Chapter 2: Relationships
In this section, we will add more entities, create a common base entity to extend from, and define relationships between them.
Created and updated timestamps
Before we add more entities, let's refactor our existing User
entity a bit. We would like to store timestamps of when the entity was created and when it was updated for the last time. To do that, introduce two new properties:
@Property()
createdAt = new Date();
@Property({ onUpdate: () => new Date() })
updatedAt = new Date();
The onUpdate
option here will be executed during the flush
operation if the ORM detects the entity was updated. For create query, we depend on the property initializers - we already said that the ORM will never call your entity constructor when creating managed entities, so it is safe to use it like this.
Custom base entity
Now let's say we want to have these timestamps on every entity in our app. We can refactor such common properties out to a custom base entity. Put the following into src/modules/common/base.entity.ts
:
import { PrimaryKey, Property } from '@mikro-orm/core';
export abstract class BaseEntity {
@PrimaryKey()
id!: number;
@Property()
createdAt = new Date();
@Property({ onUpdate: () => new Date() })
updatedAt = new Date();
}
You can see a base entity looks like any other entity, with one exception - it does not have the @Entity()
decorator. There are some use cases where you will want to use the decorator even for the base entity, in that case, add abstract: true
to the decorator options, e.g. @Entity({ abstract: true })
.
Now extend this base entity from the User
entity and remove the properties it defines:
You can see the import with
.js
extension - this is mandatory for ESM projects. If your project is targeting CommonJS, drop it.
import { Entity, PrimaryKey, Property } from '@mikro-orm/core';
import { BaseEntity } from '../common/base.entity.js';
@Entity()
export class User extends BaseEntity {
@Property()
fullName!: string;
@Property()
email!: string;
@Property()
password!: string;
@Property({ type: 'text' })
bio = '';
}
More entities
Time to add the Article
entity, it will have 4 string properties and one relationship - a ManyToOne relation pointing to the User
entity. As you expected, it will go to the src/modules/article/article.entity.ts
file.
import { Entity, ManyToOne, Property, t } from '@mikro-orm/core';
import { BaseEntity } from '../common/base.entity.js';
import { User } from '../user/user.entity.js';
@Entity()
export class Article extends BaseEntity {
@Property({ unique: true })
slug!: string;
@Property({ index: true })
title!: string;
@Property({ length: 1000 })
description!: string;
@Property({ type: t.text, lazy: true })
text!: string;
@ManyToOne()
author!: User;
}
Let's break this down, there are some new additions we haven't seen before.
slug
property is marked asunique
, this will result in a unique constraint over the columntitle
property is marked asindex
eddescription
property has alength
option, the column will result invarchar(1000)
with most SQL driverstext
property uses thet.text
mapped type, and is marked aslazy
, meaning it won't be selected automaticallyauthor
property is our first relationship
reflect-metadata
providerThe examples here are all based on the @mikro-orm/reflection
package that helps with advanced type reflection via ts-morph
, based on TypeScript Compiler API. If you are using the default reflect-metadata
provider, there are things you will need to add to the decorator options to make things work. Let's say the author
property is optional, ts-morph
would infer everything just fine, but with reflect-metadata
you would have to do this:
// with @mikro-orm/reflection package (ts-morph)
@ManyToOne()
author?: User;
// with the default provider (reflect-metadata)
@ManyToOne({ entity: () => User, nullable: true })
author?: User;
Consult the docs for more examples.
Types of relations
There are 4 types of entity relationships in MikroORM:
- ManyToOne: Many instances of the current Entity refer to One instance of the referred Entity.
- OneToMany: One instance of the current Entity has Many instances (references) to the referred Entity.
- OneToOne: One instance of the current Entity refers to One instance of the referred Entity.
- ManyToMany: Many instances of the current Entity refers to Many instances of the referred Entity.
Relations can be unidirectional and bidirectional. Unidirectional relation is defined only on one side (the owning side). Bidirectional ones are defined on both sides, while one is owning side (where references are stored), marked by inversedBy
attribute pointing to the inverse side. On the inversed side we define it with mappedBy
attribute pointing back to the owner.
When modeling bidirectional relationship, you can also omit the
inversedBy
attribute, definingmappedBy
on the inverse side is enough as it will be auto-wired.
Check the Modeling Entity Relationships section in the documentation for more details and examples for each of the types.
Working with relations
Let's get back to the server.ts
file and try a few things out with our new Article
entity, namely with its author
relation.
Creating entity graph
So far we used the entity constructor manually to create an entity instance. Sometimes we might want to create the whole entity graph, including relations. You can use em.create()
for that, it is a synchronous method that creates the entity instance for you. It allows you to create a deep entity graph, mapping foreign keys of your relations to entity references of the correct type.
You can wipe most of the contents of
server.ts
file and keep only the initial part with ORM init, up to the point where the firstUser
entity gets flushed, plus theorm.close()
call at the end. We won't be using this code going forward, it is just a playground for you.
// create new user entity instance via constructor
const user = new User();
user.email = 'foo@bar.com';
user.fullName = 'Foo Bar';
user.password = '123456';
// fork first to have a separate context
const em = orm.em.fork();
// first mark the entity with `persist()`, then `flush()`
await em.persist(user).flush();
// clear the context to simulate fresh request
em.clear();
// create the article instance
const article = em.create(Article, {
title: 'Foo is Bar',
text: 'Lorem impsum dolor sit amet',
author: user.id,
});
// and persist it to the database
await em.persist(article).flush();
console.log(article);
em.clear()
If you carefully checked this snippet, you probably found that new mysterious em.clear()
call. What it does do? It clears the context for the EntityManager
, meaning it will detach all the entities it was managing. It will bring the EntityManager
to the same state as if you would create a fresh fork via em.fork()
. You won't usually need this in your app, but it is very handy for unit testing, to simulate new requests coming in. You may as well use forks explicitly if you want.
But wait, the editor is complaining about something. You probably see this cryptic error:
Argument of type '{ slug: string; title: string; description: string; text: string; author: number; }' is not assignable to parameter of type 'RequiredEntityData<Article>'.
Type '{ slug: string; title: string; description: string; text: string; author: number; }' is missing the following properties from type '{ slug: string; title: string; description: string; text: string; author: EntityData<User> | { id?: number | null | undefined; fullName?: string | null | undefined; email?: string | ... 1 more ... | undefined; password?: string | ... 1 more ... | undefined; bio?: string | ... 1 more ... | undefined; } | EntityDataPr...': createdAt, updatedAt ts(2345)
It's indeed a bit ugly, but if you look carefully, you will the important details at the very beginning and at the very end. This error tells us the object we are passing into em.create()
is not complete - it is missing two properties, the createdAt
and updatedAt
timestamps. But we define the default value for them via property initializer, what's the problem here?
The thing is, there is no easy way to tell whether an object property has an initializer or not - for TypeScript our createdAt
and updatedAt
properties are both mandatory. To get around this while preserving the strict type checking, you can use the OptionalProps
symbol. As both of the problematic properties live in the BaseEntity
, put it there:
import { OptionalProps, PrimaryKey, Property } from '@mikro-orm/core';
export abstract class BaseEntity {
[OptionalProps]?: 'createdAt' | 'updatedAt';
@PrimaryKey()
id!: number;
@Property()
createdAt = new Date();
@Property({ onUpdate: () => new Date() })
updatedAt = new Date();
}
With this change, you can see the TypeScript error is now gone. Running the npm start
, you will see the Article
entity will get persisted and logged to the console:
[query] begin
[query] insert into `article` (`author_id`, `created_at`, `description`, `slug`, `text`, `title`, `updated_at`) values (1, 1662908804371, 'Foo is bar', 'foo', 'Lorem impsum...', 'Foo', 1662908804371) [took 0 ms]
[query] commit
Article {
id: 1,
createdAt: 2022-09-11T15:06:44.371Z,
updatedAt: 2022-09-11T15:06:44.371Z,
slug: 'foo',
title: 'Foo',
description: 'Foo is bar',
text: 'Lorem impsum...',
author: (User) { id: 1 }
}
Remember the entity references and loaded state we discussed earlier? You can see that here in action, the Article.author
is an entity reference with just the primary key. It is automatically logged as (User)
so you can easily tell the loaded state of any entity, but it is in fact the very same User
entity instance:
console.log('it really is a User', article.author instanceof User); // true
console.log('but not initialized', wrap(article.author).isInitialized()); // false
Using entity constructor
Every Article
can be identified by a unique slug - a URL fragment that can be used to look up the article. Currently, it is just a regular string property, but we can do better here. The value should be always bound to the article title. For simplicity, we will use the following function:
function convertToSlug(text: string) {
return text.toLowerCase()
.replace(/[^\w ]+/g, '')
.replace(/ +/g, '-');
}
We want the URL to remain the same after the article gets created, so let's generate the slug inside Article
constructor. Similarly, you can use the text property and store its first 1000 characters as the description:
import { Entity, ManyToOne, Property, t } from '@mikro-orm/core';
import { BaseEntity } from '../common/base.entity.js';
import { User } from '../user/user.entity.js';
function convertToSlug(text: string) {
return text.toLowerCase()
.replace(/[^\w ]+/g, '')
.replace(/ +/g, '-');
}
@Entity()
export class Article extends BaseEntity {
@Property({ unique: true })
slug: string;
@Property({ index: true })
title: string;
@Property({ length: 1000 })
description: string;
@Property({ type: t.text, lazy: true })
text: string;
@ManyToOne()
author: User;
constructor(title: string, text: string, author: User) {
super();
this.title = title;
this.text = text;
this.author = author;
this.slug = convertToSlug(title);
this.description = this.text.substring(0, 999) + '…';
}
}
With this change, the slug
and description
properties are optional too - but em.create()
complains about them. You need to add them to the OptionalProps
definition, as with the timestamps before. But these are the Article
entity properties, so we should do it in the Article
entity somehow. Maybe like this?
export class Article extends BaseEntity {
[OptionalProps]?: 'slug' | 'description';
}
Unfortunately not, you will see TypeScript error like this one:
Property '[OptionalProps]' in type 'Article' is not assignable to the same property in base type 'BaseEntity'.
Type '"slug" | "description" | undefined' is not assignable to type '"createdAt" | "updatedAt" | undefined'.
Type '"slug"' is not assignable to type '"createdAt" | "updatedAt" | undefined'. ts(2416)
Generics to the rescue!
The solution here might not be clear, but it is very simple. Instead of redefining the OptionalProps
property, we define a generic type argument on the BaseEntity
class and pass Article
specific properties down to the base entity.
export abstract class BaseEntity<Optional = never> {
[OptionalProps]?: 'createdAt' | 'updatedAt' | Optional;
// all properties remain the same
}
We picked the default value of never
for our type - this is because a union with never
will always yield the same type, e.g. string | never
is the same as just string
.
export class Article extends BaseEntity<'slug' | 'description'> {
// all properties remain the same
}
Now the em.create()
call work even without the slug
and `description:
const article = em.create(Article, {
title: 'Foo is Bar',
text: 'Lorem impsum dolor sit amet',
author: user.id,
});
console.log(article);
Running npm start
we can see the slug
and description
populated with generated values:
Article {
id: 1,
createdAt: 2022-09-11T16:08:16.489Z,
updatedAt: 2022-09-11T16:08:16.489Z,
slug: 'foo-is-bar',
title: 'Foo is Bar',
description: 'Lorem impsum dolor sit amet…',
text: 'Lorem impsum dolor sit amet',
author: (User) { id: '1' }
}
Alternative approach with Opt
type
Another way to make TypeScript aware of what properties are optional is the Opt
type, you can intersect it with the actual property type. This way the above problem with extending classes is no longer present, as we operate on property level:
export class Article extends BaseEntity {
@Property({ unique: true })
slug: string & Opt;
@Property({ length: 1000 })
description: Opt<string>; // can be used via generics too
// ...
}
Populating relationships
What if we want to fetch the Article
together with the author
relation? We can use populate
hints for that:
// clear the context to simulate fresh request
em.clear();
// find article by id and populate its author
const articleWithAuthor = await em.findOne(Article, article.id, { populate: ['author'] });
console.log(articleWithAuthor);
Run the npm start
as usual:
[query] select `a0`.`id`, `a0`.`created_at`, `a0`.`updated_at`, `a0`.`slug`, `a0`.`title`, `a0`.`description`, `a0`.`author_id` from `article` as `a0` where `a0`.`id` = 1 limit 1 [took 1 ms]
[query] select `u0`.* from `user` as `u0` where `u0`.`id` in (1) order by `u0`.`id` asc [took 0 ms]
Article {
id: 1,
createdAt: 2022-09-11T16:57:57.941Z,
updatedAt: 2022-09-11T16:57:57.941Z,
slug: 'foo-is-bar',
title: 'Foo is Bar',
description: 'Lorem impsum dolor sit amet…',
text: undefined,
author: User {
id: 1,
fullName: 'Foo Bar',
email: 'foo@bar.com',
password: '123456',
bio: ''
}
}
Lazy scalar properties
You can see the text
property being undefined
- this is because we marked it as lazy
, therefore the value is not automatically selected. If we add the text
to populate hint, we will get the value:
const articleWithAuthor = await em.findOne(Article, article.id, {
populate: ['author', 'text'],
});
Or if the entity is already loaded, you can use em.populate()
:
const articleWithAuthor = await em.findOne(Article, article.id, {
populate: ['author'],
});
await em.populate(articleWithAuthor!, ['text']);
Loading strategies
You can see two queries being fired, one for loading the Article
entity, another loading the author
relation (so a User
entity). Why not a single joined query? This is mostly a historic reason, but MikroORM was originally born as a MongoDB-only ORM, and as such, using separate queries for each database table was the best approach to bring such functionality.
This default behavior is called LoadStrategy.SELECT_IN
. If you want to have a single query that will be joining the two tables instead of this, use LoadStrategy.JOINED
:
const articleWithAuthor = await em.findOne(Article, article.id, {
populate: ['author', 'text'],
strategy: LoadStrategy.JOINED,
});
Which yields:
[query] select `a0`.`id`, `a0`.`created_at`, `a0`.`updated_at`, `a0`.`slug`, `a0`.`title`, `a0`.`description`, `a0`.`text`, `a0`.`author_id`, `a1`.`id` as `a1__id`, `a1`.`full_name` as `a1__full_name`, `a1`.`email` as `a1__email`, `a1`.`password` as `a1__password`, `a1`.`bio` as `a1__bio` from `article` as `a0` left join `user` as `a1` on `a0`.`author_id` = `a1`.`id` where `a0`.`id` = 1 [took 1 ms]
Article {
id: 1,
createdAt: 2022-09-11T17:09:10.984Z,
updatedAt: 2022-09-11T17:09:10.984Z,
slug: 'foo-is-bar',
title: 'Foo is Bar',
description: 'Lorem impsum dolor sit amet…',
text: 'Lorem impsum dolor sit amet',
author: User {
id: 1,
fullName: 'Foo Bar',
email: 'foo@bar.com',
password: '123456',
bio: ''
}
}
Both approaches have their pros and cons, use the one that suite best the use case at hand. If you want, you can change the loading strategy globally via
loadStrategy
option in the ORM config.
Serialization
What about the password? Seeing the logger Article
entity with populated author
, there is something we need to fix. We can see the user's password, in plain text! We will need to hash it and ensure it never leaks to the API response by adding hidden
serialization flag. Moreover, we can mark it as lazy
, just like we did with the Article.text
, as we rarely want to select it.
For now, let's use sha256
algorithm which we can create synchronously, and hash the value inside the constructor:
import crypto from 'crypto';
@Entity()
export class User extends BaseEntity<'bio'> {
@Property()
fullName: string;
@Property()
email: string;
@Property({ hidden: true, lazy: true })
password: string;
@Property({ type: 'text' })
bio = '';
constructor(fullName: string, email: string, password: string) {
super();
this.fullName = fullName;
this.email = email;
this.password = User.hashPassword(password);
}
static hashPassword(password: string) {
return crypto.createHmac('sha256', password).digest('hex');
}
}
Now change the part where we create our User
entity, as the constructor is now required:
const user = new User('Foo Bar', 'foo@bar.com', '123456');
console.log(user);
After running npm start
, you can see that the password is hashed, and later when you load the Article.author
, the password is no longer selected:
User {
id: undefined,
createdAt: 2022-09-11T17:22:31.619Z,
updatedAt: 2022-09-11T17:22:31.619Z,
fullName: 'Foo Bar',
email: 'foo@bar.com',
password: 'b946ccc987465afcda7e45b1715219711a13518d1f1663b8c53b848cb0143441',
bio: ''
}
That should be good enough for the time being. Don't worry, we will improve on this, later on, using argon2
via lifecycle hooks, but first things first!
Collections: OneToMany and ManyToMany
You got the User
entity opened, let's add one more property to it. You have the Article.author
property that defines the owning side of this relationship between Article
and User
entities. Now define the inverse side - for ManyToOne relation it is the OneToMany kind - represented by a Collection
of Article
entities:
@Entity()
export class User extends BaseEntity<'bio'> {
// ...
@OneToMany({ mappedBy: 'author' })
articles = new Collection<Article>(this);
}
MikroORM represents the relation via the Collection
class. Before we dive into what it means, let's add one more entity to the Article
module to test the ManyToMany relation too. It will be a Tag
entity, so we can categorize the article based on some dynamically defined tags.
The
Tag
entity semantically belongs to theArticle
module, so let's put it there, to thesrc/modules/article/tag.entity.ts
file.
import { Collection, Entity, ManyToMany, Property } from '@mikro-orm/core';
import { Article } from './article.entity.js';
import { BaseEntity } from '../common/base.entity.js';
@Entity()
export class Tag extends BaseEntity {
@Property({ length: 20 })
name!: string;
@ManyToMany({ mappedBy: 'tags' })
articles = new Collection<Article>(this);
}
And we need to define the owning side too, which is Article.tags
:
@ManyToMany()
tags = new Collection<Tag>(this);
It is enough to point to the owning side via mappedBy
option from the inverse side (or vice versa). If you want to define the relation from owning side, use inversedBy
option. A ManyToMany relation that does not define any of those two is always considered the owning side.
@ManyToMany({ inversedBy: 'articles' })
tags = new Collection<Tag>(this);
Working with collections
The Collection
class implements the iterable protocol, so you can use for of
loop to iterate through it.
Another way to access collection items is to use bracket syntax like when we access array items. Keep in mind that this approach will not check if the collection is initialized, while using the getItems()
method will throw an error in this case.
Note that array access in
Collection
is available only for reading already loaded items, we cannot add new items toCollection
this way.
To get all entities stored in a Collection
, we can use getItems()
method. It will throw in case the Collection
is not initialized. If we want to disable this validation, we can use getItems(false)
. This will give us the entity instances managed by the identity map.
Alternatively, you can use toArray()
which will serialize the Collection
to an array of DTOs. Modifying those will have no effect on the actual entity instances.
em.findOneOrFail()
So far we used em.findOne()
which can return null
if the entity is not found in the database. This results in extensive usage of the non-null assertion operator in TypeScript, which can get messy. A better solution is to use em.findOneOrFail()
, which always returns the entity or throws an error, namely an instance of NotFoundError
provided by the ORM.
// clear the context to simulate fresh request
em.clear();
// populating User.articles collection
const user = await em.findOneOrFail(User, 1, { populate: ['articles'] });
console.log(user);
// or you could lazy load the collection later via `init()` method
if (!user.articles.isInitialized()) {
await user.articles.init();
}
// to ensure collection is loaded (but do nothing if it already is), use `loadItems()` method
await user.articles.loadItems();
for (const article of user.articles) {
console.log(article.title);
console.log(article.author.fullName); // the `article.author` is linked automatically thanks to the Identity Map
}
Collection.init()
will always query the database, whileCollection.loadItems()
does only if the collection is not yet initialized, so callingCollection.loadItems()
is safe without previousisInitialized()
check.
Running this, we get the following:
User {
id: 1,
createdAt: 2022-09-11T18:18:14.376Z,
updatedAt: 2022-09-11T18:18:14.376Z,
fullName: 'Foo Bar',
email: 'foo@bar.com',
password: undefined,
bio: '',
articles: Collection<Article> {
'0': Article {
id: 1,
createdAt: 2022-09-11T18:18:14.384Z,
updatedAt: 2022-09-11T18:18:14.384Z,
slug: 'foo-is-bar',
title: 'Foo is Bar',
description: 'Lorem impsum dolor sit amet…',
text: undefined,
author: [User],
tags: [Collection<Tag>]
},
initialized: true,
dirty: false
}
}
Now try to add some tags to the first article:
// create some tags and assign them to the first article
const [article] = user.articles;
const newTag = em.create(Tag, { name: 'new' });
const oldTag = em.create(Tag, { name: 'old' });
article.tags.add(newTag, oldTag);
await em.flush();
console.log(article.tags);
And just for the sake of it, try to remove one of the tags:
// to remove items from collection, we first need to initialize it, we can use `init()`, `loadItems()` or `em.populate()`
await em.populate(article, ['tags']);
// remove 'old' tag by reference
article.tags.remove(oldTag);
// or via callback
article.tags.remove(t => t.id === oldTag.id);
await em.flush();
Refer to the Collections section in the docs for more information and examples.
Events and life cycle hooks
Time to improve our password hashing. Let's use the argon2
package, which provides hash
and verify
functions. They are both async, so we cannot use them inside the entity constructor like before. Instead, we need to use the lifecycle hooks, namely @BeforeCreate()
and @BeforeUpdate()
.
Don't forget to install the
argon2
package vianpm install argon2
.
The plan is following:
- the password will remain in plaintext when assigned via the constructor
hashPassword
function will become an event handler, we decorate it with@BeforeCreate()
and@BeforeUpdate()
- as such, it will get the
EventArgs
parameter during the flush, we use that to detect if the password property changed - the
args.changeSet
holds theChangeSet
object defining the metadata about the entity and its state ChangeSet.payload
holds the actual computed difference- we add a new
verifyPassword()
method to theUser
entity to later
import { hash, verify } from 'argon2';
export class User extends BaseEntity<'bio'> {
// ...
constructor(fullName: string, email: string, password: string) {
super();
this.fullName = fullName;
this.email = email;
this.password = password; // keep plain text, will be hashed via hooks
}
@BeforeCreate()
@BeforeUpdate()
async hashPassword(args: EventArgs<User>) {
// hash only if the password was changed
const password = args.changeSet?.payload.password;
if (password) {
this.password = await hash(password);
}
}
async verifyPassword(password: string) {
return verify(this.password, password);
}
}
⛳ Checkpoint 2
We added 2 new entities: Article
and Tag
and a common BaseEntity
. You can find working StackBlitz for the current state here:
Due to the nature of how the ESM support in ts-node works, it is not possible to use it inside StackBlitz project - we need to use
node --loader
instead. We also use in-memory database, SQLite feature available via special database name:memory:
.
This is our server.ts
file after this chapter: