Many-to-many relations

Many-to-many (m-n) relations refer to relations where zero or more records on one side of the relation can be connected to zero or more records on the other side.

Prisma schema syntax and the implementation in the underlying database differs between relational databases and MongoDB.

Relational databases

In relational databases, m-n-relations are typically modelled via relation tables. m-n-relations can be either explicit or implicit in the Prisma schema. We recommend using implicit m-n-relations if you do not need to store any additional meta-data in the relation table itself. You can always migrate to an explicit m-n-relation later if needed.

Explicit many-to-many relations

In an explicit m-n relation, the relation table is represented as a model in the Prisma schema and can be used in queries. Explicit m-n relations define three models:

  • Two models with m-n relation, such as Category and Post.
  • One model that represents the relation table, such as CategoriesOnPosts (also sometimes called JOIN, link or pivot table) in the underlying database. The fields of a relation table model are both annotated relation fields (post and category) with a corresponding relation scalar field (postId and categoryId).

The relation table CategoriesOnPosts connects related Post and Category records. In this example, the model representing the relation table also defines additional fields that describe the Post/Category relationship - who assigned the category (assignedBy), and when the category was assigned (assignedAt):

model Post {
id Int @id @default(autoincrement())
title String
categories CategoriesOnPosts[]
}
model Category {
id Int @id @default(autoincrement())
name String
posts CategoriesOnPosts[]
}
model CategoriesOnPosts {
post Post @relation(fields: [postId], references: [id])
postId Int // relation scalar field (used in the `@relation` attribute above)
category Category @relation(fields: [categoryId], references: [id])
categoryId Int // relation scalar field (used in the `@relation` attribute above)
assignedAt DateTime @default(now())
assignedBy String
@@id([postId, categoryId])
}

The underlying SQL looks like this:

CREATE TABLE "Post" (
"id" SERIAL NOT NULL,
"title" TEXT NOT NULL,
CONSTRAINT "Post_pkey" PRIMARY KEY ("id")
);
CREATE TABLE "Category" (
"id" SERIAL NOT NULL,
"name" TEXT NOT NULL,
CONSTRAINT "Category_pkey" PRIMARY KEY ("id")
);
-- Relation table + indexes --
CREATE TABLE "CategoriesOnPosts" (
"postId" INTEGER NOT NULL,
"categoryId" INTEGER NOT NULL,
"assignedAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
CONSTRAINT "CategoriesOnPosts_pkey" PRIMARY KEY ("postId","categoryId")
);
ALTER TABLE "CategoriesOnPosts" ADD CONSTRAINT "CategoriesOnPosts_postId_fkey" FOREIGN KEY ("postId") REFERENCES "Post"("id") ON DELETE RESTRICT ON UPDATE CASCADE;
ALTER TABLE "CategoriesOnPosts" ADD CONSTRAINT "CategoriesOnPosts_categoryId_fkey" FOREIGN KEY ("categoryId") REFERENCES "Category"("id") ON DELETE RESTRICT ON UPDATE CASCADE;

Note that the same rules as for 1-n relations apply (because PostCategoriesOnPosts and CategoryCategoriesOnPosts are both in fact 1-n relations), which means one side of the relation needs to be annotated with the @relation attribute.

When you don't need to attach additional information to the relation, you can model m-n-relations as implicit m-n-relations. If you're not using Prisma Migrate but obtain your data model from introspection, you can still make use of implicit m-n-relations by following Prisma ORM's conventions for relation tables.

Querying an explicit many-to-many

The following section demonstrates how to query an explicit m-n-relation. You can query the relation model directly (prisma.categoriesOnPosts(...)), or use nested queries to go from Post -> CategoriesOnPosts -> Category or the other way.

The following query does three things:

  1. Creates a Post
  2. Creates a new record in the relation table CategoriesOnPosts
  3. Creates a new Category that is associated with the newly created Post record
const createCategory = await prisma.post.create({
data: {
title: 'How to be Bob',
categories: {
create: [
{
assignedBy: 'Bob',
assignedAt: new Date(),
category: {
create: {
name: 'New category',
},
},
},
],
},
},
})

The following query:

  • Creates a new Post
  • Creates a new record in the relation table CategoriesOnPosts
  • Connects the category assignment to existing categories (with IDs 9 and 22)
const assignCategories = await prisma.post.create({
data: {
title: 'How to be Bob',
categories: {
create: [
{
assignedBy: 'Bob',
assignedAt: new Date(),
category: {
connect: {
id: 9,
},
},
},
{
assignedBy: 'Bob',
assignedAt: new Date(),
category: {
connect: {
id: 22,
},
},
},
],
},
},
})

Sometimes you might not know if a Category record exists. If the Category record exists, you want to connect a new Post record to that category. If the Category record does not exist, you want to create the record first and then connect it to the new Post record. The following query:

  1. Creates a new Post
  2. Creates a new record in the relation table CategoriesOnPosts
  3. Connects the category assignment to an existing category (with ID 9), or creates a new category first if it does not exist
const assignCategories = await prisma.post.create({
data: {
title: 'How to be Bob',
categories: {
create: [
{
assignedBy: 'Bob',
assignedAt: new Date(),
category: {
connectOrCreate: {
where: {
id: 9,
},
create: {
name: 'New Category',
id: 9,
},
},
},
},
],
},
},
})

The following query returns all Post records where at least one (some) category assignment (categories) refers to a category named "New category":

const getPosts = await prisma.post.findMany({
where: {
categories: {
some: {
category: {
name: 'New Category',
},
},
},
},
})

The following query returns all categories where at least one (some) related Post record titles contain the words "Cool stuff" and the category was assigned by Bob.

const getAssignments = await prisma.category.findMany({
where: {
posts: {
some: {
assignedBy: 'Bob',
post: {
title: {
contains: 'Cool stuff',
},
},
},
},
},
})

The following query gets all category assignments (CategoriesOnPosts) records that were assigned by "Bob" to one of 5 posts:

const getAssignments = await prisma.categoriesOnPosts.findMany({
where: {
assignedBy: 'Bob',
post: {
id: {
in: [9, 4, 10, 12, 22],
},
},
},
})

Implicit many-to-many relations

Implicit m-n relations define relation fields as lists on both sides of the relation. Although the relation table exists in the underlying database, it is managed by Prisma ORM and does not manifest in the Prisma schema. Implicit relation tables follow a specific convention.

Implicit m-n-relations makes the Prisma Client API for m-n-relations a bit simpler (since you have one fewer level of nesting inside of nested writes).

In the example below, there's one implicit m-n-relation between Post and Category:

Prisma schema
SQL
model Post {
id Int @id @default(autoincrement())
title String
categories Category[]
}
model Category {
id Int @id @default(autoincrement())
name String
posts Post[]
}

Querying an implicit many-to-many

The following section demonstrates how to query an implicit m-n relation. The queries require less nesting than explicit m-n queries.

The following query creates a single Post and multiple Category records:

const createPostAndCategory = await prisma.post.create({
data: {
title: 'How to become a butterfly',
categories: {
create: [{ name: 'Magic' }, { name: 'Butterflies' }],
},
},
})

The following query creates a single Category and multiple Post records:

const createCategoryAndPosts = await prisma.category.create({
data: {
name: 'Stories',
posts: {
create: [
{ title: 'That one time with the stuff' },
{ title: 'The story of planet Earth' },
],
},
},
})

The following query returns all Post records with a list of that post's assigned categories:

const getPostsAndCategories = await prisma.post.findMany({
include: {
categories: true,
},
})

Rules for defining an implicit m-n relation

Implicit m-n relations:

  • Use a specific convention for relation tables

  • Do not require the @relation attribute unless you need to disambiguate relations with a name, e.g. @relation("MyRelation") or @relation(name: "MyRelation").

  • If you do use the @relation attribute, you cannot use the references, fields, onUpdate or onDelete arguments. This is because these take a fixed value for implicit m-n-relations and cannot be changed.

  • Require both models to have a single @id. Be aware that:

    • You cannot use a multi-field ID
    • You cannot use a @unique in place of an @id

    To use either of these features, you must use an explicit m-n instead.

Conventions for relation tables in implicit m-n relations

If you obtain your data model from introspection, you can still use implicit m-n-relations by following Prisma ORM's conventions for relation tables. The following example assumes you want to create a relation table to get an implicit m-n-relation for two models called Post and Category.

Relation table

If you want a relation table to be picked up by introspection as an implicit m-n-relation, the name must follow this exact structure:

  • It must start with an underscore _
  • Then the name of the first model in alphabetical order (in this case Category)
  • Then the relationship (in this case To)
  • Then the name of the second model in alphabetical order (in this case Post)

In the example, the correct table name is _CategoryToPost.

When creating an implicit m-n-relation yourself in the Prisma schema file, you can configure the relation to have a different name. This will change the name given to the relation table in the database. For example, for a relation named "MyRelation" the corresponding table will be called _MyRelation.

Multi-schema

If your implicit many-to-many relationship spans multiple database schemas (using the multiSchema preview feature), the relation table (with the name defined directly above, in the example _CategoryToPost) must be present in the same database schema as the first model in alphabetical order (in this case Category).

Columns

A relation table for an implicit m-n-relation must have exactly two columns:

  • A foreign key column that points to Category called A
  • A foreign key column that points to Post called B

The columns must be called A and B where A points to the model that comes first in the alphabet and B points to the model which comes last in the alphabet.

Indexes

There further must be:

  • A unique index defined on both foreign key columns:

    CREATE UNIQUE INDEX "_CategoryToPost_AB_unique" ON "_CategoryToPost"("A" int4_ops,"B" int4_ops);
  • A non-unique index defined on B:

    CREATE INDEX "_CategoryToPost_B_index" ON "_CategoryToPost"("B" int4_ops);
Example

This is a sample SQL statement that would create the three tables including indexes (in PostgreSQL dialect) that are picked up as a implicit m-n-relation by Prisma Introspection:

CREATE TABLE "_CategoryToPost" (
"A" integer NOT NULL REFERENCES "Category"(id) ,
"B" integer NOT NULL REFERENCES "Post"(id)
);
CREATE UNIQUE INDEX "_CategoryToPost_AB_unique" ON "_CategoryToPost"("A" int4_ops,"B" int4_ops);
CREATE INDEX "_CategoryToPost_B_index" ON "_CategoryToPost"("B" int4_ops);
CREATE TABLE "Category" (
id integer SERIAL PRIMARY KEY
);
CREATE TABLE "Post" (
id integer SERIAL PRIMARY KEY
);

And you can define multiple many-to-many relations between two tables by using the different relationship name. This example shows how the Prisma introspection works under such case:

CREATE TABLE IF NOT EXISTS "User" (
"id" SERIAL PRIMARY KEY
);
CREATE TABLE IF NOT EXISTS "Video" (
"id" SERIAL PRIMARY KEY
);
CREATE TABLE IF NOT EXISTS "_UserLikedVideos" (
"A" SERIAL NOT NULL,
"B" SERIAL NOT NULL,
CONSTRAINT "_UserLikedVideos_A_fkey" FOREIGN KEY ("A") REFERENCES "User" ("id") ON DELETE CASCADE ON UPDATE CASCADE,
CONSTRAINT "_UserLikedVideos_B_fkey" FOREIGN KEY ("B") REFERENCES "Video" ("id") ON DELETE CASCADE ON UPDATE CASCADE
);
CREATE TABLE IF NOT EXISTS "_UserDislikedVideos" (
"A" SERIAL NOT NULL,
"B" SERIAL NOT NULL,
CONSTRAINT "_UserDislikedVideos_A_fkey" FOREIGN KEY ("A") REFERENCES "User" ("id") ON DELETE CASCADE ON UPDATE CASCADE,
CONSTRAINT "_UserDislikedVideos_B_fkey" FOREIGN KEY ("B") REFERENCES "Video" ("id") ON DELETE CASCADE ON UPDATE CASCADE
);
CREATE UNIQUE INDEX "_UserLikedVideos_AB_unique" ON "_UserLikedVideos"("A", "B");
CREATE INDEX "_UserLikedVideos_B_index" ON "_UserLikedVideos"("B");
CREATE UNIQUE INDEX "_UserDislikedVideos_AB_unique" ON "_UserDislikedVideos"("A", "B");
CREATE INDEX "_UserDislikedVideos_B_index" ON "_UserDislikedVideos"("B");

If you run prisma db pull on this database, the Prisma CLI will generate the following schema through introspection:

model User {
id Int @id @default(autoincrement())
Video_UserDislikedVideos Video[] @relation("UserDislikedVideos")
Video_UserLikedVideos Video[] @relation("UserLikedVideos")
}
model Video {
id Int @id @default(autoincrement())
User_UserDislikedVideos User[] @relation("UserDislikedVideos")
User_UserLikedVideos User[] @relation("UserLikedVideos")
}

Configuring the name of the relation table in implicit many-to-many relations

When using Prisma Migrate, you can configure the name of the relation table that's managed by Prisma ORM using the @relation attribute. For example, if you want the relation table to be called _MyRelationTable instead of the default name _CategoryToPost, you can specify it as follows:

model Post {
id Int @id @default(autoincrement())
categories Category[] @relation("MyRelationTable")
}
model Category {
id Int @id @default(autoincrement())
posts Post[] @relation("MyRelationTable")
}

Relation tables

A relation table (also sometimes called a JOIN, link or pivot table) connects two or more other tables and therefore creates a relation between them. Creating relation tables is a common data modelling practice in SQL to represent relationships between different entities. In essence it means that "one m-n relation is modeled as two 1-n relations in the database".

We recommend using implicit m-n-relations, where Prisma ORM automatically generates the relation table in the underlying database. Explicit m-n-relations should be used when you need to store additional data in the relations, such as the date the relation was created.

MongoDB

In MongoDB, m-n-relations are represented by:

  • relation fields on both sides, that each have a @relation attribute, with mandatory fields and references arguments
  • a scalar list of referenced IDs on each side, with a type that matches the ID field on the other side

The following example demonstrates a m-n-relation between posts and categories:

model Post {
id String @id @default(auto()) @map("_id") @db.ObjectId
categoryIDs String[] @db.ObjectId
categories Category[] @relation(fields: [categoryIDs], references: [id])
}
model Category {
id String @id @default(auto()) @map("_id") @db.ObjectId
name String
postIDs String[] @db.ObjectId
posts Post[] @relation(fields: [postIDs], references: [id])
}

Prisma ORM validates m-n-relations in MongoDB with the following rules:

  • The fields on both sides of the relation must have a list type (in the example above, categories have a type of Category[] and posts have a type of Post[])
  • The @relation attribute must define fields and references arguments on both sides
  • The fields argument must have only one scalar field defined, which must be of a list type
  • The references argument must have only one scalar field defined. This scalar field must exist on the referenced model and must be of the same type as the scalar field in the fields argument, but singular (no list)
  • The scalar field to which references points must have the @id attribute
  • No referential actions are allowed in @relation

The implicit m-n-relations used in relational databases are not supported on MongoDB.

Querying MongoDB many-to-many relations

This section demonstrates how to query m-n-relations in MongoDB, using the example schema above.

The following query finds posts with specific matching category IDs:

const newId1 = new ObjectId()
const newId2 = new ObjectId()
const posts = await prisma.post.findMany({
where: {
categoryIDs: {
hasSome: [newId1.toHexString(), newId2.toHexString()],
},
},
})

The following query finds posts where the category name contains the string 'Servers':

const posts = await prisma.post.findMany({
where: {
categories: {
some: {
name: {
contains: 'Servers',
},
},
},
},
})