DDD re-distilled
This article is an abstract of the book DDD Distilled written by Vaughn Vernon
Last updated
This article is an abstract of the book DDD Distilled written by Vaughn Vernon
Last updated
Any organization that designs a system will produce a design whose structure is a copy of the organization's communication structure. - Melvin Conway
Style of code reflects the way the people and teams in your company communicate or produce software architectures matching their current structure.
It can lead to OUTDATED AND LOWER QUALITY PRODUCTS.
Build teams that look like the architecture you want :
Cross functional teams organized around business capabilities
Structure your system into independent, self-contained services so that teams can work independently.
Microservices architectures aim to increase velocity, so they apply Conway’s law in reverse to structure the system to achieve this noble goal.
Reverse Logic follows to create fewer dependencies and can grow faster as compared to Conway Law.
Teams only focus on what they are building and how they communicate with other pieces of code or micro-services.
Domain Driven Design a.k.a DDD is an approach to software development.
DDD is definitely NOT about technology.
DDD is all about the domain.
There are 2 kinds of patterns in DDD :
Often people focuses only on the tactical ones (closer to the code) but it is really mandatory before starting anything to start with a strategy.
A domain is the reality : your real business
A domain model is an abstraction of this reality
A Bounded Context is a semantic contextual boundary.
Within the boundary each component of the software model has a specific meaning and does specific things.
Software model = language that is developed by the team
Spoken by every member of the team
Implemented in the software model
This language = Ubiquitous Language
There should be :
One team assigned to work on one Bounded Context
A separate source code repository for each Bounded Context
One team could work on multiple Bounded Contexts BUT multiple teams should not work on a single Bounded Context
Generic Subdomain : Solution may be available for purchase or be outsourced or even developed in house by a team that doesn’t have the kind of elite developers that you assign to your Core Domain or even a lesser Supporting Subdomain. NO WAY TO DIFFERENTIATE
Core domain : Where you are making a strategic investment in a single, well-defined domain model.
This is very high on your organization’s list of projects because it will distinguish it from all competitors. The heart of an organization’s business.
Supporting domain : This is a modeling situation that calls for custom development, because an off-the-shelf solution doesn’t exist. This is still an important software model, because your Core Domain cannot be successful without it.
One of the Bounded Contexts will be the Core Domain : Various Subdomains in other Bounded Contexts.
A Bounded Context should align one-to-one (1:1) with a single Subdomain
Once you have identified your Subdomains you can use the matrix behind to understand which Subdomain is what.
Find more explanation on it here :
Make sure that your core (sub)domains and / or your most valuable contexts in genesis and custom built phases are perfectly staffed
Integration between Core Domain and other Bounded Contexts is known in DDD as Context Mapping
Considering that in two different Bounded Contexts there are two Ubiquitous Languages , this line represents the translation that exists between the two languages.
A Partnership relationship exists between two teams. Each team is responsible for one Bounded Context.
They create a Partnership to align the two teams with a dependent set of goals. It is said that the two teams will succeed or fail together.
Teams will meet frequently to :
Synchronize schedules and dependent work
Use continuous integration to keep their integrations in harmony.
The synchronization is represented by the thick mapping line between the two teams. The thick line indicates the level of commitment required, which is quite high.
Describes a relationship between two Bounded Contexts and respective teams : the Supplier is upstream (the U in the diagram) and the Customer is downstream (the D in the diagram).
The Supplier holds sway in this relationship because it must provide what the Customer needs.
Downstream requirements factor into upstream planning.
Downstream team gains some influence over the priorities and tasks of the upstream team.
A Shared Kernel , is a subset of a domain model that two teams share :
Physically shared artifact between two teams (Shared JARs, DB, …)
High degree of coupling requires a high amount of coordination between the involved teams
Use it with caution.
The conformist slavishly adheres to the upstream model :
No model-to-model transformation
Motivation : simplicity, contracts
he team implementing a Conformist is always downstream
Example : Consider the need to conform to the Amazon.com model when integrating as one of Amazon’s affiliate sellers.
The Anticorruption Layer translates one model to another one : (Looser coupling) Transforms an external model from another team / bounded context / system to another internal one Reduces the amount of coupling to a single layer The team implementing the ACL is always downstream.
Defines a protocol or interface that gives access to your Bounded Context as a set of services.
The Open-host Service is a public API :
One API for several consumers
Has a common/general purpose model and functionality
The team providing the Service is an upstream team
A well documented language shared between bounded contexts :
Every bounded context can translate in and out from that language
Often combined with Open-host Service
Such a Published Language can be defined with XML Schema, JSON Schema, …
Examples: iCalendar, vCard
A bounded context has no connections to others :
Sometimes integration is too expensive or takes very long
The teams choose separate ways in order to focus on specialized solutions
Perhaps the functionality that you seek is not fully provided by any one Ubiquitous Language.
In this case produce your own specialized solution in your Bounded Context and forget integrating for this special case.
Here’s what happens over time when you are responsible for creating a Big Ball of Mud :
A growing number of Aggregates cross-contaminate because of unwarranted connections and dependencies.
Maintaining one part of the Big Ball of Mud causes ripples across the model, which leads to “whack-a-mole” issues.
Only tribal knowledge and heroics—speaking all languages at once—save the system from complete collapse.
Source : https://speakerdeck.com/mploed/visualizing-sociotechnical-architectures-with-context-maps
There is one question that divides the DDD community : which mapping can I make between bounded context and micro-services ?
Well it depends...
Some consider a micro-service to be much smaller than a DDD Bounded Context. Using that definition, a micro-service models only one concept and manages one narrow type of data.
An example of such a micro-service is a Product and another is a BacklogItem
If this is the granularity that you consider a worthy micro-service, understand that both the Product ms and the BacklogItem ms will still be in the same larger, logical Bounded Context.
The two small ms components have only different deployment units, which may also have an impact on how they interact.
EventStorming is a flexible workshop format for collaborative exploration of complex business domains.
It comes in different flavors, that can be used in different scenarios:
to assess health of an existing line of business and to discover the most effective areas for improvements;
to explore the viability of a new startup business model;
to envision new services, that maximise positive outcomes to every party involved;
to design clean and maintainable Event-Driven software, to support rapidly evolving businesses.
Rules for organizing good sessions :
Invite the right people
Business
IT
UX
Technical stakeholders, ...
Provide unlimited modeling space (surface, markers, stickies)
Model a whole business line with domain events
It is not something from IT for IT but from business to the organization.
Find more about it here : https://www.eventstorming.com/
After an Event Storming session participants have a shared understanding of the Domain. One of the major criticism about Event Storming is really the lack of structure of the production that has been done. That's why it's really mandatory to capture the outcomes of each session.
Find more about the outcomes in this great article from Philippe Bourgau : https://philippe.bourgau.net/5-views-to-capture-the-outputs-of-an-event-storming-workshop/
You can use the Bounded Context Canvas to do so :
Find more about it here : https://medium.com/nick-tune-tech-strategy-blog/modelling-bounded-contexts-with-the-bounded-context-design-canvas-a-workshop-recipe-1f123e592ab
Real life example :
An Entity models an individual thing. Each Entity has a unique identity in that you can distinguish its individuality from among all other Entities of the same or a different type.
Most times, an Entity will be mutable; its state will change over time.
Models an immutable conceptual whole. Unlike an Entity , it does not have a unique identity, and equivalence is determined by comparing the attributes encapsulated by the Value type.
Each Aggregate is composed of one or more Entities / Value Objects One Entity is called the Aggregate Root.
"When all Task instances have hoursRemaining of zero, the BacklogItem status must be set to DONE .”
At the end of a transaction this very specific business invariant must be met. The business requires it.
Memory footprint and transactional scope of each Aggregate should be relatively small.
Keep the Aggregate design :
Small and efficient
Making for lower memory requirements
Quicker loading from a persistence store.
Helps enforce the rule not to modify other Aggregate instances within the same transaction.
Store them as you want Relational Database or not.
A BacklogItem is committed to a Sprint :
BacklogItem + Sprint must react to this
As part of the BacklogItem Aggregate ’s transaction, it publishes a Domain Event.
At one point our system will be consistent.
Here is the different steps you can use to go from Event Storming to your Aggregates Design :
Find more about it here : https://www.youtube.com/watch?v=AF880t1RUU4
The DDD community is doing a really great job and they provide this modeling process that gives you a step-by-step guide for learning and practically applying each aspect of Domain-Driven Design (DDD) - from orienting around an organisation's business model to coding a domain model.
Find more about it here : https://github.com/ddd-crew/ddd-starter-modelling-process/blob/master/README.md
DDD is a :
Strategic tool that can help you
Invest through Context Distillation
Organize teams through Context Mapping
Align everyone on a business / technical vision
Design tool through Bounded Contexts
Implementation tool through Tactical patterns
Here are great books on DDD :
