Domain-Driven Design (DDD) and Hexagonal Architecture in Java

Once we dig deeper into the hexagonal architecture, we will find that it has several resemblances to the more traditional layered architecture. Indeed, you can think of hexagonal architecture as an evolution of layered architecture. However, there are some differences, especially with regard to how a system interacts with the outside world. To better understand these differences, let’s start with a recap of the layered architecture:

The principle is that the system is built up of layers stacked on top of each other. A higher layer can interact with a lower layer but not the other way around. Typically, in a domain-driven layered architecture, you would have the UI layer at the top. This layer, in turn, interacts with an application service layer, which interacts with the domain model that lives in a domain layer. At the bottom, we have an infrastructure layer that communicates with external systems such as a database.

In the hexagonal system, you will find that the application and domain layers are still pretty much the same. However, the UI layer and the infrastructure layer are treated in very different ways. Keep reading to find out how.

If you have heard of the Entity-Control-Boundary pattern before, you will find the hexagonal architecture familiar. You can think of your aggregates as entities, domain services, factories, and repositories as controllers, and the application services as boundaries.

public class MyBusinessProcess {
    // Current process state
}

public interface MyApplicationService {

    MyBusinessProcess performSomeStuff(MyBusinessProcess input);

    MyBusinessProcess performSomeMoreStuff(MyBusinessProcess input);
}

Here are two Java examples of what security enforcement in an application service could look like. The code has not been tested and should be treated more as pseudo-code than actual Java code.

Declarative Security Enforcement

@Service
class MyApplicationService {

    @Secured("ROLE_BUSINESS_PROCESSOR") // (1)
    public MyBusinessProcess performSomeStuff(MyBusinessProcess input) {
        var customer = customerRepository.findById(input.getCustomerId()) // (2)
            .orElseThrow( () -> new CustomerNotFoundException(input.getCustomerId()));
        var someResult = myDomainService.performABusinessOperation(customer); // (3)
        customer = customerRepository.save(customer);
        return input.updateMyBusinessProcessWithResult(someResult); // (4)
    }
}

1. The annotation instructs the framework only to allow authenticated users with the role ROLE_BUSINESS_PROCESSOR to invoke the method.

2. The application service looks up an aggregate root from a repository in the domain model.

3. The application service passes the aggregate root to a domain service in the domain model, storing the result (whatever it is).

4. The application service uses the result of the domain service to update the business process object. It returns it so it can be passed to other application service methods participating in the same long-running process.

Manual Security Enforcement

@Service
class MyApplicationService {

    public MyBusinessProcess performSomeStuff(MyBusinessProcess input) {
        // We assume SecurityContext is a thread-local class that contains information
        // about the current user.
        if (!SecurityContext.isLoggedOn()) { // (1)
            throw new AuthenticationException("No user logged on");
        }
        if (!SecurityContext.holdsRole("ROLE_BUSINESS_PROCESSOR")) { // (2)
            throw new AccessDeniedException("Insufficient privileges");
        }

        var customer = customerRepository.findById(input.getCustomerId())
            .orElseThrow( () -> new CustomerNotFoundException(input.getCustomerId()));
        var someResult = myDomainService.performABusinessOperation(customer);
        customer = customerRepository.save(customer);
        return input.updateMyBusinessProcessWithResult(someResult);
    }
}

1. In a real application, you would probably create helper methods that throw the exception if a user is not logged on. I have only included a more verbose version in this example to show what needs to be checked.

2. As in the previous case, only users with the role ROLE_BUSINESS_PROCESSOR are allowed to invoke the method.

Here are two Java examples of what transaction management in an application service could look like. The code has not been tested and should be treated more as pseudo-code than actual Java code.

Declarative transaction management

@Service
class UserAdministrationService {

    @Transactional // (1)
    public void resetPassword(UserId userId) {
        var user = userRepository.findByUserId(userId); // (2)
        user.resetPassword(); // (3)
        userRepository.save(user);
    }
}

1. The framework will make sure the entire method runs inside a single transaction. If an exception is thrown, the transaction is rolled back. Otherwise, it is committed when the method returns.

2. The application service calls a repository in the domain model to find the User aggregate root.

3. The application service invokes a business method on the User aggregate root.

Manual transaction management

@Service
class UserAdministrationService {

    @Transactional
    public void resetPassword(UserId userId) {
        var tx = transactionManager.begin(); // (1)
        try {
            var user = userRepository.findByUserId(userId);
            user.resetPassword();
            userRepository.save(user);
            tx.commit(); // (2)
        } catch (RuntimeException ex) {
            tx.rollback(); // (3)
            throw ex;
        }
    }
}

1. The transaction manager has been injected into the application service so that the service method can start a new transaction explicitly.

2. If everything works, the transaction is committed after resetting the password.

3. The transaction is rolled back if an error occurs, and the exception is rethrown.

Here is a Java example of what a typical orchestration could look like. The code has not been tested and should be treated more as pseudo-code than actual Java code.

Orchestration involving multiple domain objects

@Service
class CustomerRegistrationService {

    @Transactional // (1)
    @PermitAll // (2)
    public Customer registerNewCustomer(CustomerRegistrationRequest request) {
        var violations = validator.validate(request); // (3)
        if (violations.size() > 0) {
            throw new InvalidCustomerRegistrationRequest(violations);
        }
        customerDuplicateLocator.checkForDuplicates(request); // (4)
        var customer = customerFactory.createNewCustomer(request); // (5)
        return customerRepository.save(customer); // (6)
    }
}

1. The application service method runs inside a transaction.

2. The application service method can be accessed by any user.

3. We invoke a JSR-303 validator to check that the incoming registration request contains all the necessary information. If the request is invalid, we throw an exception that will be reported to the user.

4. We invoke a domain service that will check if a customer is in the database with the same information. If that is the case, the domain service will throw an exception (not shown here) that will be propagated back to the user.

5. We invoke a domain factory that will create a new Customer aggregate with information from the registration request object.

6. We invoke a domain repository to save the customer and return the newly created and saved customer aggregate root.

In the first alternative, the application services return entire aggregates (or parts thereof). The client can do whatever it wants with them, and when it is time to save changes, the aggregates (or parts thereof) are passed back to the application service as parameters.

This alternative works best when the domain model is anemic (i.e., it only contains data and no business logic) and the aggregates are small and stable (as in unlikely to change much in the near future).

It also works if the client will be accessing the system through REST or SOAP, and the aggregates can easily be serialized into JSON or XML and back. In this case, clients will not actually be interacting directly with your aggregates but with a JSON or XML representation of the aggregate that may be implemented in a completely different language. From the client’s perspective, the aggregates are just DTOs.

The advantages of this alternative are:

• You can use the classes that you already have

• There is no need to convert between domain objects and DTOs.

The disadvantages are:

• It couples the domain model directly to the clients. If the domain model changes, you must also change your clients.

• It imposes restrictions on how you validate user input (more about this later).

• You have to design your aggregates in such a way that the client cannot put the aggregate into an inconsistent state or perform an operation that is not allowed.

• You may run into problems with the lazy loading of entities inside an aggregate (JPA).

Personally, I try to avoid this approach as much as I can.

In the second alternative, the application services consume and return data transfer objects. The DTOs can correspond to entities in the domain model. Still, more often, they are designed for a specific application service or even a specific application service method (such as request and response objects). The application service is then responsible for moving data back and forth between the DTOs and the domain objects.

This alternative works best when the domain model is very rich in business logic, the aggregates are complex, or the domain model is expected to change a lot while keeping the client API as stable as possible.

The advantages of this alternative are:

• The clients are decoupled from the domain model, making it easier to evolve without changing the clients.

• Only the data that is actually needed is being passed between the clients and the application services, improving performance (especially if the client and the application service are communicating over a network in a distributed environment).

• It becomes easier to control access to the domain model, especially if only certain users are allowed to invoke certain aggregate methods or view certain aggregate attribute values.

• Only application services will interact with the aggregates inside active transactions. This means you can utilize lazy loading of entities inside an aggregate (JPA).

• You get even more flexibility if the DTOs are interfaces and not classes.

The disadvantages are:

• You get a new set of DTO classes to maintain.

• You have to move data back and forth between DTOs and aggregates. This can be especially tedious if the DTOs and entities are almost similar in structure. If you work in a team, you need a good explanation for why the separation of DTOs and aggregates is warranted.

Personally, I start with this approach in most cases. Sometimes, I end up converting my DTOs into DPOs, which is the next alternative we will look at.

In the third alternative, application services consume and return domain payload objects. A domain payload object is a data transfer object that is aware of the domain model and can contain domain objects. This is essentially a combination of the first two alternatives.

This alternative works best in cases where the domain model is anemic, the aggregates are small and stable, and you want to implement an operation that involves multiple different aggregates. I would say I use DPOs more often as output objects than input objects. However, I try to limit the use of domain objects in DPOs to value objects if only possible.

The advantages of this alternative are:

• You do not need to create DTO classes for everything. You do it when passing a domain object directly to the client is good enough. When you need a custom DTO, you create one. When you need both, you use both.

The disadvantages are:

• Same as for the first alternative. The disadvantages can be mitigated by only including immutable value objects inside the DPOs.

Here are two Java examples of using DTOs and DPOs, respectively. The DTO example demonstrates a use case where it makes sense to use a DTO than return the entity directly: Only a fraction of the entity attributes are needed, and we need to include information that does not exist in the entity. The DPO example demonstrates a use case where it makes sense to use a DPO: We need to include many different aggregates that are related to each other in some way.

The code has not been tested and should be treated more as pseudo-code than actual Java code.

Data Transfer Object (DTO) example

public class CustomerListEntryDTO { // (1)
    private CustomerId id;
    private String name;
    private LocalDate lastInvoiceDate;

    // Getters and setters omitted
}

@Service
public class CustomerListingService {

    @Transactional
    public List<CustomerListEntryDTO> getCustomerList() {
        var customers = customerRepository.findAll(); // (2)
        var dtos = new ArrayList<CustomerListEntryDTO>();
        for (var customer : customers) {
            var lastInvoiceDate = invoiceService.findLastInvoiceDate(customer.getId()); // (3)
            dto = new CustomerListEntryDTO(); // (4)
            dto.setId(customer.getId());
            dto.setName(customer.getName());
            dto.setLastInvoiceDate(lastInvoiceDate);
            dtos.add(dto);
        }
        return dto;
    }
}

1. The Data Transfer Object is just a data structure without any business logic. This particular DTO is designed to be used in a user interface list view that only needs to show the customer name and last invoice date.

2. We look up all the customer aggregates from the database. This would be a paginated query in a real-world application that only returns a subset of the customers.

3. The last invoice date is not stored in the customer entity, so we have to invoke a domain service to look it up for us.

4. We create the DTO instance and populate it with data.

Domain Payload Object (DPO) example

public class CustomerInvoiceMonthlySummaryDPO { // (1)
    private Customer customer;
    private YearMonth month;
    private Collection<Invoice> invoices;

    // Getters and setters omitted
}

@Service
public class CustomerInvoiceSummaryService {

    public CustomerInvoiceMontlySummaryDPO getMonthlySummary(CustomerId customerId, YearMonth month) {
        var customer = customerRepository.findById(customerId); // (2)
        var invoices = invoiceRepository.findByYearMonth(customerId, month); // (3)
        var dpo = new CustomerInvoiceMonthlySummaryDPO(); // (4)
        dpo.setCustomer(customer);
        dpo.setMonth(month);
        dpo.setInvoices(invoices);
        return dpo;
    }
}

1. The Domain Payload Object is a data structure without any business logic that contains both domain objects (in this case, entities) and additional information (in this case, the year and month).

2. We fetch the customer’s aggregate root from the repository.

3. We fetch the customer’s invoices for the specified year and month.

4. We create the DPO instance and populate it with data.

If you are using a lot of value objects in your domain model that are wrappers around primitive types (such as strings or integers), then it makes sense to build the format validation straight into your value object constructor.

In other words, it should not be possible to create, e.g., an EmailAddress or SocialSecurityNumber instance without passing in a well-formed argument. This has the added advantage that you can do some parsing and cleaning up inside the constructor if there are multiple known ways of entering valid data (e.g., when entering a phone number, some people may use spaces or dashes to split the number into groups whereas others may not use any whitespace at all).

When the value objects are valid, how do we validate the entities that use them? There are two options available for Java developers:

1. Constructor and setter validation: Add the validation into your constructors, factories, and setters to prevent aggregates from reaching an inconsistent state. All required fields must be populated in the constructor, any setters of required fields will not accept null parameters, other setters will not accept values of an incorrect format or length, etc. I tend to use this approach when working with domain models that are very rich in business logic. It makes the domain model very robust but also practically forces you to use DTOs between clients and application services since it is more or less impossible to bind properly to a UI.
2. Java Bean Validation (JSR-303): Use annotations on all fields and ensure your application service runs the aggregate through the Validator before doing anything else with it. I personally tend to use this approach when I’m working with domain models that are anemic. Even though the aggregate itself does not prevent anybody from putting it into an inconsistent state, you can safely assume that all aggregates that have either been retrieved from a repository or have passed validation are consistent.

Both methods can be combined, applying constructor-level validation for domain models and JSR-303 validation for incoming DTOs or DPOs.

The simplest case of content validation is to make sure that two or more interdependent attributes within the same aggregate are valid (e.g., if one attribute is set, the other must be null and vice versa). You can either implement this directly into the entity class itself or use a class-level Java Bean Validation constraint. This type of content validation will come for free while performing format validation since it uses the same mechanisms.

A more complex case of content validation would be to check that a certain value exists (or does not exist) in a lookup list somewhere. This is very much the responsibility of the application service. Before allowing any business or persistence operations to continue, the application service should perform the lookup and throw an exception if needed. This is not something you want to put into your entities since the entities are movable domain objects. In contrast, the objects needed for the lookup are typically static (see the previous article about tactical DDD for more information about movable and static objects).

The most complex content validation case would be verifying an entire aggregate against a set of business rules. In this case, the responsibility is split between the domain model and the application service. A domain service would be responsible for performing the validation itself, but the application service would be responsible for invoking the domain service.

Here, we are going to look at three different ways of handling validation. In the first case, we will look at performing format validation inside the constructor of a value object (a phone number). In the second case, we will look at an entity with built-in validation so that it is impossible to put the object into an inconsistent state in the first place. In the third and last case, we will look at the same entity but implemented using JSR-303 validation. That makes it possible to put the object into an inconsistent state but not to save it to the database as such.

Value object with format validation

public class PhoneNumber implements ValueObject {
    private final String phoneNumber;

    public PhoneNumber(String phoneNumber) {
        Objects.requireNonNull(phoneNumber, "phoneNumber must not be null"); // (1)
        var sb = new StringBuilder();
        char ch;
        for (int i = 0; i < phoneNumber.length(); ++i) {
            ch = phoneNumber.charAt(i);
            if (Character.isDigit(ch)) { // (2)
                sb.append(ch);
            } else if (!Character.isWhitespace(ch) && ch != '(' && ch != ')' && ch != '-' && ch != '.') { // (3)
                throw new IllegalArgument(phoneNumber + " is not valid");
            }
        }
        if (sb.length() == 0) { // (4)
            throw new IllegalArgumentException("phoneNumber must not be empty");
        }
        this.phoneNumber = sb.toString();
    }

    @Override
    public String toString() {
        return phoneNumber;
    }

    // Equals and hashCode omitted
}

1. First, we check that the input value is not null.

2. We include only digits in the final phone number that we actually store. For international phone numbers, we should also support a '+' sign as the first character, but we’ll leave that as an exercise to the reader.

3. We allow but ignore whitespace and certain special characters that people often use in phone numbers.

4. Finally, when all the cleaning is done, we check that the phone number is not empty.

An entity with built-in validation

public class Customer implements Entity {

    // Fields omitted

    public Customer(CustomerNo customerNo, String name, PostalAddress address) {
        setCustomerNo(customerNo); // (1)
        setName(name);
        setPostalAddress(address);
    }

    public setCustomerNo(CustomerNo customerNo) {
        this.customerNo = Objects.requireNonNull(customerNo, "customerNo must not be null");
    }

    public setName(String name) {
        Objects.requireNonNull(nanme, "name must not be null");
        if (name.length() < 1 || name.length > 50) { // (2)
            throw new IllegalArgumentException("Name must be between 1 and 50 characters");
        }
        this.name = name;
    }

    public setAddress(PostalAddress address) {
        this.address = Objects.requireNonNull(address, "address must not be null");
    }
}

1. We invoke the setters from the constructor in order to perform the validation implemented in the setter methods. There is a small risk in invoking overridable methods from a constructor in case a subclass decides to override any of them. In this case, marking the setter methods as final would be better, but some persistence frameworks may have a problem with that. You just have to know what you are doing.

2. Here we check the length of a string. The lower limit is a business requirement since every customer must have a name. The upper level is a database requirement since the database, in this case, has a schema that only allows it to store strings of 50 characters. By adding the validation here already, you can avoid annoying SQL errors at a later stage when you try to insert too long strings into the database.

An entity with JSR-303 validation

public class Customer implements Entity {

    @NotNull (1)
    private CustomerNo customerNo;

    @NotBlank (2)
    @Size(max = 50) (3)
    private String name;

    @NotNull
    private PostalAddress address;

    // Setters omitted
}

1. This annotation ensures that the customer number cannot be null when the entity is saved.

2. This annotation ensures that the name cannot be empty or null when the entity is saved.

3. This annotation ensures that the name cannot be longer than 50 characters when the entity is saved.

I’m not going to give you an example of what a God-class looks like because that would take up too much space. Besides, I think most developers who have been in the profession for a while have seen their fair share of such classes. Instead, we are going to look at an example of what a command-based application service could look like. The code has not been tested and should be treated more as pseudo-code than actual Java code.

Command-based application services

public interface Command<R> { // (1)
}

public interface CommandHandler<C extends Command<R>, R> { // (2)

    R handleCommand(C command);
}

public class CommandGateway { // (3)

    // Fields omitted

    public <C extends Command<R>, R> R handleCommand(C command) {
        var handler = commandHandlers.findHandlerFor(command)
            .orElseThrow(() -> new IllegalStateException("No command handler found"));
        return handler.handleCommand(command);
    }
}

public class CreateCustomerCommand implements Command<Customer> { // (4)
    private final String name;
    private final PostalAddress address;
    private final PhoneNumber phone;
    private final EmailAddress email;

    // Constructor and getters omitted
}

public class CreateCustomerCommandHandler implements CommandHandler<CreateCustomerCommand, Customer> { // (5)

    @Override
    @Transactional
    public Customer handleCommand(CreateCustomerCommand command) {
        var customer = new Customer();
        customer.setName(command.getName());
        customer.setAddress(command.getAddress());
        customer.setPhone(command.getPhone());
        customer.setEmail(command.getEmail());
        return customerRepository.save(customer);
    }
}

1. The Command interface is just a marker interface that also indicates the result (output) of the command. If the command has no output, the result can be Void.

2. The CommandHandler interface is implemented by a class that knows how to handle (perform) a particular command and return the result.

3. Clients interact with a CommandGateway to avoid having to look up individual command handlers. The gateway knows about all available command handlers and how to find the correct one based on any given command. The code for looking up handlers is not included in the example since it depends on the underlying mechanism for registering handlers.

4. Every command implements the Command interface and includes all the necessary information to perform the command. I like to make my commands immutable with built-in validation, but you can also make them mutable and use JSR-303 validation. You can even leave your commands as interfaces and let the clients implement them themselves for maximum flexibility.

5. Every command has its own handler that performs the command and returns the result.

In the first example, we are going to create a REST API for our Java application:

The port is an application service that is suitable to be exposed through REST. The REST controller acts as the adapter. Naturally, we are using a framework such as Spring or JAX-RS that provides both the servlet and mapping between POJOs (Plain Old Java Objects) and XML/JSON out-of-the-box. We only have to implement the REST controller, which will:

1. Take either raw XML/JSON or deserialized POJOs as input,

2. Invoke the application services,

3. Construct a response as either raw XML/JSON or as a POJO that the framework will serialize, and

4. Return the response to the client.

The clients, regardless of whether they are client-side web applications running in a browser or other systems running on their own servers, are not a part of this particular hexagonal system. The system also does not have to care about who the clients are as long as they conform to the protocol and technology that the port and adapter support.