Back to Posts

Web Development

Setting up a Hybrid REST & GraphQL API with NestJS and MikroORM

07 March, 2021
Last updated at 02 July, 2021.

Use the same scalable approach Jenyus Org uses to make your next API with the NestJS framework and one of the best Typescript ORMs available.

Jenyus CTO & GraphQL Enthusiast


While searching for a framework and toolkit to use, which would allow us to create scalable, high-performance APIs for modern web apps with realtime functionalities and the flexibility to create mobile clients as well as use modern stacks with GraphQL and NodeJS, we came across the NestJS framework and liked it so much, that we created two starter templates for it. One of which we will be using today to setup a hybrid REST and GraphQL API from scratch.

What is NestJS?

NestJS is a lightweight abstraction over the ExpressJS and Fastify libraries. It offers many common features and design patterns built-in, such as dependency injection, as well as integrations with popular libraries for tasks like authentication, field validation and interceptors, and an MVC pattern using controllers and resolvers to create modular web APIs as you might be used to from other frameworks like ASP.NET Core and Spring Boot.

NestJS is also built with Typescript in mind, which means it offers amazing type-safety and advanced ES6 features of Javascript out of the box.

You can read more about NestJS on the official website and their docs.

Why NestJS?

We defined a couple of requirements our server-side framework would have to fulfill in order for us to consider it as our main stack going forward.

  • Architecture: Features such as dependency injection, middlewares and authorization need to be implemented in a way that is scalable. ExpressJS in particular lacks this aspect when it comes to structuring larger applications;
  • Type-safety: We wanted to use a somewhat type-safe language, to make it easier to catch bugs during development and improve our workflow. This basically eliminated languages such as Javascript, PHP and Python;
  • Documentation: A good framework is useless if we have to scour pages of StackOverflow threads to find out how to do certain things, so the framework needed to have a good foundation and community support;
  • Hosting: ASP.NET Core would have been our first choice if it wasn't for the much higher demands for hosts, which might not be feasible for smaller projects.

This is how we eventually settled on the NestJS framework. Both Jenyus' CEO, Dominik Berger, and myself enjoy putting together our own toolkit. NestJS offers a lot of flexibility when it comes to integration libraries for common tasks, like ORMs, authentication, authorization and validation.

Setting up the Boilerplate

The NestJS boilerplate is available under the Jenyus Organization's GitHub profile. You can clone it using the git CLI:

git clone

If you would like to use the boilerplate which uses TypeORM for database interactions, you can use the TypeORM starter instead.

Once the project has been cloned, we need to install the dependencies and setup a couple of files. We use Yarn so run the yarn command in the project director, and then create a .env file which we will use later:

cd nestjs-auth-graphql-mikroorm-starter yarn touch .env

The .env file for now just contains a key which the JWT library will use to encode JWT tokens. It can be any random string, just make sure its unique and you don't commit this file to version control.


Project Setup

Now it's time to get into the codebase and go over the various libraries used to implement all the features a typical web API would offer. We'll start with the ORM, which connects our API to the database.

Tip: After you've setup the ORM you can run yarn start:dev to start the project, and then access the Swagger docs at localhost:3000/api or the GraphQL Playground at localhost:3000/graphql.


ORMs, or object relational mappers, are libraries that simplify working with a relational or NoSQL database in our codebase. They can automatically resolve relations using JOINs and nested SQL SELECTs, as well as map our database rows to more logical Javascript objects, or Typescript classes.

Two of the most popular ORMs for Typescript users are TypeORM and MikroORM. Both of them offer fantastic typing for querying and inserting/updating data, but TypeORM has accumulated some issues over the years and was mostly abandoned by its maintainer. MikroORM is the new kid on the block which boasts many similar features, which is why for future Jenyus projects we have decided to go with it. Recog still uses TypeORM.

MikroORM comes with a lot of features, and a very well-made NestJS integration that basically acts as a drop-in for TypeORM. In order to make the CLI aware of the database connection, you need to setup a mikro-orm.config.ts file in the src/ folder. The starter comes with an example file mikro-orm.config.example.ts, which is configured for SQLite. You can copy it and make modifications as you see fit.

const logger = new Logger("MikroORM"); const config = { type: "sqlite", dbName: "./tmp/data.sqlite", entities: ["./dist/**/*.entity.js"], entitiesTs: ["./src/**/*.entity.ts"], debug: true, highlighter: new SqlHighlighter(), migrations: { path: "./src/database/migrations", }, logger: logger.log.bind(logger), } as Options; export default config;

You can read more about how MikroORM's CLI works here.


Now it's time to setup some entities. Entities are Typescript classes, decorated with MikroORM @Entity() decorators that reflect database tables and columns. MikroORM supports one-to-many as well as many-to-many relationships, and comes with other features one might expect from an ORM.

The starter samples a forum-like application, where one user has many posts. The user entity we defined includes properties for authentication, such as username and password, which we will get back to later, as well as the profile data and posts relationship.


@Entity({ tableName: "users" }) export class User { @PrimaryKey() id: number; @Property() username: string; @Property() password: string; @Property({ nullable: true }) firstName: string; @Property({ nullable: true }) lastName: string; @OneToMany(() => Post, (post) =>, { cascade: [Cascade.REMOVE] }) posts = new Collection<Post>(this); @OneToMany(() => RefreshToken, (refreshToken) => refreshToken.user, { cascade: [Cascade.REMOVE], }) refreshTokens = new Collection<RefreshToken>(this); @Property() createdAt: Date = new Date(); @Property({ onUpdate: () => new Date() }) updatedAt: Date = new Date(); }

You can read more about defining MikroORM entities here.

Other entities that are defined in this starter are RefreshToken and Post, which can be found under the auth and posts modules respectively.


We have decided to handle authentication using JWT with a standard refresh token and access token based flow. In order to make our APIs compatible with different kinds of clients that may not support cookies, we simply return these tokens in our response bodies, and delegate the responsibility of keeping these tokens safe to our clients.

NestJS has great guides on implementing JWT token based authentication in their docs, we followed that approach and added refresh tokens on top.

For our REST API we use the PassportJS local auth strategy to validate a user logging in for the first time with their username and password combination. This is used by the /api/auth/login route to return the reusable JWT token only if the user has been verified. The configuration can be found in src/auth/strategies/local.strategy.ts:

@Injectable() export class LocalStrategy extends PassportStrategy(Strategy) { constructor(private authService: AuthService) { super(); } async validate(username: string, password: string) { const user = await this.authService.validateUser(username, password); if (!user) { throw new UnauthorizedException(); } return user; } }

LocalStrategy calls the authService.validateUser() method, which itself uses the UsersService to retrieve a user by their username and then uses BCrypt to compare the hashed passwords. If they match, it returns the user, otherwise null.

As I alluded to before, I use this strategy in the AuthController's login method to verify the user and generate tokens. We need to create a guard that NestJS will use, to make sure the route handler is only called if the request body contains the correct username and password combination.


import { Injectable } from "@nestjs/common"; import { AuthGuard } from "@nestjs/passport"; @Injectable() export class LocalAuthGuard extends AuthGuard("local") {}

Afterwards we can decorate our controller method with @UseGuards(LocalAuthGuard) and generate our access and refresh tokens.


@Controller("auth") export class AuthController { constructor(private authService: AuthService) {} @UseGuards(LocalAuthGuard) @Post("login") @ApiBody({ type: LoginUserBody }) @ApiOkResponse({ description: "User has been logged in.", type: LoginUserResponse, }) async login(@Request() req) { const accessToken = await this.authService.generateAccessToken(req.user); const refreshToken = await this.authService.generateRefreshToken( req.user, 60 * 60 * 24 * 30, ); const payload = new LoginUserResponse(); payload.user = new UserDto(req.user); payload.accessToken = accessToken; payload.refreshToken = refreshToken; return payload; } }

The equivalent is possible in GraphQL, with the login mutation.


@Resolver() export class AuthResolver { constructor(private authService: AuthService) {} @Mutation(() => LoginUserPayload) async login(@Args("input") input: LoginUserInput) { const user = await this.authService.validateUser( input.username, input.password, ); if (!user) { return new UserInputError("Username or password incorrect."); } const accessToken = await this.authService.generateAccessToken(user); const refreshToken = await this.authService.generateRefreshToken( user, 60 * 60 * 24 * 30, ); const payload = new LoginUserPayload(); payload.user = user; payload.accessToken = accessToken; payload.refreshToken = refreshToken; return payload; } }

Since PassportJS has no strategies like the one we used earlier in REST for GraphQL, we simply created a LoginUserInput model, which expects the username and password, and then verifies those manually using the authService (this is also the only place where we duplicate code 😉).


@InputType() export class LoginUserInput { @Field() username: string; @Field() password: string; }


To protect subsequent routes and queries, we can use guards just like did before. First we need to use the JWT strategy from PassportJS, which follows a similar interface to the PassportJS local strategy to retrieve a user based on the bearer token passed in the authorization headers.

The JWT strategy validates our JWT bearer token, and then uses the information we stored in it in our AuthService to retrieve our user by ID from the database, allowing us to check roles and permissions in the future.


@Injectable() export class JwtStrategy extends PassportStrategy(Strategy) { constructor( configService: ConfigService, private usersService: UsersService, ) { super({ jwtFromRequest: ExtractJwt.fromAuthHeaderAsBearerToken(), ignoreExpiration: false, secretOrKey: configService.get<string>("auth.jwtKey"), signOptions: { expiresIn: "15m" }, }); } async validate(payload: any) { const { sub: id } = payload; const user = await this.usersService.findOne({ id }); return user; } }

After this we can implement the actual guard. NestJS provides an AuthGuard that expects the name of the strategy we defined earlier to protect routes. In this case jwt.


@Injectable() export class JwtAuthGuard extends AuthGuard("jwt") {}

As seen earlier in the AuthController, routes can be protected by decorating them with @UseGuards(LocalAuthGuard). Most routes will use the JwtAuthGuard, though, since a client expects to be able to authenticate the user with just the access token.


In order to protect GraphQL queries and mutations, we need to create a separate guard that lets NestJS access the actual request, since the GraphQL implementation has to modify our request context in order to store additional information.


@Injectable() export class GqlAuthGuard extends AuthGuard("jwt") { getRequest(context: ExecutionContext) { const ctx = GqlExecutionContext.create(context); return ctx.getContext().req; } }


Now we get to the fun part, implementing actual CRUD functionality into our app using the entities and authentication middleware we setup previously. We're going to implement a simple PUT /api/users/profile route which lets the user update their own profile.

In order to make it a little easier to access the user from our request context, instead of directly accessing request.user, we're going to create a NestJS ParamDecorator that returns the user for us.


export const CurrentUser = createParamDecorator< unknown, ExecutionContext, User >((_, ctx) => { const request = ctx.switchToHttp().getRequest(); return request.user; });

Now we can use this decorator in whichever controller methods we need to get the database user and check their roles, permissions, etc.

Next it's time to create some DTOs. DTOs, or data transfer objects, are interfaces and classes that help us define what a client needs to send to our API and what we return. This separates the database entity from our APIs schema, which makes it easier to add computed values, and hide fields we might not want to show publicly, such as the password in the case of a User.


@Exclude() export class UserDto { constructor( partial: Pick<User, "id" | "username" | "firstName" | "lastName">, ) { Object.assign(this, partial); } @Expose() @ApiProperty() readonly id: string; @Expose() @ApiProperty() readonly username: string; @Expose() @ApiProperty() readonly firstName: string; @Expose() @ApiProperty() readonly lastName: string; }

As you can see, we use the class-transformer library's @Exclude() and @Expose() decorators to hide all fields assigned to this object by default, and only expose those we actually want to show. I do it this way in order to maintain the best possible security standards for our application as otherwise Object.assign() in our constructor will assign all the values, which will then be displayed in our request's return values.

Also for the values our users will pass us we use DTOs. This has two benefits, it informs OpenAPI docs of all the properties that we expect, and it gives us better type-hints in our controllers than if we simply accepted objects of the any type.


export class UpdateUserDto { @ApiProperty() readonly username?: string; @ApiProperty() readonly firstName?: string; @ApiProperty() readonly lastName?: string; }

Now lastly, to allow the user to update their profile we need to implement the controller method. It expects an UpdateUserDto and returns a UserDto. We pass that information to the OpenAPI decorator, so Swagger docs can display those in the frontend, and also use them as types to get type-hinting in our controllers and services.


@Controller("users") export class UsersController { constructor(private readonly usersService: UsersService) {} @UseGuards(JwtAuthGuard) @Put("profile") async update( @CurrentUser() user: User, @Body() updateUserDto: UpdateUserDto, ) { const res = await this.usersService.update(, updateUserDto); return res && new UserDto(res); } }


The GraphQL implementation of this same functionality is even easier. We're going to be creating a updateProfile mutation, that takes a UpdateUserInput input type, but first, just like before, let's make it a little easier for us to get the user from the request context.


export const GqlCurrentUser = createParamDecorator< unknown, ExecutionContext, User >((_, context) => { const ctx = GqlExecutionContext.create(context); return ctx.getContext().req.user; });

Now it's time for our DTOs. For our projects we use the code-first approach to generate GraphQL object and input types. You can read more about how NestJS handles GraphQL schema generation here, but we've decided to use code-first as it reduces the amount of code duplication we have, and foregoes the need for a generator script that converts our schema into interfaces for type-hinting.

Just like before, we won't be mixing entities and GraphQL objects, so we're creating a separate UserObject with NestJS's code-first approach. You can read more about it here.


@ObjectType("User") export class UserObject { @Field(() => Int) readonly id: number; @Field() readonly username: string; @Field({ nullable: true }) readonly firstName: string; @Field({ nullable: true }) readonly lastName: string; }

Defining the update input type is way easier in GraphQL, as the input type basically follows the same structure as the UserObject, we can use NestJS's mapped types to easily make the ID a required type, while all other types are optional.


@InputType() export class UpdateUserInput extends IntersectionType( PickType(UserObject, ["id"] as const, InputType), PartialType(OmitType(UserObject, ["id"] as const, InputType)), ) {}

Then, in our resolver, we define the updateProfile mutation, which uses the DTOs we just defined for type-hinting and schema-generation.


@Resolver(() => UserObject) export class UsersResolver { constructor( private readonly usersService: UsersService, private postsService: PostsService, ) {} @Mutation(() => UserObject) updateProfile(@Args("input") input: UpdateUserInput) { return this.usersService.update(, input); } }

Optimizing Relations using GraphQL-Utils

GraphQL implements some very unique functionality which REST does not offer: Requesting relationships in the same query as the root object. Say we have a user, which owns a list of posts. In REST this would require a separate request to the /api/posts endpoint, but in GraphQL we can retrieve this data all in a single request using the following query:

{ users { id username firstName lastName posts { id title body } } }

The user's posts are resolved using field resolvers. You can read more about how NestJS handles GraphQL code-first resolvers here.

Essentially, the users query returns a list of users, and then GraphQL calls the user's posts field resolver, with the resolved user as the parent argument in order for us to return the posts from the database that belong to the given user. You can read more about GraphQL's execution flow here and the arguments that are passed to resolvers here.

Jenyus Org has published a NPM library for NestJS called nestjs-graphql-utils which lets us use declarative syntax and NestJS ParamDecorators to determine whether we should already fetch the user's posts in our original SQL query to improve performance. By doing so we reduce latency and thus speed up queries wherever possible. The implementation is also very straightforward using an ORM like MikroORM, since it's capable of populating relationship collections automatically using dot-notation.


@Resolver(() => UserObject) export class UsersResolver { constructor( private readonly usersService: UsersService, private postsService: PostsService, ) {} @Query(() => [UserObject]) users(@Selections("users", ["posts"]) relations: string[]) { return this.usersService.findAll({ relations }); } @ResolveField(() => [PostObject]) async posts(@Parent() user: User) { if (user.posts.isInitialized()) { return user.posts; } const { id } = user; return this.postsService.findAll({ authorId: id }); } }

In our users query we use the @Selections() decorator from @jenyus-org/nestjs-graphql-utils to check which fields were selected and directly store them in an array. This array is passed to our UsersService which relays that information to MikroORM.

We still need to implement the User.posts field resolver, though. MikroORM lets us check if a collection has been populated, which is useful since an empty collection is still technically initialized. You can read more about MikroORM collections here. If the collection wasn't initialized by previous resolvers, we can use the PostsService to fetch them after the fact and in the best case simply return the initialized collection.


interface FindAllArgs { relations?: string[]; } @Injectable() export class UsersService { constructor( @InjectRepository(User) private usersRepository: EntityRepository<User>, ) {} findAll(args?: FindAllArgs) { const { relations } = args; return this.usersRepository.find({}, relations); } }

I wrote more about this technique of pre-fetching required relations in my personal blog, using a more barebones implementation of GraphQL and the graphql-utils package by Jenyus Org here.

Closing Words

NestJS gives us the best of Typescript and Javascript, while offering many useful features and built-in integrations to handle common use-cases.

The Jenyus starter template bundles all those in a properly architectured structure and can be used as a boilerplate for hybrid REST and GraphQL APIs, or just support one! The choice is up to you.