This module provides a set of functions to do dependency injection in a manner similar to ASP.NET, but agnostic of framework.
First you need to setup scopes around each request. Use a middleware from your framework to do that.
For sdkgen:
import { setupScope } from "@cubos/inject";
api.use((ctx, next) => setupScope(next));
For Koa:
import { setupScope } from "@cubos/inject";
app.use((ctx, next) => setupScope(next));
For Express:
import { setupScope } from "@cubos/inject";
app.use((req, res, next) => setupScope(next));
Then you can begin creating, registering and using your services. See below.
The concept of "scope" can vary depending on the framework, but here we call "scope" a ongoing request from the client. Everything a request is received a new scope is created. All operations executed because of this request happens inside the same scope. You can also have nested scopes, if you need that (most likely you don't).
setupScope
receives a single argument, a callback function. This function will be called immediately and its result will be returned. Any asynchronous task created by this callback will inherit the same scope. This is possible by using AsyncLocalStorage
, available since Node 12.17. Thus this module does NOT work in Node 10.
You are not required to setup a scope if you are not going to use scoped services or values.
Registers a new service with the specified lifetime.
"singleton"
lifetime: A new instance will be created on the first usage and will be kept for reuse. The same instance will be returned always."scoped"
lifetime: A new instance will be created on the first usage inside each scope. A instance created for one scope won't be returned for another. It is safe to store that such as user credentials in such service."transient"
lifetime: A new instance will be created every time. This kind of service must be lightweight.
For example:
class LargeCacheService {
expensiveData = new Map<string, string>();
constructor() {
// populate this.expensiveData
}
}
registerService("singleton", LargeCacheService);
class FeatureFlagService {
constructor() {
// load features from current user
}
isEnabled(feature: string): boolean {
// ...
}
}
registerService("scoped", FeatureFlagService);
If the class constructor need arguments, you can pass them to registerService
:
class CoolIntegrationService {
constructor(private axiosClient: AxiosInstance, private apiKey: string) {}
// ...
}
registerService("transient", CoolIntegrationService, axios.create(), env.COOL_API_KEY);
If you need some custom construction logic and for some reason the class constructor can't be used for that, you can use registerServiceWithFactory
. This function will be called only when needed.
registerServiceWithFactory("singleton", SomeService, () => SomeService.getInstance());
This function can be called from anywhere (including from the constructor of a service) to obtain an instance of a service ready to use. The lifetime of the service is handled behind the scenes.
class UserInformationService {
constructor() {
// populate information of current user
}
}
class FeatureFlagService {
dbConnection = use(DatabaseConnectionService);
constructor() {
const userId = use(UserInformationService).userId;
}
}
The intended use is to call use
from your controllers to obtain services and interact with them. Member functions of services can call use
themselves as well. If the service is scoped, you must be inside a scope to use it.
Not everything is a service. Maybe you have a stand alone function or some data you want to share. For this registerValue
can be used.
registerValue("dbConnection", pool);
registerValue("handleError", err => console.error(err));
Those values can be later used with use
:
use("dbConnection").query("SELECT 1+2");
use("handleError")(new Error("failed"));
// Or:
use.dbConnection.query("SELECT 1+2");
use.handleError(new Error("failed"));
By default the return of use
is typed as unknown
, but you can provide a custom typing to improve your experience. You can change to you strict typing by adding the following to your project:
// This must be a .d.ts file
import type { ConnectionPool } from "some-package";
module "@cubos/inject" {
export interface UseTypeMap {
dbConnection: ConnectionPool;
handleError(err: Error): void;
}
}
Similar to registerValue
, registerScopedValue
can be used to register values to be consumed with use
. But they will only be available within the same scope.
The goal of this library is to make dependency injection during unit tests easy. This is our suggested architecture:
-
Create your service classes and values. They can
use
freely, but they should not register themselves. -
Create your controllers and routes consuming those services with
use
. -
Create your application entry point file. This is the only file that won't be tested. It should register all services and start the application server.
-
Create a test setup file. Assuming Jest, you should pass this file as
setupFilesAfterEnv
on jest config. It should have abeforeAll
that can register some default mock or services/values that all tests can use. Then it must call the following:beforeEach(pushInjectionContext); afterEach(popInjectionContext);
This pair of functions are responsible for creating an isolated context for each test.
-
Write your tests. Each test can
registerService
its own mocks as needed, not need to cleanup. After each testpopInjectionContext
will take care of eliminating anything this test registered.