Notes on Domain Modeling Made Functional

I’ll divide my notes from Domain Modeling Made Functional into four main sections: fundamentals of domain-driven design; functional architectures; domain modeling; and high-level takeaways. I’ll include page numbers where applicable in case you’ve purchased the book and want to use these notes as a reference.

Overall, I greatly enjoyed this book, and I’d likely advocate for using its guidance if I become an architect of an enterprise software project in the future. In practice, I plan to use its guidance on domain modeling quite heavily: I want to understand the domain of freight forwarding (and specifically, warehousing) deeply enough such that I can model the core components effectively in software. In addition, I will continue to look for ways to apply the principles of functional programming (particularly around designing programs as pipleines) despite my usage of object-oriented languages in my day-to-day (Ruby, JavaScript, and Java).

Fundamentals of domain-driven design (DDD)

• Fundamental components of DDD (p. 22)
  • Bounded contexts: subsystems with clear boundaries, each of which own their data stores. These systems consist of workflows, and each has its own contextual language.
  • Domain events: generated via techniques such as event storming (and related to event sourcing)
  • Commands
  • Ubiquitous: a language with nouns — data structures — and verbs — business processes — that is shared between all stakeholders in a project. These terms should be maintained in a living document.

Functional architecture

• What are the main components of a functional architecture? (pp. 43-55)
  • Bounded contexts as autonomus subsystems with well-defined [data] boundaries. These contexts could be implemented in a variety of ways, depending on the scale of the system.
    • Could be simple software modules/classes, separate network-isolated services, or even microservices.
    • In general: start with a monolith and refactor into services at sufficient scale.
  • Components stay decoupled when communicating (via events/message queues), or simple function calls
  • Contracts between bounded contexts is a hard concept to get right, but incredibly important (p. 48)
  • Code structure within a bounded context: onion architecture (p. 52)

Domain modeling

• The idea of documentation as compilable code — i.e. types — is the central to domain modeling. (p. 57)
• Type patterns (p. 78)
  • Simple values (e.g. identifiers, messages). These values are often constrained in the domain. For example, if there are only 100 products in a catalog, an integer-based product code could only go up to 100.
  • AND combinations. These types represent closely linked data (i.e. documents), and are implemented with records.
  • Choices with OR. Represents a choice between this or that.
  • Workflows. Business processes with well-defined inputs and outputs. Often, these workflows cause side effects. (Workflows correspond to functions.)
• Value Objects vs. Entities for classifying data(p. 88)
  • Value Objects: objects without a persistent identity. For instance, if the fact that the fields of two objects are the same means the “value” of those objects are the same, you’re dealing with value objects.
  • Entity: objects with a persistent identity. Often, these objects are documents of some kind. They have a unique identity even if their components (fields) change.
  • Aggregate: collection of entities, where the top-level entity is called the aggregate root. Aggregates enforce consistency and invariants in the domain: if an entity in its collection changes, it will reflect those changes across other associated entites in the domain.
• Core principle: make illegal states unrepresentable in the domain via integrity and consistency (p. 103)
  • In distributed systems, we should be able to trust that “eventual consistency” will get the system to the state we expect in spite of transient failures.
• Workflows implemented as pipelines (p. 119)
  • Data is passed from one function to another, in the form of a pipeline. (And more concretely, using the pipeline operator in FP.)
  • State machines in particular can be highly useful for modeling workflows.
• Idea: documenting side effects in function type signatures.

Takeaways

• F# is a great language for DDD: it has useful primitives such as algebraic data types, modules, map/bind methods, Result/Async type wrappers, and computation expressions.
  • I don’t believe that the third section of the book in particular, since it was so implementation-heavy, is immediately useful to most developers if there is already lots of code in a different language. (Perhaps if the existing language is part of the JVM, though, this assertion might not hold.)
• I like to think of domain types as an instantiation of the ubiquitious language.
  • However, I don’t think this book dove as deep into requirements gathering as I would have liked. Without understanding the domain, you can’t attempt to domain model.
• With rigorous and intentional domain modeling, I can’t imagine the design getting out-of-sync with the code nor the real world requirements.
• Pushing input/output to the boundaries of a workflow/application can make business logic significantly easier to reason about. More practically, I think this pattern can be implemented in simple method calls as well: push logs, jobs, etc. to the very end of void (unit-returning) methods to make it crystal-clear where the side effects are occuring. Burying logs or other side effects in helper functions can make a system’s behavior incrementally more difficult to reason through (not to mention harder to test).