Microservice-Driven Architecture

What is a Microservice?

Microservice-driven architecture, or simply microservices, is an architectural style that structures an application as a collection of loosely coupled services.

Each of these services is designed to execute a specific process or business capability, making them easy to understand, develop, and maintain.

For instance:

• FileService - This service is responsible for handling all file-related operations, such as file creation, reading, updating, and deletion.
• ConfigurationService - This service is tasked with managing all the configuration settings of an application.
• LoggerService - It captures and stores logs, which are records of events or transactions taking place within the application, typically used for debugging and monitoring purposes.
• ProcessService - The ProcessService is involved with the execution and management of application processes.
• LifecycleService - It manages the various stages an application or component goes through from initialization to termination.
• WindowService - It handles operations such as opening, closing, resizing, and moving windows.

Microservices and Dependency Injection (DI)

MDA can perfectly work with the idea of Dependency Injection (DI) to achieve a much less coupled system from an overall perspective. In order to understand the concepts in depth, I will show some real applications of what we can benefit from those two concepts.

For instance, consider the following class:

/**
 * @class The main class of the application. It handles the core business of the 
 * application.
 */
export class ApplicationInstance extends Disposable implements INotaInstance {

    constructor(
        @IInstantiationService private readonly mainInstantiationService: IInstantiationService,
        @IEnvironmentService private readonly environmentService: IMainEnvironmentService,
        @IMainLifecycleService private readonly lifecycleService: IMainLifecycleService,
        @ILogService private readonly logService: ILogService,
        @IFileService private readonly fileService: IFileService,
        @IMainStatusService private readonly statusService: IMainStatusService,
    ) {
        super();
        // ...
    }

    // ...
}

• The ApplicationInstance depends on all sorts of microservices in abstraction (in our context, depends on interfaces). ApplicationInstance is not responsible for creating those microservices (less coupled systems).
• Those prefixes in front of every field like @ILogService are known as decorators. These are language-specific techniques used to achieve DI in TypeScript. We will talk about this technique later.

Microservices Depends on Microservices

Similarly, microservices are not special classes, they can also depend on other microservices as well. As an example, the following MainWindowService also depends on all sorts of different microservices:

class MainWindowService implements IMainWindowService {
    constructor(
        @IInstantiationService private readonly instantiationService: IInstantiationService,
        @ILogService private readonly logService: ILogService,
        @IFileService private readonly fileService: IFileService,
        @IMainLifeCycleService private readonly lifeCycleService: IMainLifeCycleService,
        @IEnvironmentService private readonly mainEnvironmentService: IMainEnvironmentService,
    ) {
        // ...
    }
}

From a framework perspective, I can construct MainWindowService as follows:

• Regardless of the specific microservice instance passed through the framework, as long as that instance implements the corresponding interface, the system will function seamlessly.
  • For instance, as long as FileService, TestFileService, and NullFileService have all implemented the interface IFileService, the system will operate without issues.

// normal case
const mainWindowService = new MainWindowService(
    new InstantiationService(),
    new LogService(),
    new FileService(),
    new MainLifeCycleService(),
    new MainEnvironmentService(),
);

// unit test case
const mainWindowService = new MainWindowService(
    new TestInstantiationService(),
    new TestLogService(),
    new TestFileService(),
    new TestMainLifeCycleService(),
    new TestMainEnvironmentService(),
);

// null case
const mainWindowService = new MainWindowService(
    new NullInstantiationService(),
    new NullLogService(),
    new NullFileService(),
    new NullMainLifeCycleService(),
    new NullMainEnvironmentService(),
);

Godfather of All Microservices

Let’s address some potential questions that might arise so far:

• What happens when I’ve instantiated a FileService somewhere in my program, and another microservice, that depends on a FileService, needs to be constructed? From where does the new microservice obtain the previously constructed FileService?
• Does one have to manually construct all microservices, or is there a more efficient way to automate the construction process?
• Moreover, how do those decorators work, which we’ve previously seen and which are placed before each constructor parameter?

  All these questions find their answers in a sophisticated solution, a class called InstantiationService. InstantiationService lies at the heart of the microservice ecosystem. It’s responsible for managing the lifecycle of various services in the system.

• The primary task of InstantiationService is to create instances of services, injecting any dependencies as needed.
• InstantiationService is a concrete solution to achieve the Dependency Injection (DI) principle.

To illustrate how InstantiationService operates, consider the following API example:

// initialization (acting as DI)
const instantiationService = new InstantiationService();

// register a dependency (microservice) into DI (we will talk about it later)
instantiationService.register(IFileService, new FileService());

// create the service by its corresponding dependency tree (how it works?)
const mainWindowService = instantiationService.createInstance(MainWindowService);

• The above example provokes some thoughts:
  • First, what is a dependency tree of a microservice?
  • Second, how do we determine the dependency tree of each microservice?
  • Third, how do we create a microservice by its corresponding dependency tree? Don’t worry, I will explain these concepts to you step by step.

1. What Is a Dependency Tree?

A dependency tree represents a hierarchical relationship between different modules (or classes). Recall the previous example, ApplicationInstance, its dependency tree would look like the following:

ApplicationInstance
├─ IInstantiationService
├─ IEnvironmentService
├─ IMainLifecycleService
├─ ILogService
├─ IFileService
└─ IMainStatusService
    ├─ IInstantiationService
    ├─ ILogService
    ├─ IFileService
    ├─ IMainLifecycleService
    └─ IEnvironmentService

2. How Do We Determine a Dependency Tree?

Following the earlier example, where MainWindowService depends on IFileService. We can draw a few conclusions about IFileService:

• IFileService serves as an abstraction. IFileService as an interface, is a syntax sugar from TypeScript, which only exists on compile-time.
• IFileService can be implemented as concrete classes, such as FileService, TestFileService or NullTestService, and so forth.

• Since the interface is just a syntax sugar, it will be removed once compiled. We need a runtime solution to identify the existence of IFileService. Otherwise, the InstantiationService will not be able to establish the connection between IFileService and FileService/TestFileService/NullFileService, which is essential for constructing a MainWindowService based on IFileService.
• Luckily we can do that by using decorators. Let me introduce createDecorator decorator.

2.1 Runtime Identifier for Every Microservice - createDecorator Decorator

VSCode provide a useful function, createDecorator, which creates a unique decorator that can be served as an identifier for each microservice.

Sidenote: In TypeScript, a decorator is essentially a function.

• The decorator created by createDecorator, acts like an identifier, which can be stored in DI to make a connection between the microservice and the concrete class implementation as follows:

// fileService.ts
const IFileService = createDecorator('file-service'); // note: a variable in TypeScript can have the same name as an interface.

export interface IFileService {
    // ...
}

export class FileService implements IFileService {
    // ...
}

// playground.ts
const instantiationService = new InstantiationService();

// DI registration (⭐)
instantiationService.register(IFileService, new FileService()); // we've seen this line of code before

The decorator performs two functionalities in our cases:

1. First, since the decorator is a variable, thus it exists in run-time, it can be used in our DI system (InstantiationService). As previously mentioned, It establishes a connection between the abstraction concept (IFileService) and a concrete implementation (our case is FileService), as we’ve just done in the above example.
   • At line 16, the DI system now recognizes an abstraction (IFileService), and a way to construct its corresponding class (FileService).
2. Second, this decorator also helps us to construct a dependency tree at runtime, which leads us to the next sub-topic.

Variables and interfaces can have the same names. In this case, we have a variable and an interface both named IFileService.

Sidenote: Decorators will be immediately executed once the JavaScript script is loaded.

2.2 Build a Dependency Tree at Runtime Using Decorator

Let’s see what can this decorator be used for:

class MainWindowService implements IMainWindowService {
    constructor(
        @IInstantiationService private readonly instantiationService: IInstantiationService,
        @ILogService private readonly logService: ILogService,
        @IFileService private readonly fileService: IFileService,
        @IMainLifecycleService private readonly lifecycleService: IMainLifecycleService,
        @IEnvironmentService private readonly mainEnvironmentService: IMainEnvironmentService,
    ) {
        // ...
    }
}

A decorator, as the name tells, is used to add something extra to a class parameter. It does one key job:

• It marks the decorated class (in this case, MainWindowService), which depends on the parameter (e.g. IFileService), at runtime.

You do not need to worry about the black magic behind the decorators created by createDecorator. It works and works elgantly, I can promise you.

To give you a better illustration of what decorators do, let us consider the following class:

class TestService {
    constructor() {}
}

Currently, the class does not depend on anything. We can use an imagined function getDependencyTreeFor:

const dependencies = getDependencyTreeFor(TestService);
console.log(dependencies);
// []

Now, we let TestService depends on two other services:

class TestService{
    constructor(
        @IService1 private readonly service1: IService1,
        @IService2 private readonly service2: IService2,
    ) {}
}

If we print out its dependency tree again, we will see different things:

const dependencies = getDependencyTreeFor(TestService);
console.log(dependencies);
/**
[
  {
    id: [Function: serviceIdentifier] {
      _: undefined,
      toString: [Function (anonymous)]
    },
    index: 1,
    optional: false
  },
  {
    id: [Function: serviceIdentifier] {
      _: undefined,
      toString: [Function (anonymous)]
    },
    index: 0,
    optional: false
  }
]
 */

In essence, the decorator’s job is to create and store the above dependency tree at runtime.

Continuing our example, the complete dependency tree of MainWindowService will look like the following:

MainWindowService
└─ IInstantiationService
└─ ILogService
└─ IFileService
└─ IMainLifecycleService
└─ IMainEnvironmentService

Furthermore, the dependency tree can be more than just one level. The following dependency tree is also valid:

 MainWindowService
└─ IInstantiationService
└─ ILogService
└─ IFileService
└─ IMainLifecycleService
    └─ ILogService
└─ IMainEnvironmentService

However, the DI system will throw a runtime error when encounters a cyclic dependency. That is, encountering A depends on B, B depends on C, and C also depends on A.

3. How Do We Create a Microservice?

Now, we already covered a way to create a dependency tree for our classes. We finally reach a stage to discuss how we create our microservices.

3.1 Register a Microservice

For every dependency tree, its tree leaf (the dependency that depends on nothing) cannot be constructed automatically. It means the leaf must be provided by ourselves. We call this the registration process.

It is quite simple, recall our previous example:

// initialization (acting as DI)
const instantiationService = new InstantiationService();

// `FileService` does not depend on anything
// register a dependency (microservice) into DI (we just covered)
instantiationService.register(IFileService, new FileService());

// register other dependencies...

Of course, the other tree nodes other than just leafs can also be registered optionally.

3.2 Assembly Factory - InstantiationService

Here comes the final step of constructing a microservice: assembling. This is done inside the method createInstance:

// construct the service (by its corresponding dependency tree)
const mainWindowService = instantiationService.createInstance(MainWindowService);

When the InstantiationService (the DI System) is constructing a microservice, it will look for its dependency tree, iterate each dependency, and try to perform the following two operations to that dependency:

1. If the dependency already exists, meaning it has been constructed before, then the process is done and it continues to the next dependency.
2. If the dependency has not been constructed yet, the DI system recursively constructs this dependency.

After all the dependencies have been constructed, the InstantiationService uses them to construct the final target, the original microservice that was requested to be built.

Lazy Loading

Recall the second step when assembling in InstantiationService, where it checks ‘if the dependency has not been constructed yet’.

One might wonder how a microservice can be registered into the DI system without actually constructing one. For that, VSCode introduce a utility tool named SyncDescriptor:

export class SyncDescriptor<T> {

	readonly ctor: any;
	readonly staticArguments: any[];
	readonly supportsDelayedInstantiation: boolean;

	constructor(ctor: new (...args: any[]) => T, staticArguments: any[] = [], supportsDelayedInstantiation: boolean = false) {
		this.ctor = ctor;
		this.staticArguments = staticArguments;
		this.supportsDelayedInstantiation = supportsDelayedInstantiation;
	}
}

With the help of SyncDescriptor, we can achieve lazy loading when registering a microservice:

// initialization (acting as DI)
const instantiationService = new InstantiationService();

// register a dependency into DI using lazy loading
instantiationService.register(IFileService, new SyncDescriptor(FileService));

• When registering a microservice using SyncDescriptor, known as lazy loading technique, it ensures that the microservice is only constructed when it is actually needed.
• This strategy can significantly improve the efficiency and scalability of an application.