Datamodel & Migrations

Datamodel (MongoDB)

Overview

The datamodel of your service configuration has two major roles:

  • Define the underlying database schema (the defined models are mapped to MonogDB collections).
  • It is the foundation for the auto-generated CRUD and realtime operations of your Prisma API. Learn more.

The datamodel is written using a subset of the GraphQL Schema Definition Language (SDL) and stored in one or more .prisma-files. These .prisma-files need to be referenced in your 1 under the datamodel property. For example:

endpoint: __YOUR_PRISMA_ENDPOINT__
datamodel: datamodel.prisma
Copy

Building blocks of the datamodel

There are several available building blocks to shape your datamodel.

  • Types consist of multiple fields and typically represent entities from your application domain (e.g. User, Car, Order). Each non-embedded type in your datamodel is mapped to a MongoDB collection and CRUD operations are exposed in the generated Prisam client API.
  • Relations describe relationships between types. With MongoDB, a relation can be modeled as an embedded type or as link relation.
  • Directives covering different use cases such as type constraints or cascading delete behaviour.

Example

A simple example datamodel.prisma file:

type Tweet {
  id: ID! @id
  createdAt: DateTime! @createdAt
  text: String!
  owner: User!
  location: Location!
}

type User {
  id: ID! @id
  createdAt: DateTime! @createdAt
  updatedAt: DateTime! @updatedAt
  handle: String! @unique
  name: String
  tweets: [Tweet!]! @relation(link: INLINE)
}

type Location @embedded {
  latitude: Float!
  longitude: Float!
}

This example is based the on datamodel v1.1 format which is currently in Preview.

This example illustrates a few important concepts when working with your datamodel:

  • The two types Tweet and User are each mapped to a MongoDB collection.
  • There is a bidirectional relation between User and Tweet (via the owner and tweets fields). Note that in the underlying database, only User documents store references to Tweet documents (not the other way around), because the tweets field on User specifies link on the @relation-directive. However, the Prisma API still lets you query the owner of a Tweet directly.
  • There is a unidirectional relation from Tweet to Location (via the location field). Location is an embedded type, which means there are no Location documents are not stored in their own Location collection - instead each Location is stored inside a Tweet document in the Tweet collection.
  • Except for the name field on User, all fields are required in the datamodel (as indicated by the ! following the type).
  • The fields annotated with the @id-, @createdAt- and @updatedAt-directives are managed by Prisma and read-only in the exposed Prisma API, meaning they can not be altered via mutations (except when importing data using the NDF format).
  • The @unique directive expresses a unique constraint, meaning Prisma ensures that there never will be two nodes with the same values for the annotated field.

Creating and updating your datamodel is as simple as writing and saving the datamodel file. Once you're happy with your datamodel, you can save the file and apply the changes to your Prisma service by running prisma1 deploy:

The values of these fields are currently read-only (except when importing data using the NDF format) in the Prisma API but will be made configurable in the future. See this proposal for more information.

$ prisma1 deploy

Changes:

  Tweet (Type)
  + Created type `Tweet`
  + Created field `id` of type `GraphQLID!`
  + Created field `createdAt` of type `DateTime!`
  + Created field `text` of type `String!`
  + Created field `owner` of type `Relation!`
  + Created field `location` of type `Relation!`
  + Created field `updatedAt` of type `DateTime!`

  User (Type)
  + Created type `User`
  + Created field `id` of type `GraphQLID!`
  + Created field `createdAt` of type `DateTime!`
  + Created field `updatedAt` of type `DateTime!`
  + Created field `handle` of type `String!`
  + Created field `name` of type `String`
  + Created field `tweets` of type `[Relation!]!`

  Location (Type)
  + Created type `Location`
  + Created field `latitude` of type `Float!`
  + Created field `longitude` of type `Float!`
  + Created field `id` of type `GraphQLID!`
  + Created field `updatedAt` of type `DateTime!`
  + Created field `createdAt` of type `DateTime!`

  TweetToUser (Relation)
  + Created relation between Tweet and User

  LocationToTweet (Relation)
  + Created relation between Location and Tweet

Applying changes... (22/22)
Applying changes... 0.4s

Files

You can write your datamodel in a single .prisma-file or split it accross multiple ones.

The .prisma-files containing the datamodel need to be referenced in your prisma.yml under the datamodel property. For example:

datamodel:
  - types.prisma
  - enums.prisma

If there is only a single file that defines the datamodel, it can be specified as follows:

datamodel: datamodel.prisma

Object types

An object type (or short type) defines the structure for one model in your datamodel. It is used to represent entities from your application domain.

Each object type is mapped to a collection in your MongoDB database. Note that Prisma enforces a schema even for schemaless databases!

A type has a name and one or multiple fields. Type names can only contain alphanumeric characters and need to start with an uppercase letter. They can contain at most 64 characters.

An instantiation of a type is called a node. This term refers to a node inside your data graph.

Defining an object type

A object type is defined in the datamodel with the keyword type:

type Article {
  id: ID! @id
  title: String!
  text: String
  isPublished: Boolean! @default(value: false)
}

The type defined above has the following properties:

  • Name: Article
  • Fields: id, title, text and isPublished (with the default value false)

id and title and isPublished are required (as indicated by the ! following the type), text is optional.

Generated API operations for types

The types in your datamodel affect the available operations in the Prisma API. Here is an overview of the generated CRUD and realtime operations for every type in your Prisma API:

  • Queries let you fetch one or many nodes of that type
  • Mutations let you create, update or delete nodes of that type
  • Subscriptions let you get notified of changes to nodes of that type (i.e. new nodes are created or existing nodes are updated or deleted)

Fields

Fields are the building blocks of a type, giving a node its shape. Every field is referenced by its name and is either scalar or a relation field.

Field names can only contain alphanumeric characters and need to start with a lowercase letter. They can contain at most 64 characters.

Scalar fields

String

A String holds text. This is the type you would use for a username, the content of a blog post or anything else that is best represented as text.

String values are currently limited to 256KB in size on Demo servers. This limit can be increased on other clusters using the cluster configuration.

Here is an example of a String scalar definition:

type User {
  name: String
}

Integer

An Int is a number that cannot have decimals. Use this to store values such as the weight of an ingredient required for a recipe or the minimum age for an event.

Int values range from -2147483648 to 2147483647.

Here is an example of an Int scalar definition:

type User {
  age: Int
}

Float

A Float is a number that can have decimals. Use this to store values such as the price of an item in a store or the result of complex calculations.

In queries or mutations, Float fields have to be specified without any enclosing characters and an optional decimal point: float: 42, float: 4.2.

Here is an example of a Float scalar definition:

type Item {
  price: Float
}

Boolean

A Boolean can have the value true or false. This is useful to keep track of settings such as whether the user wants to receive an email newsletter or if a recipe is appropriate for vegetarians.

Here is an example of a Boolean scalar definition:

type User {
  overEighteen: Boolean
}

DateTime

The DateTime type can be used to store date and/or time values. A good example might be a person's date of birth or the time/data when a specific event is happening.

Here is an example of a DateTime scalar definition:

type User {
  birthday: DateTime
}

When used as arguments in an operation, DateTime fields have to be specified in ISO 8601 format and are typically passed as strings, here are a few examples:

  • "2015"
  • "2015-11"
  • "2015-11-22"
  • "2015-11-22T13:57:31.123Z".

Enum

Like a Boolean an Enum can have one of a predefined set of values. The difference is that you can define the possible values (whereas for a Boolean the options are restriced to true and false). For example you could specify how an article should be formatted by creating an Enum with the possible values COMPACT, WIDE and COVER.

Enum values can only contain alphanumeric characters and underscores and need to start with an uppercase letter. The name of an enum value can be used in query filters and mutations. They can contain at most 191 characters.

Here is an example of an enum definition:

enum ArticleFormat {
  COMPACT
  WIDE
  COVER
}

type Article {
  format: ArticleFormat
}

Json

Sometimes you might need to store arbitrary JSON values for loosely structured data. The Json type makes sure that it is actually valid JSON and returns the value as a parsed JSON object/array instead of a string.

Json values are currently limited to 256KB in size.

Here is an example of a Json definition:

type Item {
  data: Json
}

ID

An ID value is a generated unique 25-character string based on cuid. Fields annotated with the @id-directive are system fields.

ID fields can only be used once on a type and always need to be annotated with the @id directive:

type User {
  id: ID! @id
}

Type modifiers

In a field definition, a type can be annotated with a type modifier. SDL supports two type modifiers:

  • Required fields: Annotate the type with a !, e.g. name: String!
  • Lists: Annotate the type with a pair of enclosing [], e.g. friends: [User]

List

Scalar fields can be marked with the list field type. A field of a relation that has the many multiplicity will also be marked as a list.

You will often find list definitions looking similar to this:

type Article {
  tags: [String!]!
}

Notice the two ! type modifiers, here is what they express:

  • The first ! type modifier (right after String) means that no item in the list can be null, e.g. this value for tags would not be valid: ["Software", null, "Prisma"]
  • The second ! type modifier (after the closing square bracket) means that the list itself can never be null, it might be empty though. Consequently, null is not a valid value for the tags field but [] is.

Required

Fields can be marked as required (sometimes also referred to as "non-null"). When creating a new node, you need to supply a value for fields which are required and don't have a default value.

Required fields are marked using a ! after the field type:

type User {
  name: String!
}

When adding a required field to a model that already contains nodes, you receive this error message: "You are creating a required field but there are already nodes present that would violate that constraint."

This is because all existing nodes would receive a null value for this field. This would violate the constraint of this field being required (or non-nullable).

Here are the steps that are needed to add a required field:

  1. Add the field being optional
  2. Use updateManyXs to migrate the field of all nodes from null to a non-null value
  3. Now you can mark the field as required and deploy as expected

A more convenient workflow is discussed in this feature request on Github.

Field constraints

Fields can be configured with field constraints to add further semantics and enforce certain rules in your datamodel.

Unique

Setting the unique constraint makes sure that two nodes of the type in question cannot have the same value for a certain field. The only exception is the null value, meaning that multiple nodes can have the value null without violating the constraint.

Unique fields have a unique index applied in the underlying database.

A typical example would be an email field on the User type where the assumption is that every User should have a globally unique email address.

Only the first 191 characters in a String field are considered for uniqueness and the unique check is case insensitive. Storing two different strings is not possible if the first 191 characters are the same or if they only differ in casing.

To mark a field as unique, simply append the @unique directive to its definition:

type User {
  id: ID! @id
  email: String! @unique
  name: String!
}

For every field that's annotated with @unique, you're able to query the corresponding node by providing a value for that field as a query argument.

For example, considering the above datamodel, you can now retrieve a particular User node by its email address:

TypeScript
JavaScript
Flow
Go
const user = await prisma.user({
  email: 'alice@prisma.io',
})
Copy

More constraints

More database constraints will be added soon. Please join the discussion in this feature request if you have wish to see certain constraints implemented in Prisma.

Default value

You can set a default value for non-list scalar fields. The value will be applied to newly created nodes when no value was supplied during the create-mutation.

To specify a default value for a field, you can use the @default directive:

type Story {
  isPublished: Boolean @default(value: false)
  someNumber: Int! @default(value: 42)
  title: String! @default(value: "My New Post")
  publishDate: DateTime! @default(value: "2018-01-26")
}

System fields

Fields annotated with the @id-, @createdAt- or @updatedAt-directives have special semantics in Prisma.

Even when not specified on a type, the underlying database will always manage fields with their behaviour. In case you decide to add these fields later, all existing documents will therefore already maintain those fields.

System field: @id

A node will automatically get assigned a globally unique identifier when it's created, this identifier is stored in the field annotated with the @id-directive. For MongoDB, the field that's annotated with the @id-directive is mapped to the MongoDB object ID which is stored in a field called _id.

Values for fields annotated with the @id-directive have the following properties:

  • Consist of 25 alphanumeric characters (letters are always lowercase)
  • Always start with a (lowercase) letter, e.g. c
  • Follows cuid (collision resistant unique identifiers) scheme

System fields: createdAt and updatedAt

The datamodel further provides two special directives which you can add to your fields:

  • @createdAt: Fields annotated with this directive store the exact date and time for when a node of this object type was created.
  • @updatedAt: Fields annotated with this directive store the exact date and time for when a node of this object type was last updated.

If you want your types to expose these fields, you can simply add them to the type definition, for example:

type User {
  created_at: DateTime! @createdAt
  updated_at: DateTime! @updatedAt
  # ... other fields
}

Relations

A relation defines the semantics of a connection between two types. Two types in a relation are connected via a relation field. When a relation is ambiguous, the relation field needs to be annotated with the @relation directive to disambiguate it.

Relations in MongoDB

One of the biggest differences between document and relational databases is the way how relations between data types are handled.

While relational databases use database normalization to store flat data records that reference each other via keys, document databases are able to store an object physically co-located to a related object within the same collection. The latter is called embedding data (i.e. a document in a collection can have embedded sub-documents/arrays which live inside the same collection).

With MongoDB, relations can be expressed either by embedding data directly in a parent document or by using references. A good overview on the differences can be found in the MongoDB documentation. Prisma adopts the idea of embedded types when modeling data for an underlying MongoDB.

Embedded types

The MongoDB connector introduces the concept of embedded types. An embedded type ...

  • ... always is annotated with the @embedded directive.
  • ... always has (at least) one parent type.
  • ... always is stored directly within its parent type's collection in the underlying Mongo database (i.e. an embedded type never has its own collection).
  • ... can not have unique fields (i.e. fields annotated with the @unique directive).
  • ... can not have have a (back-)relation to its parent type (but it can have relations to other non-embedded types).
  • ... can not be queried directly using the Prisma API but only via nested operations through the parent type.
  • ... can not be created, updated or deleted directly using the Prisma API but only via nested operations through the parent type.

Here is an example of a datamodel where Coordinates is defined as an embedded type:

type City {
  id: ID! @id
  name: String!
  coordinates: Coordinates
}

type Coordinates @embedded {
  latitude: Float!
  longitude: Float!
}

Here is an example of the data that's stored in the underlying MongoDB database based on this datamodel:

With this setup, it is not possible to query any instances of Coordinates directly because Coordinates is an embedded type. Coordinates can only be queried via the City type. Similarly, you can not create, update or delete coordinates directly, instead you need to create, update or delete a City in order to perform such operation on an instance of Coordinates.

You can learn more about embedded types in the specification.

With MongoDB, you can model relations in two ways:

  • Using an embedded type as explained above
  • Using references which are called links in Prisma terminology

A link relation with the MongoDB connector works in the way that:

  • One side (A) of the relation stores the ID of the document on the other side (B), this is called an inlined link
  • The other side (B) of the relation has no reference at all to the document on the initial side (A)
  • Each side of the relation is represented by its own collection in the underlying MongoDB, i.e. a link relation always spans accross multiple collections.

You can denote the side of the relation that should store the ID using the link parameter of the @relation directive. In the following example, the User type stores the ID values of all the Post documents it's related to. A Post document however doesn't store any information about its author in the underlying Mongo database:

type User {
  id: ID! @id
  name: String!
  posts: [Post!]! @relation(link: INLINE)
}

type Post {
  id: ID! @id
  title: String!
  author: User!
}

Here is an example of the data that's stored in the underlying MongoDB database based on this datamodel:

While this approach enables querying for Post documents directly in the Prisma API (as opposed to embedded types which can only be queried through nested operations via their parent types), there are performance considerations when modeling relations this way.

Operations that go from a Post to a User via the author field will be more expensive. This is because the underlying Post objects don't know who their authors are and Prisma needs to filter all Users to find who the author of Post is.

Here is an example for a many-to-many relation:

type Article {
  id: ID! @id
  categories: [Category!]! @relation(link: INLINE)
}

type Category {
  id: ID! @id
  articles: [Article!]!
}

Here is an example for a one-to-one relation:

type Human {
  id: ID! @id
  brain: Brain! @relation(link: INLINE)
}

type Brain {
  id: ID! @id
  human: Human!
}

Self-relations

A relation can also connect a type with itself. It is then referred to as a self-relation.

Note that self-relations also need to include the link argument on the @relation directive:

type User {
  id: ID! @id
  friends: [User!]! @relation(link: INLINE)
}

A self-relation can also be bidirectional:

type User {
  id: ID! @id
  following: [User!]! @relation(name: "FollowRelation", link: INLINE)
  followers: [User!]! @relation(name: "FollowRelation")
}

Note that in this case the relation needs to be annotated with the @relation directive and the name argument has to be provided.

Required relations

For a to-one relation field, you can configure whether it is required or optional. The ! type modifier means that this field can never be null. A field for the address of a user would therefore be of type Address or Address!.

Nodes for a type that contains a required to-one relation field can only be created using a nested mutation to ensure the respective field will not be null.

Consider again the following relation:

type User {
  id: ID! @id
  car: Car! @relation(link: INLINE)
}

type Car {
  id: ID! @id
  owner: User!
  color: String!
}

A Car can never be created without a User and the other way around because that would violate the required constraint. You therefore need to create both at the same time using a declarative nested write:

TypeScript
JavaScript
Flow
Go
const newUser = await prisma.createUser({
  car: {
    create: {
      color: 'Yellow',
    },
  },
})
Copy

Note that a to-many relation field is always set to required. For example, a field that contains many user addresses always uses the type [Address!]! and can never be of type [Address!], [Address]! or [Address].

The @relation directive

When defining relations between types, you can use the @relation directive which provides meta-information about the relation. Sometimes, the @relation directive can be required, e.g. if a relation is ambiguous, you must use the @relation directive to disambiguate it. Or if you're defining a link relation, one side of the relation must be annotated with the @relation directive to specify the link argument.

It can take two arguments:

  • name (required): An identifier for this relation, provided as a string.
  • link (required on exactly one side of a link relation): Specifies which side of the relation should store references to the other side.
  • onDelete: Specifies the deletion behaviour and enables cascading deletes. In case a node with related nodes gets deleted, the deletion behaviour determines what should happen to the related nodes. The input values for this argument are defined as an enum with the following possible values:

    • SET_NULL (default): Set the related node(s) to null.
    • CASCADE: Delete the related node(s). Note that is not possible to set both ends of a bidirectional relation to CASCADE.

Here is an example of a datamodel where the @relation directive is used:

type User {
  id: ID! @id
  stories: [Story!]!
    @relation(name: "StoriesByUser", link: INLINE, onDelete: CASCADE)
}

type Story {
  id: ID! @id
  text: String!
  author: User @relation(name: "StoriesByUser")
}

The relation is named StoriesByUser and the deletion behaviour is as follows:

  • When a User node gets deleted, all its related Story nodes will be deleted as well.
  • When a Story node gets deleted, it will simply be removed from the stories list on the related User node.

It is currently not possible to rename relations that are specified via the @relation directive.

Using the name argument of the @relation directive

In certain cases, your datamodel may contain ambiguous relations. For example, consider you not only want a relation to express the "author-relationship" between User and Story, but you also want a relation to express which Story nodes have been liked by a User.

In that case, you end up with two different relations between User and Story! In order to disambiguate them, you need to give the relation a name:

type User {
  id: ID! @id
  writtenStories: [Story!]! @relation(name: "WrittenStories", link: INLINE)
  likedStories: [Story!]! @relation(name: "LikedStories", link: INLINE)
}

type Story {
  id: ID! @id
  text: String!
  author: User! @relation(name: "WrittenStories")
  likedBy: [User!]! @relation(name: "LikedStories")
}

If the name wasn't provided in this case, there would be no way to decide whether writtenStories should relate to the author or the likedBy field.

Using the onDelete argument of the @relation directive

As mentioned above, you can specify a dedicated deletion behaviour for the related nodes. That's what the onDelete argument of the @relation directive is for.

Consider the following example:

type User {
  id: ID! @id
  comments: [Comment!]!
    @relation(name: "CommentAuthor", onDelete: CASCADE, link: INLINE)
  blog: Blog @relation(name: "BlogOwner", onDelete: CASCADE, link: INLINE)
}

type Blog {
  id: ID! @id
  comments: [Comment!]!
    @relation(name: "Comments", onDelete: CASCADE, link: INLINE)
  owner: User! @relation(name: "BlogOwner", onDelete: SET_NULL)
}

type Comment {
  id: ID! @id
  blog: Blog! @relation(name: "Comments", onDelete: SET_NULL)
  author: User @relation(name: "CommentAuthor", onDelete: SET_NULL)
}

Let's investigate the deletion behaviour for the three types:

  • When a User node gets deleted,

    • all related Comment nodes will be deleted.
    • the related Blog node will be deleted.
  • When a Blog node gets deleted,

    • all related Comment nodes will be deleted.
    • the related User node will have its blog field set to null.
  • When a Comment node gets deleted,

    • the related Blog node continues to exist and the deleted Comment node is removed from its comments list.
    • the related User node continues to exist and the deleted Comment node is removed from its comments list.

SDL directives

Directives are used to provide additional information in your datamodel. They look like this: @name(argument: "value") or simply @name when there are no arguments.

Unique scalar fields

The @unique directive marks a scalar field as unique. Unique fields will have a unique index applied in the underlying database.

# the `User` type has a unique `email` field
type User {
  email: String @unique
}

Find more info about the @unique directive above.

Relation fields

The directive @relation(name: String, onDelete: ON_DELETE! = SET_NULL) can be attached to a relation field.

See above for more information.

Default value for scalar fields

The directive @default(value: <type>!) sets a default value for a scalar field:

type Post {
  title: String! @default(value: "New Post")
  published: Boolean! @default(value: false)
  someNumber: Int! @default(value: 42)
}

Naming conventions

Different objects you encounter in a Prisma service like types or relations follow separate naming conventions to help you distinguish them.

Types

The type name determines the name of derived queries and mutations as well as the argument names for nested mutations.

Here are the conventions for naming types:

  • Choose type names in singular:

    • Yes: type User { ... }
    • No: type Users { ... }

Scalar and relation fields

The name of a scalar field is used in queries and in query arguments of mutations. The name of relation fields follows the same conventions and determines the argument names for relation mutations.Relation field names can only contain alphanumeric characters and need to start with an uppercase letter. They can contain at most 64 characters. Field names are unique per type.

Here are the conventions for naming fields:

  • Choose plural names for list fields:

    • Yes: friends: [User!]!
    • No: friendList: [User!]!
  • Choose singular names for non-list fields:

    • Yes: post: Post!
    • No: posts: Post!

More SDL features

In this section, we describe further SDL features that are not yet supported for data modeling with Prisma.

Interfaces

"Like many type systems, [SDL] supports interfaces. An interface is an abstract type that includes a certain set of fields that a type must include to implement the interface." From the official Documentation

To learn more about when and how interfaces are coming to Prisma, check out this feature request.

Union types

"Union types are very similar to interfaces, but they don't get to specify any common fields between the types." From the official Documentation

To learn more about when and how union types are coming to Prisma, check out this feature request.