney & Monson-Haefel • 97 Things Every Software Architect Should Know
Henney, Buschmann & Schmidt • Pattern-Oriented Software Architecture Volume 5
Eric Evans • Domain-Driven Design
Zhamak Dehghani • Data Mesh
Sam Newman • Monolith to Microservices
Sam Newman • Building Microservices
Ronnie Mitra & Irakli Nadareishvili • Microservices: Up and Running
Mitra, Nadareishvili, McLarty & Amundsen • Microservice Architecture
Rishu Mehra • What is Data Observability

Twitter
LinkedIn
Facebook

Looking for a unique learning experience?
Attend the next GOTO conference near you! Get your ticket at gotopia.tech

Dev Interrupted
What the smartest minds in engineering are thinking about, working on and investing in.

Listen on: Apple Podcasts   Spotify

Catalyst 360: Health, Wellness & Performance!
Your trusted resource for engaging, evidence-based health, wellness & performance

Listen on: Apple Podcasts   Spotify

Twitter
LinkedIn
Facebook

Looking for a unique learning experience?
Attend the next GOTO conference near you! Get your ticket: gotopia.tech

SUBSCRIBE TO OUR YOUTUBE CHANNEL - new videos posted almost daily

Intro

Kevlin Henney: Hello, and good morning, good afternoon, good evening, wherever you are, wherever you are. Welcome to another GOTO Unscripted conversation. My name is Kevlin Henney and I'm here with Audun Strand. If you'd like to say something, introduce yourself.

Audun Strand: Sure. As Kevlin said, my name is Audun Strand. My title is fundamental of principle engineer, but I work at NAV, which is basically a big government agency in Norway. It's I think in almost every other country. NAV (Norwegian Labour and Welfare Administration) does what 15 or 16 different organization does. So we're big and we say that we support Norwegians from before they're born till after they're dead. So that's kind of the scope.

Kevlin Henney: So that is a full life cycle. That is full life cycle support.

Audun Strand: Your parents get money and if you can't afford your own funeral, we support that as well. So it's basically everything you can imagine doing. I've been a developer for 15, 16 years, mostly, I realized, I think when I was working for six or seven years that the funniest thing is to solve problems for the developers instead of... So basically, I've been taking all my bitterness as a frustrated back-end developer to make application platforms and trying to make it easier for people or developers who make products to help them instead of trying to figure out what normal people like because I've never... I know what developers like. I don't understand what normal people like.

Kevlin Henney: So in other words, developers as clients.

Audun Strand: Yes.

Kevlin Henney: I like that idea that you kind of take the bitterness that you have got. I think many people have that and it's just like, "Yes, but it should be done like this. This would be a much better way," and actually turn it into something constructive. I think that's kind of almost the basis of a lot of open-source development is just "I've always wanted this," or "This should be better. Well, I can do it." And they do it. As I understand it, you've been working on a platform that's open-source itself.

Open-source platform NAIS by NAV

Audun Strand: Yes, I started to work on platforms when I worked at FINN, the former company, which is Norway's...at least for some ways of measuring, it's Norway's biggest website, and I made an application platform there. Then NAV used to be...we used to think of it as the worst place, at least in Norway, to make software. It was all consultants and they made a strategic decision not to hire developers. They only use Accenture and Capgemini and everything, and architects, of course. Then they've got this new CTO, and he said, "We need to in-source and we need to hire developers." So my friend Truls was the first one and I was like the sixth or seventh. Now we have 400 developers. So we kind of built everything from inside a really large...coming through a recession, we built a new software department.

Kevlin Henney: That's really cool.

Audun Strand: The first thing I did was build an application platform and try to use that as a tool to make people understand that it can actually be...first of all, it can have normal, even good software in the government and the public sector. It's kind of cool to work at now. We open-source it because it's the one thing that no one can argue about open-sourcing is basically just its go code talking to Kubernetes and Docker, so it's not in any state secrets. We use that to build our brand basically. The platform is called the NAIS. We found a funny word first and then we retrofitted it. So it means NAV Infrastructure Service or System. NAV Application Infrastructure, and you can choose between service and system because it doesn't really matter.

Kevlin Henney: Because it doesn't matter. Because you made up NAIS first. It's like, "Right, what can we do to fit that?" That's really interesting. It's kind of a transformation story because I think what you described is a common experience for many people in many different countries, is the idea that government software is not the place that you want to work, whereas logically we should be thinking that that's the place where we want the highest quality. If you think about it from a social responsibility perspective, it's sure they should have the best software.

Audun Strand: There are many other things. Like, there's really no competition, and the scope is to have some mainframe stuff, the old COBOL stuff, and have the strangest Oracle thing you probably never heard of this. It's what Oracle had before Oracle and IBM invented application servers. The instances of the Java objects live inside the Oracle database, so it's not like stored procedures.

Kevlin Henney: I remember that.

Audun Strand: To develop this, you have to have your own instance of Oracle database running on your own machine, and nobody wants that. So we have 40-year-old systems that have been important for Norway for that time. And to realize I'm never ever going to write any software that's going to be important for 40 years. I understand that's what we're going to replace. It's both scary, but it's also nice.

Changing governmental software

Kevlin Henney: So that's really interesting. I think one of the other things people associate with government software is, obviously, there's the whole legacy. There are systems that date back decades that we know little about, and yet a lot of people rely on them. But also, you're dealing with ridiculous amounts, and particularly in modern society, and the kind of the amount of data that is associated with any individual in any society, you're dealing with ridiculous amounts of data. There are all these little domains and you've got all these different systems, which I'm presuming don't necessarily talk to each other comfortably. I presume we want one view of the data or to have a coherent data flow.

Audun Strand: Yes, very much, not very good connectivity for that. The first step was basically just to make it sustainable to be able to make a system that is changeable for more than the project. That, of course, takes a bit of time, because that means you need to build teams and you need to build an organization before you start to build systems. And then, after that, we tried to build the system around the domains, as I said, so we have something for the sickness benefit and something for the parent key benefits and age pensions and everything, and then trying to figure out good ways of making it possible for these applications and these teams to change in their own pace but still being able to communicate in between. So trying to have the loose coupling, as everybody talks about, and share just enough but not too much data.

Kevlin Henney: I think that's one thing that when people are working within companies that have a single focus, they might not appreciate. And likewise, anybody who's worked in a startup culture where you get to choose your focus. But here, you're actually dealing with lots of things that are pulling you in different directions, and have done for decades before you were there and long after you were there. Exactly as you said, this is an interesting kind of compromise. You need looseness to allow change. But you also need to have something that is together enough.

So giving the migration, the journey that you've kind of taken on. You mentioned Kubernetes, you mentioned Docker, and you mentioned creating the NAIS platform. You also mentioned actually that NAV covers traditionally what in other governments will be 15 or 16 departments. That's a lot of...I don't know, is there a term for it, domain share. It's just one unified view. There isn't one unified. There isn't a single coherent model for everything.

Audun Strand: No, no, there used to be, because that was the old way of doing it. People realized that doesn't really work because it kind of blocks everything. There's this old saying that...I don't remember the English one, "When two people live together, then they're not married." In Norwegian, I don't know what that means in English. Anyway, people say there are, 14 different definitions of that word in our systems. The funny thing is that they think, "Well, then we need one definition." Then we come in and say, "Well, we don't because it means 14 different things." So instead we need to understand that it means 14 different things. And whenever two systems talk together and have a different understanding of that term, then they need to figure out what it means. But to try to make this big model of one specific meaning of every word is almost impossible, and they're as big as we are.

Kevlin Henney: I think that's a kind of...there's an interesting collision between what we desire, particularly from a developer perspective"Wouldn't it be nice if everything could be...you know, we can have one definition of this?" And what the real world does. I think even in the real world, we're not very good at acknowledging this, that there are actually...that words have context and specific meaning, and they also drift and change in their own way.

Audun Strand: Yes.. Because it feels like common sense to have one specific definition. But if you take that over time and say, "I'm going to keep that model over 10 years, and everything is going to change at different paces." It's almost impossible to keep that.

Domain-driven design vs the law

Kevlin Henney: I think I first experienced that trying to talk to somebody about kind of a domain model, actually, it was an entity model, specifically what they refer to it in a medical environment in the 1990s. What does patient mean? And it turns out, that different parts of the medical space have different meanings and different information requirements for the word patient. That's just what most people think of as one domain, and actually, it has multiple subdomains. The minute you move outside that, then suddenly you get all of these difficulties. So I think that kind of takes us into the era of domain-driven design, which I know that's something you're interested in. And it seems to really apply here.

Audun Strand: Yeah, definitely, I think I've spoken about it. I think it's chapter 14 in the big blue book about strategic design and how we can have different models connecting and customer supply and everything. I'm trying to explain that too, for instance, the legal experts. Because another funny thing...because everything we do is kind of grounded in the law, so the benefits are different chapters in the law. There's the same thing there. It's reuse. So the parental benefit and the sickness benefit points to the same chapter saying we're going to calculate the amount from this. Then everybody says, "Well, then this has to be some common service, of course, because it's the same chapter." And then when you try to implement it, we see that "Well, it doesn't." But it feels like the law has if statements.

So we easily figure it's not exactly the same. The contexts are extremely different, because parental benefits, that's something you enjoy. That's a good thing that happens to you, and the sickness benefit, it's something bad that happens to you. So trying to make that into one system doesn't really make sense, even though it shares the same chapter of the law book. So trying to...and that's kind of the domain-driven design kind of gone further because it's not enough to think about the technical side but you have to think about the legal side and the communication side and everything.

Kevlin Henney: I think that's an interesting one because you mentioned of "The Bluebook", you know, it's the strategic design chapter, which it's...Eric, said, "That's the bit that people always ignored." A lot of people never got that far into the book. And yet, I think in recent years, it's an area...people normally grab onto the early part of the book, and the patterns in the book seem very intuitive from a developer perspective. But this other stuff seemed is not immediately obvious. But things like bounded context have become very popular in recent years, and this idea of saying, "Well, actually, we don't have to have a uniform..." Although we use the term ubiquitous language, it's not actually uniform and universal, it's the idea of little local subdomains. And as you say, in law, it is... the law has a lot of subtleties to it, speaking as a developer, and I've read a couple of things in law and no shortage of legal documents. I'm always surprised at how loose they seem compared to code. It's quite interesting. You get used to code. And people say, "Oh, legal documents are so boring." It's just, like, have you tried reading code? Legal documentation is actually surprisingly imprecise, or they rely on a lot of contexts that are unspoken.

Audun Strand: I think it's also because the code is not...at least, normally, it's much faster to change. It's probably a good thing that laws take a bit of time to kind of evolve. It's easy to change the code. But that's why they think they can't define everything, because when the world changed, when there's a pandemic, or whatever, they have to change everything, and that's too slow of a process.

How to organize data in the back-end for a complex system

Kevlin Henney: That kind of sets us up with the kind of really interesting thing that you've got a number of different teams. You've got a number of different teams, a number of different software requirements, and dividing them. That kind of leads to very different data flows, and different applications, but at the same time, you're still trying to get something coherent. How does the data look at the back-end? How are you organizing that? Because you're talking about there's COBOL, there's Oracle, there's...

Audun Strand: Almost all the new stuff we write is Kotlin and JVM. For communication parts, I would say we kind of divide at NAV to what we call product areas, which is we have about 10 I think. We might have 15 when we're done, or not done, but then we kind of reached our limit. Then we try to say that between those, we should have asynchronous communication streams, so we use Kafka, but that doesn't really matter. So you should communicate events and then you should listen to events and that kind of makes it easier to achieve the loose coupling and easier to evolve differently. That makes the teams, at least some of the teams, that make our most complicated systems become more event-driven as well. And that's really good. My experience is that when you go from a monolithic approach to microservices, and until you kind of realize that you need to be asynchronous, you won't get all the benefits of dividing up and making microservices.

Kevlin Henney: So I think that's kind of an interesting one because I think people often overlook the fact that synchronous is coupling in time, whereas, people often look at coupling as a very immediate, "This box depends on this box. This service depends on this, or this class depends on this." And they think of it as a very immediate thing, whereas actually, there's a temporal, there's a coupling in time, and asynchrony allows you to just let the parts move at a slightly different rate and execute at a different rate.

Audun Strand: When you kind of go from events to streams. So we have the Kafka streams where we have the events stored for some time. We can play them back as well. And then, you realize you kind of have a superpower because then you can...when you have immutable data, so you can always understand that what I read is never going to change. When you play everything back, you can kind of...you can change some rules and you can play back and you can get a new reality afterward, and then that makes it even better to be asynchronous.

Kevlin Henney: So that's kind of interesting, because, in other words, you get to manipulate time, which I think is everybody's kind of fantasy. "If only I could be a time lord." But there's another interesting thing. From the traditional perspective, there's a kind of an inversion of dependencies, and I mean that in the deepest sense as opposed to the way that people often mean it, but often when people have organized large systems, they have looked at it in terms of we are organizing it around the services, and the services are the first-class concept, or they've organized it around the entities, which goes back to our problem with, you know, how do we have a unified entity model. But here we're saying actually, no, events are the first-class citizen, and then we hang everything off that. So as long as we have a truth about events, then actually we can actually loosen how we build the applications.

Audun Strand: We had a guy called Fred George. He calls himself the grandfather of microservices.

Kevlin Henney: I will say I first learned about microservices by attending a Fred George talk.

Audun Strand: Yes, me too. He has this actually, this is a GOTO conference. And it started because me and my colleague, we wanted to go to Noma, which was the best restaurant in the world in Copenhagen. Of course, we were working at software companies. We tried to find a conference that made it possible for us to go to Copenhagen.

Kevlin Henney: Another version of dependencies. Yeah, we want this restaurant. What conference will allow us to go to that restaurant?

Audun Strand: Then we went to GOTO in Copenhagen, and we had Fred George's microservice workshop. Actually, and I didn't realize this till, after this, Eric Evans was a participant in that workshop. But in that workshop, which I think is the best workshop I've ever done, because you get to code, six hours out of an eight-hour workshop on actual problems, not just fill in these tests, but you get to solve problems. He has this pattern, which he calls Rapids and Rivers, which is basically a way of organizing microservices that makes it easy for developers to think asynchronously. So what you do is you have...every single service publishes an event on the same topic. Then that's the rapid. Rapid is the big thing where everything flows. And then the services create rivers, which are basically filtered views of the rapid.

So you get a tiny bit of the event flow into our service and then do something and then we put it back on the rapid. That's basically the core pattern of almost all the new systems we build now. It kind of cheats, it makes a shared database that works inside the domain, and also, it takes away the doubt developers feel, "Should I be asynchronous or should I be synchronous?" Because it's natural to do it asynchronously. So everything kind of turns into events. Then you get the benefits of it without people thinking, "Well, this is probably synchronous," and then you stop and then you start doing it differently.

Kevlin Henney: Well, I think there's a really interesting metaphor, again, with the whole thing to do with. It's rivers, that's basically the right rapid. Everything that happens goes there.Then the rivers are the read layer for everyone. Then ultimately, the storage is in the ponds.

People were talking about data lakes. So this is a kind of interesting idea the persistence is in ponds, which are smaller and local. You maybe have a pond per microservice, so to speak, rather than this lake of a single, coherent, and centrally-managed data model.

Audun Strand: And it's so easy to keep separation of concerns. If you want to add observability, you just add another service that listens to all events on the rapid and does it separately. You can really add logic where it belongs, although the problem might be that you get too many services and it'd be difficult to see, then you just add another service to add observability, and then you can see again, that it's…

Kevlin Henney: So it's very composable by having these separations and actually saying that if we're allowed to loosen time a little bit, that it allows all of these possibilities and I think that's quite an interesting kind of observation. And I hadn't come across this one too recently and didn't realize that it actually goes back to when I first saw Fred George just after when I first saw him. He was talking about programmer anarchy, about the way he organized teams. That's where I first encountered microservices, and that's like 2012. And he had a very strict definition of how he thought about microservices. And he did do it on size. We seemed to have given up the...I think people have given up the idea of... You know, it's just like, you know, we ask people. It's like, "Well, why are they micro?" Well, let me show you this micro. That's not micro in my book.

Audun Strand: The only reason we can't call them services, is because then you go back to, this whole thing.

Kevlin Henney: Except that, I think modern services are now bigger than the stuff that we had under...

Audun Strand: So we should probably just call them services.

Microservices vs CQRS

Kevlin Henney: Time is cyclic. I think that that's sort of an interesting observation there. And I hadn't realized that there was this broader view. How does that relate to things? I don't know, because this idea of having different read and write sources, introducing a little bit of asynchronicity to allow the kind of like...the parts to move more freely. This is sort of on a parallel brush, CQRS. They seemed to have a kind of a shared goal if you like. They have very different histories, but they appear to have this kind of shared goal. How would you compare them?

Audun Strand: I kind of agree. We've never really had CQRS as a big thing because it kind of comes almost... Well, a few of our teams managed to think more like that because they have the big...some of them have really big microservices where they kind of modeled the lower part where they had the model kind of, and a whole person's sickness history to be able to, and that becomes quite a big object-oriented model. There, they try to have the CQRS. So they have kind of the events and the commands coming in. It feels, at least from my perspective, it feels less like something that's on the developer's minds than it used to be

Kevlin Henney: It was interesting because that's...I think that one is more...I found originally that was more intuitive because it was closer to something that I had done, but it was not exactly the same, and seeing this rapids, rivers, ponds kind of approach, I thought, "Ah, that's a really interesting, different way of looking at it."

Audun Strand: So the teams don't really do that much. For multiple reasons.

Kevlin Henney: They live within it, as it were. In other words, the architecture is around them. So in that sense, it seems more like that when we normally talk about architecture, people like to use building metaphors. But this feels much more like urban architecture. The developers stay in their own houses and they do what they want in their own houses. But how the streets and the services are organized, that's really what you're talking about at that level.

Audun Strand:I think it's important too when you have this Kafka topic that you just keep the idea of what data on the inside versus with data on the outside. So the rapid is only for one domain. It's for one team, for one thing, and then if you want to share data with others, you need to make a different topic. And at least before I had the really bad problems where people thought that every event was available to everybody. So we kind of forgot everything about public and private, and just because it was an event and the service, it doesn't mean you can't...you still have to keep your boundaries.

Kevlin Henney: Again, there's this really interesting tension, isn't there? There's this kind of idea, but yeah we still have to have some separations, but at the same time, we want to get rid of separations. And these two are pulling in opposite directions. So I think software architecture...

Audun Strand: When there are different teams you have to change slower than when you're just exchanging data with the guy sitting next to you.

The languages of the NAIS platform

Kevlin Henney: It seems like gear matching in manual-driven cars, cutting it just right. So I think that's an interesting kind of way of looking at it. We kind of take the big picture. Now, you mentioned a couple of technologies. You also mentioned a couple of programming languages. So you mentioned that the NAIS platform is being developed in Go, but you also mentioned from...I think the application developers are using primarily Kotlin. So you've got to kind of Go lower level, and then the JVM up a level but with a bias towards Kotlin. Just out of interest, what informs those decisions?

Audun Strand: That's a good question. I think from a platform perspective, Go has the best libraries I think, at least had four years ago when they started doing this and then it kind of grew from there. Kubernetes is made in Go. Docker is made in Go, so it was easy to start building that in Go and it also feels. I don't know if you've written those before, but I find Go to be a frustrating language to write and a good language to read.

Kevlin Henney: I don't have deep experience. I consider myself a Go tourist. There is a certain frustration because the language has certain limitations in certain cases. And familiar as I am with lots of language models, it's kind of, like, "Oh, I wouldn't have put that limitation there." But that readability thing does make a difference.

Audun Strand: So it feels like... Well, at least in the start, we thought that maybe the application platform would be slower moving, but it turns out that because the team is very good, they can also manage to move very quickly. Then, from the application side, we normally say that we have...the language we have the most code in is Java, but we write most Kotlin. And I never, at least from my...I've never learned a language as quickly as I did when I went from Java. The bitterness comes from Java.

Kevlin Henney: Okay, that we completed the story. Where did the bitterness come from? Now, I understand.

Audun Strand: And then to go on to Kotlin was quite as an easy transformation I've ever had I think going from one programming language to another.

Kevlin Henney: So it's kind of interesting because I know a number of people who...their reason for looking at Kotlin is Androids, but that's not the reason here.

Audun Strand: No, no, we have no native apps at all. I can't totally figure out who started. But it became a movement, I think, and we really...one of our philosophies when it comes to technology choices in NAV is that everything is allowed, as long as you broadcast it to the company. So we have our own internal technology radar to kind of persist all those tracers and make it available for all the teams to see. So it started out in some corners and people saw that this actually works. And then some libraries, the library we built around rapid observers is Kotlin, so it kind of...it grew from there. But we still allow a few people to write Spring and Java as well.

Kevlin Henney: Just to keep them on board.

Audun Strand: Yes.

Building a developer culture

Kevlin Henney: So there's an interesting thing here. Let me just pick up on the kind of the theme of...it just occurred to me. So, in that case, this idea of building a developer culture. And also that NAIS is open-source. There's a lot of this idea of like, "Well, you know, as long as you broadcast, as long as you make this information available," and that seems to be a very...again, it's another inversion of how sometimes people try and get things done. In other words, .it's not a sort of a command and control. It's very much a ground-up kind of approach. And I guess leading by example, "Here's the thing I've built." That team is using this, and you can see the examples there, which kind of makes. I wish more organizations would be aware of that, but it's very interesting to hear it from within a government organization.

Audun Strand: I think we managed to build a really good culture around that because it kind of...it makes it seem responsible as well. We respect the fact that they know best what tools they can use, or what they say they can use. But also, we need to be able to see what we're doing. Before we had a big problem with shadow IT where people used stuff but nobody said anything, specifically for government organizations. We can't use stuff without paying for it. When it's in the shadows, no one will say, "I'm using it," so no one will pay for it. So to be able to get everything out in the light, it helps us also to follow the regulations. As I keep saying to people, software developers should be the first ones to voluntarily pay for software they use because it's basically it's our business. Right now, we don't work for a tools company, but you might sometime later, and then it's good to be able to know I paid for the stuff I used before.

Kevlin Henney: That's an interesting one. But also the point you make about shadow IT I think it's very common in a lot of government-related work. And I know somebody who worked, I don't know which bit of the Norwegian government, but it was many years ago. So definitely, you know, it was in the 2000s. It was interesting hearing what they described as what they're supposed to be doing what they're actually doing to work around what they're supposed to be doing. In other words, a lot of games rather than something that was spoken.

Audun Strand: I think, before we had the tech radar, we had this list of allowed technologists, and I think you had some version on there, at least four years after everybody started using it. And that kind of shows it's almost impossible to have such a list centrally managed and up to date. So they're saying, "That's not possible. It's better for you to do your own choices and broadcast it and it actually scales."

Kevlin Henney: Again, interesting enough, it goes back to this idea of the cost of centralized locking. It's the same in development as it is in the runtime, that there is a huge cost to try and keep everything synchronous. So the scale that you're talking about, is asynchronicity both in execution but also actually in terms of how teams evolve and are organized.

How does data mesh fit in?

Kevlin Henney: Now, one other area, which I think is interesting, and I mentioned in the past is data mesh stuff, because that's a kind of a more recent trend. Again, it seems to fit with everything that you're saying, the idea that perhaps we don't try and have a single source of truth that is a physical single source of truth, that actually what we should...I guess data mesh is for data as microservices are to model lists.

Audun Strand: The first time I read the article on martinfowler.com I thought, "Well, this is basically someone that managed to articulate their domain-driven-design thinking but with analytical data in mind." Then I spent almost a year being frustrated, for reasons which I now understand more. All the data, people can see that. For me, it was very clear. Well, this basically summarized what I've been doing in the application space for 5 to 10 years. Someone has actually explained to you how it works in data. It's a bigger change for them because they're very tool-driven, this very monolithic, the central data warehouse is like...you can't be more monolithic than that, I think.

So to be able to find ways to explain and maneuver our organization into thinking more about splitting their ownership and taking the analytical creativity, and the analytical minds into the teams so that they're closer to where the data originates so they stop being frustrated because the thing that gets data up to date but doesn't work because the database is changing. Before we had like a few really, really big databases that changed four times a year, and being the central database team, it was difficult already then. But now, we have 2000 databases under change daily. It's almost impossible to follow through. So trying to make everybody think and understand, how should we think about getting the analytical data and the data products now when architectural applications are changing?

Kevlin Henney: I think that's a really interesting way of looking, but you also highlight something else. There's a difference in sometimes cultures. In other words, you talked about the sort of the data people, and we had the application people, and they pull in slightly different directions, and there are these different movements within each. It's like the data world is kind of catching up. We rely so heavily on data. The modern government relies on data.

Audun Strand: All the things we can get out of these analytical minds and what they think about when they decide on data models and understand other uses of data than just defecting something from an API and comparing it and throwing it away, but to understand that the data is actually even more valuable than the codes most of the time.

Kevlin Henney: Yes, indeed, I think that's one of those things that we always recognize, but, ultimately, it's the data that will outlive all the applications. I think in the past, that has meant that people have become complacent and they've assumed that because of that, they won't touch the data, or we can't change the schema, we can't do anything new. The data's on a different geological period. The applications move at this speed and then everything else moves much more slowly. This is really kind of like bringing data in at the same speed.

Audun Strand: Because there's so much value as well, and trying to figure out... Sometimes we have been able to build the teams where the developer or the application developers and the data engineers and analytical people have worked in the same team and to see the value they get from when the team wonders about should I do this, or should I do that. They can just do some SQL magic and then two seconds later, he knows well, 45% of it is this and 35% is this, so you should do this. And to have that ability really causes real development.

Kevlin Henney: So again, there's a kind of interesting thing here about time and turnaround time, and responsiveness, but also, at the same time, loosen the coupling, allow so I presume that looking to the future, that you're anticipating that NAV was gonna continue growing, but also, it's kind of the software perspective. That they're also all of the products that you're dealing with are also gonna be growing.

Observability for the legal system

Audun Strand: Well, we still have over four biggest legacy systems. We only really managed to modernize most of one of them. We started in the second one and then we have slight plans for the third one and really no plans at all for the fourth one yet. We have loads of modernization. Well, I'm looking forward to what happens when we're kind of done with the modernization when we get to proper software products that can evolve with its contents. For instance, just having analysis and feedback going back to the people making the law.  We have a really cool thing now where we managed to, in the code, that evaluates a case, it connects it to the different paragraphs and the different sentences in the law, and actually puts it into the...actually, it puts each evaluation of a sentence on a Kafka topic and presents it to another database. So we can do an analysis by saying, "Well, these two sentences in the law always go the same way at the same time. So maybe we don't need both. Maybe we may take one away, or..." This is never evaluated at all. So being able to really get data-driven on how to make laws.

Kevlin Henney: That's really interesting. Observability for the legal system. I had not thought about that.

Audun Strand: Then we can do simulations that politicians are gonna say, "Well, what happens if we do this," well, at least someone in the future will say, "Well, if we did this five years ago, this would have been the consequences," at least for the one part of the system.

Kevlin Henney: It gets back to this idea of replaying time. Just what are the consequences of doing that? That's incredibly powerful. That’s a feedback loop that many people probably wouldn't even imagine having. I think that the point there is that the fact that is now a possibility is itself exciting.

Audun Strand: So that's what kind of drives us on. We just need to finish all this, get away from all these mainframes and Oracle systems and try to get evolvable software that can change.

Legacy in software development

Kevlin Henney: But I think that was that idea that information should be as...you should be able to have the information that you need, rather than... I think a phrase that I found myself using a lot is that we spend a lot of time programming in the past. In other words, you know, we might be sitting there going like, "Here's the code I see in my ID," but actually, when I look at the code, it's actually 5 years old, 10 years old, 20 years old. I'm talking to a database that is maybe 30 or 40 years old, or that, you know... Although we think of ourselves as programming in the present, actually a lot of the things are the decisions that are in the past and not necessarily the ones we want today. And we feel powerless to change them.

Audun Strand: But, at the same time, we always try to think of the code we have now as the code that survived, and for a reason. So legacy, in our terms, normally is kind of a bad word. But in normal English, at least legacy is a good thing. So we always try to focus on the fact that this system has been important for 40 years. That's a good thing. We need to change it now, but that doesn't mean it's been bad for 40 years. It's been really good and it's built the world's best welfare state. So to be able to treat it with respect at the same time they're saying, "Well, now we need to change them because now we have other options."

Kevlin Henney: I think that's a very interesting way of looking because we do. Legacy is a funny word, where it's gotten used. I discovered a Twitter comment and then I sort of started digging around, I quite like language. So I actually dug into the Oxford English Dictionary. And the earliest citation for using legacy in reference to software was 1989, which actually feels more recent, but in language terms, it feels very recent. What I found actually fascinating is that it got taken up very, very quickly, because it clearly fulfilled a need. Then we've also got this other use of the word. In other words, that legacy we'd like to think of as positive, but it isn't always.

But I think one of the more interesting things is:  Legacy code has a history, which means you can tell something about, which you can't tell about new code. This is a new code. We don't yet know how good it is or, you know, how relevant it is.

Audun Strand: Does it solve the right problem?

Kevlin Henney: Right. Does it solve the right problem, or if we solved the problem in our head, or the one in the world, which, you know, I think a lot of developers might be guilty of, and I know I have been. But that idea of the...this idea, it has a history that we can actually go and look at and go, "Actually this is useful, or actually it hasn't been useful for a while. We just haven't noticed, whereas new code is experimental and should be experimental." We should be allowed to throw it away.

Audun Strand: Yes, because that's our superpower. We can change software on a whim. I remember reading an interview with one of the guys who worked at NAV from the very beginning. He was retiring for years, actually, basically working 40 years on the same system. And he kind of divided everything into decades. The first decade was quite good. And then the second decade came and then some politicians say it's bad, but the second was better again. And just to have that time span of understanding how the system you work on affects the real world and the other way around.

Kevlin Henney: I think it's sometimes when people properly appreciate software architecture and development organizations, they start talking, "Oh, yeah, it's a socio-technical system," and you start to see it with these kinds of long views. But I think that idea, as you say, is the idea that in software development, we have a superpower. Sometimes we don't use it. You know, that software should be soft. We should be able to change it and we should be able to do so as quickly as possible. We should also be able to understand it in a bigger historical context. And that would be...you know, it's a superpower in that sense if we get it right.

Outro

Kevlin Henney: So I'm gonna bring this discussion to a close. We've kind of covered a remarkable amount of space. What kind of things that you think are interesting from a personal perspective, but also, in the context of NAV and NAIS. Over the next few years, where you kind of...you've already mentioned a couple of things you're kind of looking forward to.

Audun Strand: One thing we're working on right now is continuing our migration to the cloud and trying to do that in a way that actually makes everything better and cheaper and everything. And another really interesting thing about me and a colleague has been working on lately is to try to balance the creative autonomy of teams with kind of the large-scale efficiency alignment to try to figure out a way to control the teams without actually making them feel like they're being controlled.

Kevlin Henney: Control without control.

Audun Strand: Yes, and we've been working quite a lot. Technology radar is one example of one technique you can use to do that. And then we have some other techniques where we try to balance the normative and the descriptive parts somewhat. We spend some time saying, "This is a good way of doing stuff." And at the same time saying, "But you can choose yourself. Everything is allowed if broadcast it." And trying to find that balance so that a system of teams that are more than 100 can both be similar, but also tried to figure out what's the best way of moving forward, because we don't want to create a new system that's impossible to change because it's 40 years old. It has to keep being changeable.

Kevlin Henney: Yeah, you don't want to create somebody else's bad story. That's brilliant. Thank you very much, Audun.

Audun Strand: Thank you.

Intro
The languages of the NAIS platform
Open-source platform NAIS by NAV
Changing governmental software
Domain-driven design vs the law
(Cont.) Domain-driven design vs the law
How to organize data for complex systems
Microservices vs CQRS
The languages of the NAIS platform
Building a developer culture
(Cont.) Building a developer culture
How does Data Mesh fit in?
Observability for the legal system
Legacy in software development
Outro