Domain-Driven Design
verraes.net
verraes.net
When I joined, my team had been building the backend for the first version of our app for about 4 months. I would describe the state of the code base when I joined as an 'Anemic Domain Model' as defined by Martin Fowler
- There was a 'domain model' in the loosest possible sense - Each model type was just a POJO with raw getters/setters for each field - Almost all fields were primitives (mostly strings with a few int/doubles/dates) - All validation code for these types was done in application code and was fragmented throughout the code base - Internal DB identifiers were fully exposed into the code model - Internal service types were liberally mixed with external service types - No notion of aggregate roots - every entity was just accessed ad-hoc
It was a highly unsustainable approach, and one of the first thing I did was attempt to implement strategic DDD in the areas that were the most painful. This included
- Adopting rich value objects to represent domain concepts instead of raw strings - Enforcing business invariants inside the model classes - Enriching the domain entities with methods that matched business behavior and performed validation - Creation of repositories that shifted much of the persistence details out of the application code - Defining aggregates based on our required access patterns which simplified our data access - Bounded contexts for our internal domain - mapping external service types to our own internal representation
The result of all this was the creation of a core 'domain model' that captured business behavior expected of our service and most importantly, ended up significantly simplifying the rest of the application code. If DDD makes your app code more complex, you're doing something wrong.
This makes intuitive sense to me: it’s much easier to define a principle that applies in many situations but much harder to define concrete technical implementations that do.
In your situation it sounds like the multiplier you applied was in pushing your team towards a rich domain. That’s a much more sensible approach than, say, using CQRS everywhere.
If an experienced developer can barely understand you, there's a communication problem.
Then, I started working as an AI Software Engineer (mix of a software engineer + devops + data scientist), and it all clicked. DDD is a wonderful design pattern for anything related to Data Science, AI, ML. Why ? because 90% of your problems is retroactively making sense, organizing, sorting, filtering, aggregating all the data you got from your favorite Data Base/Lake/Wharehouse. DDD let you have a unified language, invariant definition and expectations between your existing business challenges and the analytics your are running on it. It's very good for validating assumptions accross a dataset, for example: Sales amount can never been < 0 ? Let's check that... Oh well, you forgot about returns, so now you can define them and be explicit when to include or exclude them.
The ubiquitous language alone, is highly valuable.
Even if you dislike the ideas about software architecture that DDD introduces, having a clear, broadly carried, living dictionary, helps a lot! No more EntityTreeSectionBatchBuilder, but CompanyDivisionBuilder. No more 'yea, the ActorModel is really what should have been the User, but since User has legacy in the Authentication part (which also does some authorization, it's on the backlog), we needed another model name.
But also, learning about your domain, writing that down, outside (or before) patterns, frameworks, libraries start enforcing their domain, is very liberating and valuable in the long run.
What's that? I haven't encountered that terminology before.
You know, stuff you usually put in assert statements. Except in real life you don't have strict application of these laws in the data. So you need a way to find out what percentage of your data is "misshapen" and what to do with them. You can't necessarily filter them out, because they might reflect some business process with relevant information. You goal is to easily identify these outliers without crashing your processing (that would be stupid).
> DDD is not prescriptive. It doesn’t have rules of how to do it, and is open to new interpretation. It doesn’t prescribe methods, or practices, and even the patterns in the book1 are meant to be illustrative rather than a final set. ...
> That makes DDD notoriously hard to define.
I don't know anything about DDD, but if you can't concretely describe it and how it's different to how teams naturally work together, how can it be actionable and what stops it becoming another cargo cult when nobody can agree on what it is?
I found it a great practical introduction to DDD.
They are the probably the best concrete and actionable introductions to DDD out there, and do a very good job of separating practical DDD from the OOP dogma that it is sometimes tied to. The book uses F# as it's language to implement and demonstrate its examples, but most of the concepts are presented in a language agnostic way.
The most important part of DDD is the principles. It's a way of looking at the world and at problems, of seeing things first and foremost from the business perspective. How you apply those principles to achieve that perspective must vary dramatically from project to project, which means that the details of action are left to you to decide.
The one time it worked is a very complex domain of insurance policy administration. It had highly complex business rules and had to maintain invariants. The original developers were all experienced and senior from a consulting company. I joined as a permanent member of the team before handover.
The problem was that over several years it was a hard time to get new hires up to speed to maintain the complex code and test it. Moreover even the ubiquitous language starts to change with new CEO, SMEs retiring, change in market etc. E.g. Member became Customer. Now you either refactor the entire codebase or have the code different to the new business jargon. Guess which one was chosen? Over time it became hard to refactor. I have now left the company many years back, but I am assuming this code still lives.
Then I worked in enterprises where this kind of core systems were built in some humongous monolith like SAP, maintained my an army of people from WITCH companies. All the digital apps, web portal etc. was a frontend, API and caching layer. The DDD complexity was in the bowels of SAP or PEGA or Dynamics etc.
ES (Event Sourcing) and CQRS are technical patterns people like to use while doing DDD because of various reasons, but they're in no way required to practice DDD.
But DDD, by no means, needs ES.
Another part of the reason, I think, is because many other architectural patterns have taken certain concepts 'hostage'. Take for example Rails. MVC. 'Model' in Rails has a clear and distinct meaning, place and implementation. There is no way to have a bounded context, isolated domain models in Rails, without at least the terminology becoming very confusing. Having root aggregates in Rails makes all that worse, even.
What I'm trying to paint, here, is that ES naturally embraces DDD, whereas other architectures often conflict, or require effort to fit in DDD. Therefore, you'll probably find many talks about ES or CQRS and DDD, but hardly any about Actor, MVC, etc., and DDD.
To those experienced in DDD: You can do DDD without CQRS? And you can do CQRS without it being DDD too? This is not clear to me as an observer.
CQRS is associated with DDD because it is the kind of smart pattern made out of bits that the other smart patterns frown against -- like data duplication/denormalization.
* The essence of DDD: make the implicit explicit (language, boundaries, code) and evolve your model so it matches the domain *
But to be honest, in hindsight I think it was nothing more than a drip in the ocean, and does not clarify a lot, unless you are already well versed in DDD; it's like those explanations about monads...
As for the CQRS-as-a-top-level architecture:
There are a lot of "beginner experts" and self-proclaimed thought leaders emerging, as DDD is getting more popular... (~= Agile movement)
CQRS-all-the-things is typically a phase that you go through if you are studying DDD. (I've been there, done that, got the T-shirt.)
In my personal opinion a good heuristic to detect "beginner experts" is that they prefer to use a lot of DDD lingo, and focus on the more technical aspects instead of truly trying to understand the business domain first.
Update:
Added the first paragraph with an attempt for a DDD definition.
For example, DDD stresses the idea of an "isolated domain model" - and that has a very practical reason, being that your business rules are already complicated enough so mixing in database transactions, HTTP caching, response serialization, authorization, etc. etc., into your domain logic makes it harder to understand. You want to be able to have a conversation with a customer / domain expert where they explain something and you can adjust the code quickly to meet their needs. That's the ideal goal at least.
So some patterns enable that goal better. For example, this implies that you need some kind of data access layer pattern so that code that cares about the database is separate from code that cares about domain logic. Well, there's a million different ways to get that separation, and they have different tradeoffs, and like everything else, certain patterns become trendy. That's what's going on with CQRS. People feel that it leads to a better expression of the domain model, because creating data and querying it are often radically different from the customer's perspective.
Whether or not that goal is achieved, I'm not sure. But that's the reason people are experimenting with patterns like that, as I see it.
[1] https://academy.lightbend.com/courses/course-v1:lightbend+LR...
The above idea is hardly revolutionary anyway, I'm sure many people operated this way DDD or not.
This seems like another one in the endless stream of innovative project management methodologies that promise to make everything better, but whose main purpose is creating lucrative make-work opportunities to the next batch of consultants and evangelists.
Much of DDD is common sense, what it offers is a way to be explicitly common sense. The important factors, the way I see it, is to explicitly make sure both developers and stakeholder have a shared mental model of the business process that maps to the domain in question, they all agree on which 'jargon' terms map to which parts of that model AND that that model (with the agreed upon terms) is explicitly represented in the code.
Keeping in close contact with the business domain people is most useful if you actually understand each other. Too often I've seen developers and business people using the same 'every day' word when discussing a problem and only much later do they find out that their obvious understanding of that word was very different.
How complex your methodology around this has to be depends entirely on how complex the domain is and how many different definitions of "sold" is used with different parts of that domain.
doStuff({soldFromSys1, soldFromSys2, soldType, price, priceType, priceTypeIdKey, customer})
¯\_(ツ)_/¯
And, conversely, I have seen many people get extremely excited about solving technical problems that aren't related to the underlying business in any way. That is why the book needed to be written - people are more concerned with the newest database or frontend framework, but have lost sight of the purpose of software: to provide value to people.
So, sure, there are DDD consultants and trainers. They're going to capitalize off of the idea like vultures, as is done with all ideas and movements. But, the core beliefs of DDD seem like a good idea to me.
1. Understand the business of the customer, or any requirement won't make sense.
2. Learn the jargon of the business, or you won't be able to communicate.
3. Use that jargon when naming database tables, software modules, variables, everything, or you'll have to translate between your possibly abstract names and the jargon of the business.
So, apparently nothing new but I should read the PDF (too bad it's not HTML) because there could be new useful ideas about the process.
Edit: oops, the PDF is only that page as PDF. I'll google for it but does anybody have a good link to a detailed explanation of the method?
DDD is more about managing your business software needs than making software. If you read "Implementing DDD" a good thing to take from it the fact you should focus your efforts on your business core value add. It's where you put your best developers, architects and money. Anything outside this core will get less resources and can often be outsourced.
And to evaluate what is this core and what is needed, you need your technical team to speak often with the domain experts. Using some common language.
The coding aspect is itself agile as usually there will be miscommunication at first between your tech team and your expert giving you a less than good result. More communication, more knowledge shared and understood will make you think differently about your product and its architecture: that's when you refactor.
My experience is that, unfortunately, majority of developers are either not experienced enough, or smart enough (these are closely related), to keep the abstraction creep at bay and apply only meaningful abstractions. And others are simply not interested enough, and they know they can get a LGTM with following misjudged patterns.
They also jump at the opportunity to use identical terms from the DDD book, and then keep explaining them with more colloquial terms everybody understands: it's like social sciences all over again (sorry social sciences, but it's what it is) where they invent terms so they'd be more "scientific". Why not simply use the terms everybody understands?
I get the argument that a new language allows for consistency, but it still gets misapplied, because it's the same humans doing the work.
I can't think of a better parallel than SQL-vs-noSQL databases: sure, no-ACID makes a bunch of things simpler, but then every developer has to think through all of the same problems ACID DBs solve, and majority will get them wrong (or maybe even everybody would get them wrong _most of the time_).
Jump to coding or architecting because that's their happy place. That's easy, that's what they know so they think that's what they've been hired for.
Shelling some dollars to buy an off-the-shelf solution for some ancillary application? No thanks we think we can do it ourselves and have more control over it. Let's go coding.
To be precise probably. Every domain develops a jargon at some point. It can be used to exclude or sound smart but most of the time it just to be precise. Common terms are too much vague / polysemic.
Why do you need words like compiler, linker, dynamic and static types and linter instead of using common words?
My (limited) experience with social sciences (as obligatory courses in Math/CS studies) was that you'd have things like "Someone's Continuum" (I am, thankfully, at a loss for the name of Someone :)) to say that two extremes of learning are rote learning and learning with understanding (this, by virtue of what "two extremes" mean in a natural language, means that there are things in between), all explained in 3-4 pages of dense prose that says, literally, nothing else. There is no precision gained from introduction of these terms for the most part, mostly the number of things to memorize is increased.
To contradict that, a "continuum" in mathematics is given a very precise meaning to indicate that no matter how small your "in between", you can still find something there: you get very particular properties and can easily differentiate if something is a continuum or not. You also build up other tools to work on things which are "continuous", like derivatives, differentials or integrals.
Some of the language of DDD is quite like it, which probably makes sense, since it's more closely related to social sciences (not rigorous science), rather than formal sciences. That's not to slight it, because that's the best we can do with some things, but that means that in any particular setting, you should understand what you are trying to get out of it, and if you are struggling with language, move on to the point of the exercise instead (eg. establishing domains, boundaries and shared language as it relates to your problem).
I would appreciate an explanation of what DDD is by counterexample - i.e. what are examples of non-domain driven design process? Especially, examples of successful or mostly successful projects - i.e. where the disconnect from the problem domain does not feel wrong from the start.
What's particularly contrary to DDD principles and common is mixing up different concerns without regard for model integrity (e.g. ad hoc incorrect caching of stale data for performance reasons, messing up all business logic).
Don't get me wrong, DDD has meaning and purpose, but some companies are applying it as a badge to be obtained instead of pondering the question, do you really need to rewrite everything following DDD?
In our case, simple CRUD APIs that in "regular programming" might take a couple 200 line files have turned into unmanageable nightmares in DDD that take you at least a couple of days of really intensive investigation to understand, because it have been divided in more that 25 files that hold 3 or 4 lines of code at most, with so many abstraction layers that it's impossible for the best of us to follow in one go.
Now, you could make the argument "You Are Doing It Wrong(tm)" but since I'm just a drone in this specific scheme and there's no wiggle room for anything (the team is quite inflexible on this) I have to follow it to the letter.
Just giving my two cents, again, not depreciating DDD, it has its purpose but in my opinion, it's for very specific projects.
When you put engineers in charge you get overengineering and when you put managers you get underengineering.
Is there a way out?
This means you picked the wrong engineers. An engineer's job and skill is to be able to make this sort of call.
The same goes with managers or any other role.
So the way out is always to carefully select experienced, competent, no nonsense people.
I've worked with developers with 5+ years of experience that haven't went trough that (either corporate culture allowed them to deliver minimum value in 5 days or jumping projects before it gets to the WTF stage of complexity). It's hard to learn if you never get burned.
The whole idea is to bring them together "in the same room" and allow common understanding of the domain so they stay on the same page throughout the project. Then they'll do each what they are best at (managers whip up glossy slides, and devs crank out reams of code ;)
Which programming language? There was some work shown in .net recently
I agree with you that a minimal implementation should handle it but... someone had wet dreams with DDD and everything has to be DDD now :)
Perhaps that's a perfect example to implement it and push it. ( depends on how far you're willing to go and push back against who)
Look for a fancy name to describe the "pattern" could help.
Your story sounds like they implemented a "technical DDD top-level architecture" (TM), whatever that may be. (I'd assume layers of abstractions coupled with logic spread all over the place, without any added benefit.)
You see this a lot when people read some stuff about DDD, and they start experimenting with the technical/tactical patterns, because this is the aspect that makes most sense to a technical audience.
In reality the tactical/technical DDD patterns should only be applied in the core part of your business (i.e. the thing that gives you a strategic advantage over your competitors.), because that typically needs to change a lot, so having a common language/model with the business tends to be worth the extra upkeep required when opting for more flexible models.
Identifying what the core part is of your business (most likely it's not authentication, billing, invoicing, content management, ...) is one of the more important (and most difficult) aspects of DDD.
Exactly :)
I cannot imagine anything where it would make sense unless you're implementing an actual framework such as Spring in javaland.
And at that point I'd say you're wasting effort and just using spring boot should be preferable.
It does make sense with these super low level frameworks though, but which corporation makes them in-house at this point?
If your business is talking about updating/deleting/inserting data, you are not describing the actual reasoning behind the change. For DDD, it makes sense to figure out why exactly you are doing the things you do, and model these explicitly in your systems for those parts that matter.
The stereo-typical example for this is an address change: if you model this as an "AddressUpdated", you might as well use CRUD, as this does not specify the intent of the change.
You could change an address because it contained a typo, but you can also change it because someone moved. These might lead to different outcomes, so in DDD you would typically model these as "AddressTypoCorrected" and "ContactMoved".
There are other fine-grained aspects, for example the need of an identity for an object: physical money is considered a value object (so it has no specific ID per instance) in almost all contexts, unless you are the national bank: all of a sudden the identity (serial number of the bill) does matter, so the same "thing" might have different "models" within different parts of the business.
Other examples might be value objects for specific areas, for example weight. Typically weight starts out as a number, and all of a sudden there might be a need to add a precision, mark it as an estimate, or have it "unknown". In order to avoid if-statements all over the code, you construct a "weight value object" that properly manages all of these peculiarities in a single place (i.e. what's the result of an estimate+undefined etc.)
I'm sure there are successful and maintainable projects that utilize this design approach.
Nonetheless, the only thing I could think of after reading your example is just how many subtle bugs and inconsistent behaviors this engine will have with various edgecases, so I'm still pretty convinced I'd rather implement it with less abstraction/indirection.
It might just be a difference of character at the end of the day, because I do agree that what you write sounds great. I just see it more like a triangle in the CAP theorem, where the edges are speed, abstraction/extensibility and stability/consistency
This example clarifies the intent behind DDD: make the implicit explicit and make sure there is awareness about all the edge cases.
It might be as simple as contacting a user in case you have an uncovered edge case, but at least you'd be aware that your system is unable to handle edge case X. (In non-DDD scenarios this would just be a bug that emerges - implicit behavior.)
This line of reasoning cuts both ways: how many bugs and inconsistent behaviors often pop up because developers rushed to write code without gathering enough requirements on the domain, how many productivity problems are caused by growing the system by accretion where it can, and how many rewrites were required just to fit the system's domain to the problem and shed technical debt from the accretion.
To me, what matters are the strategic patterns, i.e. how you think and talk about your domain. What a lot of (microservice) software gets wrong is bounded contexts and APIs, which can be improved through event storming (discussing what kind of things happen in your domain) and context mapping (how that [sub]domain interacts with others). And then there's ubiquitous language, or calling the thing what it is to the business, not some programmer gobbledygook.
And then the client data has all fields named in Spanish with abbreviations, lettering and numbering everywhere.
In general it's preferable to use only english because the code is more readable (because the language's keywords and APIs are in english), but also to maintain consistency with noun-adjective order, verb tenses, etc. Some local concepts can be easily translated (or at least it would seem so) like invoice-factura ... but then inevitably I arrive at concepts for which there is no obvious translation (so I could make one up, but it would not be clear enough for others) or for which the translation doesn't exactly line up.
Any tips of how to do DDD when every area is considered a core priority of the business because uncomfortable conversations are hard?
[0] https://www.slideshare.net/TomJanssens1/selling-ddd
[1] https://twitter.com/huizendveld/status/1440683623628230665
Reasonable people (including DDD advocates) clearly understand that tactical patterns can easily be misapplied, and that the strategic part of it is more important, but inexperienced programmers jump in on the "new" fad (it's not even that new, which is most perplexing to me), and misapply all the patterns they possibly can.
I see there's a lot of similar sentiment here, so I think the question really is: how to convince inexperienced "converts" where the right boundary of applying tactical DDD solutions is?
A. Don't learn things.
B. Dedicate some amount of time to learn something in a sandbox before using it on the job.
C. Once you've learned something well enough to see the error of your ways, dedicate some amount of time to clean up your old work.
B and C are a single thing, and unfortunately, C does not happen, which is why any particular methodology gets a bad rep. And it's obviously already happening with DDD (judging by the polarized sentiments around here).
And finally, while I do believe abstraction is the ultimate tool of the human mind (and mathematics is the purest form of abstraction we are capable of), I do not think it suits all brains equally, and not everybody will be equally capable of ever getting the right understanding. Basically, your architecture can be _too smart_ if you are looking to hire actual, real-world developers and software engineers, and have them be efficient.
"There is some benefits to what you're proposing. My primary concern is cost. I see no technical difficulties with doing it <this other way>, and one of the benefits of that is that you'd have it ready in a tenth of the time. If I'm wrong, we can always change our minds and build it your way. That would barely take any additional time at all, since what I'm proposing is so simple to begin with.
"So, what do you say? Would you like to try to get it done by November, or should we labour over it until next June? If I understand the business figures right, if we can get it out in November, we'll make $6,000,000 more – and it would all be because you made the right technical decision here."
... <End of Year Review>
"We were able to get this out by end of November, thanks to your good technical decision. $6,000,000 saved! Congrats and great job. Accordingly, you've qualified for an annual bonus of a $100 Amazon Gift Card and the default 1.5% raise.
Keep up the good work!"
To get buy-in you have to provide value for them that aligns with their needs. If they desire a delusional feeling of outsized importance within an organization then you need to be quite creative. It's more likely that their needs are simpler though. They need to feel they are getting value from you, even if the other parts of the business hold more of the cards.
Try to find simple things to fix for them, choose to build out features with their input, and when building new features for the main stakeholders try and prioritize items which help multiple stakeholders. This is good practice in any case because in the long-term things can change quite significantly within organizations and you don't want to be perceived as only an ally of some within the business.
UPDATE: Beware of quantifying too much. This really impresses some people but will make others feel really small. You may completely lose a connection with one of the smaller stakeholders by quantifying every detail and calling out big numbers around the large stakeholders. You need to work qualitatively for the most part for them. If you want to use numbers, pilot something with a small stakeholder and make a 25% increase to their sales (or a cost reduction). If big holders can get a similar percentage then that translates to big numbers. It feels like a big number to both groups.
On one hand, there are quite a few smart concepts in there. Everyone should know what aggregates are, for example.
On the other hand, many of them have little to do with the basic premise of DDD. You can have DDD without these specific patterns.
On the third hand, people start applying every single pattern everywhere, like they did with GoF patterns.
https://www.wiley.com/en-us/Patterns%2C+Principles%2C+and+Pr...
Honestly, I see no relation at all. Those technical patterns can be applied the same way whether you model your code after the domain or not, and you don't need them at all to model your code after the domain.
They look a lot like "true Agile" 2.0.
He said if he wrote the book again, he'd have put all the patterns as an appendix
He thought people concentrate on applying the patterns rather than seeing DDD as a way to communicate. Between developers but also to the business
Let me put it this way, when Twitter started out, they did not have tweets. They had posts, and the act of posting to Twitter was called twittering. They were not associated with birds (actually more with whales lol). The idea of birds and tweeting actually came later with a third-party client interacting with their API.
Eric Evans in the early aughts now makes a big splash with this outrageous statement, where many of us graybeards would instead say “if it ain't broke don't fix it”: Eric Evans would recommend that the posts table in the database be renamed to the tweets table. Version 2 of the API should not reference “posts” or post_ids, but rather tweets and tweet_ids.
Why?! Those sorts of migrations are painful and clumsy! Yes, Eric says. (He is not stupid.) Maybe it's a lost cause. But, Eric remarks on two things:
1. There is no reason to believe, given software’s previous performance, that any amount of upfront planning is going to generate the most consistent useful model before the software is built and we can interact with it. So you're going to want to iterate. What are the systematic obstacles to renaming the table and the API, and can we overcome them so that we can do lots of little experiments?
2. Something else that is clumsy and painful, is when your users come to you reporting a problem with Widgets or whatever, and you go off and you fix the FactoryService to add some new functionality to widgets, tell the user that their problem is fixed, and they go and do the thing again and run into the same problem that they ran into, “it's not fixed yet!”. Why did this happen? One big reason is that the word “widget” means something different in the database versus the backend, or in the backend versus the frontend, or in the frontend versus the real world. Twitter might get some other notion of “topics” and they roll it out and everyone starts to call them “posts”, now the topic table holds posts and the posts table holds tweets, and you're always looking for “posts” in the wrong table now.
So, you should rename the table because first, this should be a possible thing for you to do and building up that sort of leverage is going to pay dividends later, and second, the less friction we can have by transforming the way we developers speak into the way that our users speak, is going to pay dividends too.
This anthropology is kind of the core part of DDD, I don't understand why people try to do DDD as design patterns rather than saying that it's the users who unwittingly dictate the design, as we redesign around them to reduce friction.
It's similar to, I don't understand why people find it hard to draw context boundaries in DDD. So bounded contexts are a programming idea, in programming we call them namespaces, they exist to disambiguate between two names that are otherwise the same. DDD says that we need to do this because different parts of the business will use the same word to refer to different things, and trying to get either side of the business to use some different word is error-prone and a losing proposition. So instead we need namespaces, so that both of our domain experts can speak in their own language and we can understand them both because in this context we use this namespace, in that context we use that namespace. So: where do you draw the boundary? In other words how big should your modules be? (Or these days, for “module” read “microservice”.)
Simple: you partition users into groups, based on the sorts of things that they seem to care about when they are interacting with the system, and the different ways that they talk about the world. The bounded context is not an “entity” or a “strong entity” or a service-discovery threshold, rather it is an anthropological construct just like everything else in DDD. “The people in shipping care about this for one reason, the people in billing care about it for another, they don't usually talk to each other, but I guess sometimes they do...” sounds like you've got a shipping module/microservice and a billing module/microservice. The boundary is the human boundary.
Similarly for “should I use events or RPC?” ... Does someone from shipping ever come up to the billing department and say “The delivery costs a ton more because XYZ, the customer said they preferred to pay more rather than cancel the order, I am gonna stay here in billing until this critical task is complete,” or whatever, or would they prefer an asynchronous process like email, “we will just put it on the shelf until we can pay to safely ship it.” Different industries would have different standards here! If it's something that has no shelf life, that delivery does not want to keep in the shelves for one second longer than it has to, then that drives the different behavior. Only way you can know is by observing your users in their natural habitat.
DDD has some great ideas about how to model things and talk about things. But it is not a concrete architecture prescribing a particular number of layers or lines of code.
As someone who had been in the industry 6-7 years at that point, it really resonated - he was describing modes of success and failure I had seen but didn’t really have names for. Much of the usefulness of the book was just to put names on these things.
What has happened to ‘DDD’ in the meantime surprised me. It never occurred to me from the original book that a methodology of strict practices could emerge from it. To me that wasn’t the sense if it at all.
The ideas made sense on logically complex code that required frequent refactoring, but the strict separation between all the different layers simply led to a lot of code in most instance. Far too much.
We also couldnt agree on where the limits of the bounded contexts really lay. Most documentation on this issue is a mere handwave saying "you figure it out" or "it'll become clear when you do these exercises with the business" (it didnt), which is odd given how vitally important it is and how damaging it is to bound the wrong things.
This is the hardest part of software design. No wonder there are no clear cut rules on how to do it. You have to be both a domain and implementation expert to get the boundaries right on the first try.
I'm not so convinced that it's something you can get just by better communicating with "the business" either.
This is partly what infuriated me so about the extreme amount of code required to follow DDD patterns - 3-4x the amount of code means 3-4x the cost of that rewrite if you get the boundary wrong.
Doesn't matter if your code has a type named `Aggregate` in it. Matters if you get your consistency boundaries right.
> I'm not so convinced that it's something you can get just by better communicating with "the business" either.
I don't have a good answer for this. I personally try to keep my modules small so that there's not more than a ~week worth of stuff to redo if understanding of business (or business itself) changes. I often fail too.
I tended to find a lot written about the former and very little written about the latter.
In fact I found essentially zero practical or actionable advice about getting the boundaries right beyond just "it's important to get it right". DDD doesnt appear to have a coherent opinion beyond that.
Not in the books nor in blog posts written about the topic.
It reminded me a bit of how scrum had a lot to say about standups (the color of that shed was well defined) but very little to say about refactoring.
Examples:
Are you writing to disk? Printing to the screen? Saving to the database? This is a different "abstraction level" than your data processing, those things should be separated by a boundary.
Are you doing hardcore math to procedurally generate a terrain? It shouldn't be on the same layer you push the triangles to the screen.
It's also about isolation: you want to control the Framebuffers in your 3D renderer? Don't let the Framebuffer class with OpenGL calls ever leak outside the renderer, even though there's encapsulation. Just use a data class to communicate between renderer/non-renderer code.
--
However, the part that is not really discussed on these books is that we should be vigilant not to "over-abstract". Sometimes you have a certain level of abstraction distributed over two or more classes, only to mirror some kind of external structure or mental mode, but in reality you want the code for those things to be together in the same class/method.
One example I run across a lot in 3D renderers is wrapping internal 3D library concepts like "vertex buffer", "context" and "framebuffer" in separate abstract objects, even when they really don't need to be abstract.
For example: "Open GL renderer" will only ever be able to call "OpenGL vertex buffer", "OpenGL context" and "OpenGL framebuffer", while Vulkan will only call the Vulkan equivalents. This means you don't need a "framebuffer" abstraction, you can have it on the same layer as the renderer. You might need a data-only, non-abstract "framebuffer" class to control it from the outside, though.
To be fair, I've never really seen any process/paradigm address this. I've mostly done it based upon gut feel - sometimes abstractions seem like too much of an imposition. Other times it feels critical.
DDD just says "do way too much. all the time".
I imagine one day there will be an abstraction calculus but software engineering aint there yet.
Now imagine you need to encode behaviours for a system where domain experts use terms and jargon you've never heard before. Even worse, users of the same system coming from different departments use the same terms to mean different things. How do you draw the boundaries there? That's what the GP finds disappointing - there is no single guide or reliable process to jump into a new domain and get the boundaries right.
Same with programs.
Why is that a bad thing?
* Try to mininimize the overlap between bounded contexts - loosely coupling domain models.
* Default to "too large" to begin with and institute a pattern for breaking down a bounded context into two smaller contexts.
I tend to find any process that defaults to "have more conversations/interactions" defaults to wheel spinning without some sort of specific plan about what those interactions would entail.
What if this doesn't lead to a more accurate representation of the domain? What if two contexts are just coupled in the business, for good reasons?
> Default to "too large" to begin with and institute a pattern for breaking down a bounded context into two smaller contexts
This is exactly the ambiguous advice that you are rallying against. How do you know when to break down a context?
It's like, building any system involving many actors and actions is hard, that has nothing to do with software. We're just digitizing the same patterns and behaviors that people have used to run companies for hundreds of years.
People want a playbook to be followed to arrive at a "perfect" domain model or architecture. I'm sorry, that sounds pretty farfetched to me. It reminds me of how we first started thinking about computability theory, when David Hilbert proposed that we should be able to devise an algorithm that could decide the truth or falsity of any logical statement (the Entscheidungsproblem). Hilbert was one of the smartest mathematicians to ever live, and he was very confident that this could be done.
Well, Alan Turing, Kurt Gödel, and Alonzo Church (not slouches in their own right) all smashed that idea with various proofs. The truth can often be counterintuitive. I am sorry that the world is complex, I also wish it weren't so.
This is a very important question. I will bite and give a guidance because no one is willing to give advice here. If you development team grows over 10 SE. Then you need to split it. So your bounded context has a complete team on it becoming experts in the subject as well as experts in the software implementation. Each team should be able to work independently of other teams.
Is "larger than 10 software engineers" a rule that domain experts would suggest? Would that be in the ubiquitous language?
I ask because only ever heard personal opinions in answer to this question. I researched it heavily and found nothing. As far as I can tell it doesnt.
As far as what the goal of the art is - the goal is to avoid linguistic and semantic ambiguities in the ubiquitous language. There is even a section entitled "Recognizing Splinters Within a Bounded Context" where specific examples are given:
* duplicate concepts * false cognates
If you have truly duplicate concepts across contexts, this is a symptom of the lines not properly being drawn, and perhaps a new, shared context is missing.
False cognates are the bread and butter of bounded contexts though - these occur when two areas of a business use the same exact word for something, but they mean slightly different things _depending on the business context_. The example in the book given is the notion of a "Charge," which customer invoicing and bill payment departments might both use. But each department only cares about certain pieces of data of a charge, so if one "Charge" model were created, it would be more complicated because it had to worry about all the different ways that the teams use it. And even worse, sometimes they are used in _conflicting ways_. That is a semantic collision, creating ambiguity in the model.
This is what a bounded context is meant to address. Each department gets its own model, each with its own version of Charge. The code and data is fit for the specific business purpose it's serving, instead of having a one-size-fits all model that gets the job done, but is more complicated to use in all contexts.
Honestly curious, have you read the book? I still don't think it will give you what you're looking for in terms of a prescriptive formula for "doing DDD right," but there's quite a bit of guidance in there.
It's conceptually similar to answering the question "How do I know where the borders of Germany lie?" by saying "ask the first person you see if they speak German".
It also conflicted with a process I followed, which was to essentially create a team glossary and agree to semantically disambiguate terms which had multiple different meanings (e.g. linux user/website user instead of just user) and even just "ban" the usage of terms which got overloaded too much.
(I discovered that semantic collisions didn't just present problems in code, it often prevented you and your team from having coherent conversations).
This could, of course, then put everything we touched as a team into the same bounded context. Or not...?
>The example in the book given is the notion of a "Charge," which customer invoicing and bill payment departments might both use. But each department only cares about certain pieces of data of a charge
It sounds like they're essentially saying (not explicitly, but via assumption), that your software should follow conway's law.
Nonetheless, this example screams "bug alert" to me, since assumptions made by departments (and, as a consequence software systems) about what they should care about are where the really nasty bugs lie - frequently driven by misunderstanding between departments about terms (e.g. what counts as a user).
It's a critical topic that is right at the heart of DDD. I researched up and down on this topic and unlike 'where to use a factory" the DDD community refuses to go into even as much detail on this topic than I just did with my half baked comment.
I am not trying to "complete" DDD here. I think DDD should largely be consigned to the trash heap.
Data-DD is focus on data and keep it simple. Domain-DD is another one of these architecture astronauts fads where you introduce new abstractions and then spend most of the time wondering whether penguin is a bird or a fish for the purpose of your application.
Reading this I guess you followed the wrong course. One that shoved you Tactical Patterns down the throat without explaining well-enough why you should use them. DDD in the core is not about architecture, but about *common understanding* of the domain, *before* you start coding and hacking on a particular architecture. The domain understanding should help you make appropriate architecture decisions, but in now way dictate what architecture designs to use.
That doesn't sound right. Domain-driven design is just a technique to design a data model that fits your application domain, and designs the whole application around that data model.
To put it differently, with DDD first you define a data structures, and afterwards the application is just operators over said data structure.
The only reason why in DDD you would care about whether "a penguin is a fish or a bird" is if your businesses required handling fishes or birds in fundamentally different and incompatible ways, which would require you to write special purpose code that crossed all layers.
It sounds like you've been experiencing a kind of analysis paralysis with the DDD tag.
Are you talking about design patterns by any chance? Maybe things like repositories, adapters, small and focused service classes and the like?
I think the main topic and real difficultly of DDD is figuring out specifically where and how to separate out different contexts (the so-called 'bounded context' in DDD parlance).
Creating microservices is a good example. Where should the responsibility for a single microservice start and finish? What might the implications for scalability and extensibility be? What does this mean for data storage? What data will be shared or replicated between microservices and how will this be done?
Answering these kinds of questions is hard and has big implications for your teams and for your business.
It's very specific that, for instance, your domain model "should only be created by factories", though.
It's all a bit bikesheddy.
shrug maybe im reading all the wrong things but IME most DDD discussions, blog posts and books sound more like this:
https://stackoverflow.com/questions/555241/domain-driven-des...
And few delve into the real "essence" as you put it.
Event storming is one of few times, but it doesn't seem to be a core part of DDD and i found the outcome to be underwhelming.
I think most important to realize that DDD is just another tool in your toolbox, and can be used alongside all / most of the other tools you already use. Event storming can be a nice way to quickly kick off the elaboration process, should the method appeal to you.
Money.
Most non-fiction literature is like that.
In particular, pulverized source files and excessive abstraction layers are characteristic symptoms of dogmatic, value-oblivious impractical design: quite the opposite of thinking hard about a meaningful domain model in order to use it as a shared language.
But here’s where you have to be careful. It’s still useful to go through things like Event Storming and modeling domains to understand the problem space before deciding that basic crud is good enough.
Your colleagues need to remember that DDD is supposed to be applied pragmatically. If the structure causes more navigation work than needed, simplify it. If the problem could be solved with a simple CRUD system, do that. If most of the problem is CRUD, but there's one particularly complex bit that changes a lot and requires a lot of flexibility, isolate that part, so that the simple and complex parts can have a simple integration, don't leak into each other, and can evolve at their own speeds.
I'd say part of the problem is that DDD critics conflate DDD with overly complex, enterprisey models that don't match their personal preferences on the acceptable tradeoffs between complexity and correctness.
As DDD comes up sounding like too much work to implement too much complexity that brings too little value, they flag it as a concern.
What I believe is missing from this discussion is the scenario where DDD practices are not followed and consequently teams are forced to iterate and reimplements projects or parts of it just to fit requirements that emerged because some aspects of the domain model weren't looked into. Design by accretion is largely accepted, as is technical debt, but they do have a cost.
This said, in my situation it's like trying to kill flies with a muon cannon (fyi. this gun is fictional and it's exaggerated to drive the point), it's cool'n's*it but the same could have been done with a newspaper or your hand. To maintain the muon cannon you need the entire Fermilab team, your hand... well, it's your hand.
Apologies for the exaggeration, can't avoid it :D
Every text I've ever written on DDD has made it pretty clear that these patterns are at the very heart of what it is. I've never seen one that says "look, all this stuff is optional, write your software however, here's how to really get to grips with the domain model".
I don't think it's the critics conflating. It is what it is.
Implementing DDD on the other hand is a lot better about this. Surely because it has been written 10 years later. So most people should start with it.
But anyway when people say they're using DDD, if they can't point you to some domain experts, a dictionary of the ubiquitous language or a mapping of what they do they're not using DDD.
In my experience it's extremely safe to assume that, even with the best intentions, DDD code can become stupid big balls of mud.
I always hate these arguments - for me, whether a particular programming paradigm is 'good' or 'bad' for an organisation comes down to: "what will my least senior developer do with this?". If it tends to produce tangled nightmares, then it's not a good paradigm, it's about how the weakest link will use it, not the strongest ones.
- try to write code that can be easily unit tested
- composition over inheritance
- immutable over mutable + avoid side effects
- avoid recursion unless your data is recursive
etc. etc.
The more general purpose it purports to be the vaguer and more misapplied it will end up being.
DDD could be a lot better as a movement if it tried to limit its scope a bit.
Until then, we have to stick to patterns that are very simple, easy to teach and have wide pits of success. We also have to lean heavily on systems that easy to replace and easy to refactor, and can be managed by the few very experienced devs. That's why platforms like Ruby on Rails did so well, even though they can be divisive.
To paraphrase Brooks, no pattern or methodology is gonna make development "more productive, more reliable or simpler". All patterns and methodologies require a skilled practitioner, but someone skilled enough can even eschew them and still make good software, so in the end they don't really matter.
The point of patterns/methodologies is purely to facilitate communication between experts, not to guide.
In the end, good software is not about following recipes, especially complex recipes. It's a craft.
As for where do we go, what we need is the foundation for personally re-discovering and internalising those same methodologies and patterns, rather than blindly following them. Sweezyjeezy already enumerated some ideas above, and I have done the same in the past: https://news.ycombinator.com/item?id=27987498
I wish ideas like these were more prevalent or that, at least, people considered that in many cases the quality of a method or practice is not an intrinsic attribute, but a matter of suitability to a particular environment.
That's the problem I have with agile enthusiasts in general. I've seen agile methods operating wonders on many projects, but I've seen it failing on much more, and it's completely OK to assume that the method has its assumptions/conditions/limitations but enthusiasts, instead, blame the company and practitioners for not having understood and applied it appropriately.
There's nothing about DDD that says you can't make a simple CRUD API, if that's all that is required.
DDD's principle value is one of ubiquity (sold as "Ubiquitous Language" but I posit that "Ubiquity" is more accurate) - does your code do what the organisation does, and vice versa?
Not just using the same terminology, but using the same workflow.
Now if what the organisation does is basic stuff, then your code should be basic, too.
If there is an asymmetry between what the org and code do.. there's pain.
That sounds kind of annoying, but not very hard to understand.
Instead I just try to absorb the major takeaways that I got from what I've read:
1. Bring in people with domain knowledge to help you understand the expected behavior of the system.
2. Try to establish a consistent language that is used in both verbal conversations as well as code.
I feel like those are good, easy-to-understand principles, and I've never understood why the whole DDD space is taken up by this insanely complicated terminology and theory. It's so off-putting.
It's a godsend for the top-down inside-out approach of requirements engineering that I like and teach.
Then I met some people the actual local DDD meetup group and was shocked that that part basically made up effectively zero percent of their discussion and the rest was taken up by talk about adapters, hexagonal architecture and all kinds of artificial design patterns that cultivate complexity and self-importance. I've been careful to call myself a DDD advocate ever since.
IMHO DDD went through the same unfortunate descent that Agile did: An originally really great idea and common-sense approach that went on to get bastardized into a cargo cult by coaches who like to produce sheets for BS bingo.
I also had the unfortunate experience of working with some die-hards who believe DDD dictates that "the code and only the code should reflect the business requirements 1:1", with emphasis on the "only". Which means that things that are often configurable in other systems are instead hardcoded and scattered in the codebase done by those. In early stage startups this is a death kneel.
Hugs. I'm sure we've all suffered this Kafkaesque torture. But it still sucks.
Have you heard of the CIA (nee OSS) book on Simple Sabotage Field Manual? http://www.simplesabotage.com
It predates Brazil, Office Space, Dilbert, etc. After reading this book, and observing management, it's hard to imagine it's not all deliberate. There's just something inherently evil in bureaucracy.
> ...you could make the argument "You Are Doing It Wrong™"
Ages ago my company's study group tackled Applying Use Cases: A Practical Guide. https://www.amazon.com/Applying-Use-Cases-Practical-Guide/dp... After all the monkey motion with UML, schemes, design patterns, etc, this book was like a clarion blasting away ignorance and ambiguity.
It was so clear. Do the use cases. Then directly derive architect from those use cases. Voila! Impossible to fuck up.
However. Young me learned a very valuable lesson.
Nothing is so obvious and virtuous and good that some whackadoodles cannot, will not comprehend it.
But why?
Obstinance? Actual confusion? Inability to suspend disbelief? White knuckled desperate grasp on prior beliefs? Fear? Moral and philosophical opposition? Refusal to concede control (power)?
I have no idea why.
Whatever the root cause, I've experienced these impasses so many times, I've simply given up.
I eventually learned to do whatever it takes to publicly appease the tyrannical gods of confusion, then do any actual work as able on the down low.
However, you _are_ correct to say that DDD has a very limited use - as does the domain model pattern itself. Martin Fowler even calls this out in Patterns of Enterprise Application Architecture, imploring that it is necessary only when there are “complex and ever-changing business rules”. Most business systems should be multi-player Access databases, and instead have “a few sums and not-null checks”, thus should not use a domain model, thus should not use a technique aimed at designing and validating a domain model.
Honestly, I’d find a new job.
That's the issue. If they weren't doing with DDD they would be doing the exact same thing with whatever the CEO read in Gartner instead anyways.
Anecdotal: I chatted with Eric Evans, author of the DDD book, at a conference once, and he stressed that DDD was only appropriate for certain parts of the system that called for it. I think he'd be as frustrated as you are by the situation you describe.
I can see how excessive detail division into Microservices can create a nightmare. I have seen it over and over. We need better guidance on how to breakdown services and how much. Erik Evans touches on this https://www.youtube.com/watch?v=sFCgXH7DwxM but I still think he is to shy on giving advice. Microservices are valuable because they allow your team to work autonomously. So you you should roughly have one Microservice per team. It is not a hard rule at all, but it can help to see if your microservices are too granular. That is the number of Microservices should be about 5 to 7 times larger than the number of developers. If you have less than 2 developers per Microservice, you are probably creating a maintenance nightmare.
The design patterns and actual implementations traditionally associated with DDD, on the other hand, are quite bad, and tend to lead to Enterprise Software levels of cruft and/or excessively dogmatic use of OOP.
What's even funnier is that none of the acronyms listed there is a soft-porn reference as claimed in the twitter thread.
A more real world story of DDD in practice is this blog post:
I think of game software as the entire system and not just the game client. That means any backend or infrastructure is a piece of the complex whole.
DDD for backend systems is well understood. The difference is it needs to serve game clients as well as any other clients (e.g. a website to market the game, forums for game related posting, maybe even run logic for chatbot(s)). If you want to expose online services to your game client, the backend system is where you will want to think about what logic to run and what state to persist.
I like DDD for some categories of game client code, not all. There are some simple gameplay elements that are driven from game engine physics and collisions. There are more complex gameplay elements that benefit from more complex domain modeling techniques. Basically allow yourself to write untested/lightly tests components for simple gameplay elements. For complex systems, it may be beneficial to think of unity as concrete implementations for domain specific logic. Model the code in the language of the game system expert and think about how to translate unity runtime events (user input, collisions) into terms meaningful for the domain.
In practical terms, unity game objects represent domain entities and interactions between game objects can trigger domain logic. For example, a projectile game object colliding with a combatant game object is a good place to execute combat service logic, which may trigger domain events (observable behavior) that can then be listened for by subscribers (e.g. a damage UI, a toast notification, or any other unity UI).
This is turning into more of a brain dump but I wanted to send this out into the world since it’s a topic I’m studying heavily right now.
Does anyone have more experience applying DDD to video games out there?
With DDD "done right" there's no issue with understanding what the code does because it's split into tons of files. It's actually the opposite. But, indeed, it could be a nightmare when applied blindly, or forcibly because of some ivory-tower architect decided to "use DDD" without even knowing what it is. I know as I've been that guy.
1. Listen to how your customer refers to their business and internalize this perspective as if your life depends on it (your career certainly may).
2. Find a way to build your product and communicate about it in similar terms. The methodology should be understood to be domain-specific.
Anything that extends the philosophy beyond these points is making assumptions about your product that really only you and your customers should be making.
The only concrete technical assumption I would make here is that you probably want to at least consider SQL for a few minutes if you are dealing with tons of distinct types and complex, rapidly-changing requirements from the business. If you can find a way to make the business write their own logic/queries in a language they understand, you have discovered the most powerful force multiplier I am personally aware of right now.
Reading about what it's meant to mean, it seems pretty common sense. As often implemented, however, it seems to lead to a lot of accidental complexity and a bunch of dubious usefulness abstraction layers.
I have a theory why this is. It seems that enterprise development as a field have a penchant for methodologies and acronyms. So, whatever silver bullet the enterprise-y shops get interested in eventually gets enterprisified. Be it UML, OOP, Agile, or DDD.
EDIT: I've found this an interesting read that sheds some more light on DDD using counter-examples:
[0] Video of the talk: https://vimeo.com/167722768
My guess was wrong, but I'm still confused what it is about :)
Other than that, most DDD concepts are a bit dated, and really oriented around JAVA/C# in the early 2000s.
That was like 25 years ago.
It sounds very reasonable, but as soon as you face the issue of communicating to the user or other components the fact that something is wrong and what is wrong, you'll discover that it is very hard to inform about illegal states if you cannot represent them.
Otherwise you may never notice that illegal state was there in the first place.
I encountered many systems that would take invalid inputs and produce invalid (but sometimes even valid) results [1].
Illegal internal states are even harder to catch. When using FSM / StateCharts I usually automatically print error to log (and crash) for all illegal state/event combinations in the State Transition Table, instead of silently ignoring them.
[1] GIGO - Garbage In Garbage Out
Let's say you're trying to parse a user object. Let's further say users always have to have a last name - a user without last name would be illegal state.
Now if you're parsing some data for a user that really doesn't have a last name for some reason, there's two approaches to this problem - either you return a User with a null last name (which is essentially giving incorrect information to the caller). Or you make it impossible to set User.lastName to null (for example by making it an Optional) and fail with an error about what went wrong.
Of course you wouldn't NEED to restrict User.lastName to never be null - but if you always fail with an error in that case anyways, why not? That way any consumer of the User object knows that the lastName will always be there and valid.
There should be only one place for name-handling logic, and since other people types besides users could appear in the domain model (commercial customer, social network "friend", relative, etc.) the user class isn't that place.
Unvalidated data (the DTO) is just the raw representation of whatever was inputed, read from storage, received from external services, etc. Any possible input should be acceptable and faithfully representable as a DTO.
Then, by passing the DTO to a validation function, you return either a validated object (a model) which is in fact constrained to only contain a legal data state; or a set of validation errors which can be acted upon.
Your business logic should operate only on validated objects, so that you can actually rely on your basic assumptions, and actual workflow rules (eg. "you can't checkout an empty cart") can be expressed and separated from trivial validation (eg. "quantity must be greater than 0").
I always took it as such.
Mathias is one of the original instigators of the European DDD community: he spent the last decade giving talks all over Europe about DDD and organizing one of the largest DDD conferences in the world: DDD Europe [0].
Next to this he also has a consulting org (recently rebranded to Aardling [1]), and spends a lot of time with DDD gurus like Eric Evans (the "inventor" of DDD, author of the DDD blue book), Alberto Brandolini ("inventor" of event storming), Nick Tune, Paul Rayner, Yves Reynhout, Cyrille Martraire, Romeu Moura, Marco Heimeshoff, ...
As far as I'm concerned, he's the real deal, and has skin in the game. Highly recommended!
100 software engineers with decade of experience,
then 30 of them would have no idea or understood it bad, and then you'd have 70 different interpretations
funny anecdote: at my last semester of eng. degree there was presentation to perform about some fancy topic. One person who had I guess like 2 or 3 years of xp decided to pick DDD and thought he'll manage to do it like a day or a few before presentation
He failed kinda hard with understanding it