Prisma Plugin for Pothos

This plugin provides tighter integration with prisma, making it easier to define prisma based object types, and helps solve n+1 queries for relations. It also has integrations for the relay plugin to make defining nodes and connections easy and efficient.

This plugin is NOT required to use prisma with Pothos, but does make things a lot easier and more efficient. See the Using Prisma without a plugin section below for more details.

Features

🎨 Quickly define GraphQL types based on your Prisma models
🦺 Strong type-safety throughout the entire API
🤝 Automatically resolve relationships defined in your database
🎣 Automatic Query optimization to efficiently load the specific data needed to resolve a query (solves common N+1 issues)
💅 Types and fields in GraphQL schema are not implicitly tied to the column names or types in your database.
🔀 Relay integration for defining nodes and connections that can be efficiently loaded.
📚 Supports multiple GraphQL models based on the same Database model
🧮 Count fields can easily be added to objects and connections

Example

Here is a quick example of what an API using this plugin might look like. There is a more thorough breakdown of what the methods and options used in the example below.

// Create an object type based on a prisma model
// without providing any custom type information
builder.prismaObject('User', {
  fields: (t) => ({
    // expose fields from the database
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    bio: t.string({
      // automatically load the bio from the profile
      // when this field is queried
      select: {
        profile: {
          select: {
            bio: true,
          },
        },
      },
      // user will be typed correctly to include the
      // selected fields from above
      resolve: (user) => user.profile.bio,
    }),
    // Load posts as list field.
    posts: t.relation('posts', {
      args: {
        oldestFirst: t.arg.boolean(),
      },
      // Define custom query options that are applied when
      // loading the post relation
      query: (args, context) => ({
        orderBy: {
          createdAt: args.oldestFirst ? 'asc' : 'desc',
        },
      }),
    }),
    // creates relay connection that handles pagination
    // using prisma's built in cursor based pagination
    postsConnection: t.relatedConnection('posts', {
      cursor: 'id',
    }),
  }),
});

// Create a relay node based a prisma model
builder.prismaNode('Post', {
  id: { field: 'id' },
  fields: (t) => ({
    title: t.exposeString('title'),
    author: t.relation('author'),
  }),
});

builder.queryType({
  fields: (t) => ({
    // Define a field that issues an optimized prisma query
    me: t.prismaField({
      type: 'User',
      resolve: async (query, root, args, ctx, info) =>
        prisma.user.findUniqueOrThrow({
          // the `query` argument will add in `include`s or `select`s to
          // resolve as much of the request in a single query as possible
          ...query,
          where: { id: ctx.userId },
        }),
    }),
  }),
});

Given this schema, you would be able to resolve a query like the following with a single prisma query (which will still result in a few optimized SQL queries).

query {
  me {
    email
    posts {
      title
      author {
        id
      }
    }
  }
}

A query like

query {
  me {
    email
    posts {
      title
      author {
        id
      }
    }
    oldPosts: posts(oldestFirst: true) {
      title
      author {
        id
      }
    }
  }
}

Will result in 2 calls to prisma, one to resolve everything except oldPosts, and a second to resolve everything inside oldPosts. Prisma can only resolve each relation once in a single query, so we need a separate to handle the second posts relation.

Install

yarn add @pothos/plugin-prisma

Setup

This plugin requires a little more setup than other plugins because it integrates with the prisma to generate some types that help the plugin better understand your prisma schema. Previous versions of this plugin used to infer all required types from the prisma client itself, but this resulted in a poor dev experience because the complex types slowed down editors, and some more advanced use cases could not be typed correctly.

Add a the `pothos` generator to your prisma schema

generator pothos {
  provider = "prisma-pothos-types"
}

Now the types Pothos uses will be generated whenever you re-generate your prisma client. Run the following command to re-generate the client and create the new types:

npx prisma generate

additional options:

clientOutput: Where the generated code will import the PrismaClient from. The default is the full path of wherever the client is generated. If you are checking in the generated file, using @prisma/client is a good option.
output: Where to write the generated types

Example with more options:

generator pothos {
  provider = "prisma-pothos-types"
  clientOutput = "@prisma/client"
  output = "./pothos-types.ts"
}

Set up the builder

import SchemaBuilder from '@pothos/core';
import { PrismaClient } from '@prisma/client';
import PrismaPlugin from '@pothos/plugin-prisma';
// This is the default location for the generator, but this can be
// customized as described above.
// Using a type only import will help avoid issues with undeclared
// exports in esm mode
import type PrismaTypes from '@pothos/plugin-prisma/generated';

const prisma = new PrismaClient({});

const builder = new SchemaBuilder<{
  PrismaTypes: PrismaTypes;
}>({
  plugins: [PrismaPlugin],
  prisma: {
    client: prisma,
    // defaults to false, uses /// comments from prisma schema as descriptions
    // for object types, relations and exposed fields.
    // descriptions can be omitted by setting description to false
    exposeDescriptions: boolean | { models: boolean, fields: boolean },
    // use where clause from prismaRelatedConnection for totalCount (will true by default in next major version)
    filterConnectionTotalCount: true,
  },
});

It is strongly recommended NOT to put your prisma client into Context. This will result in slower type-checking and a laggy developer experience in VSCode. See this issue for more details.

You can also load or create the prisma client dynamically for each request. This can be used to periodically re-create clients or create read-only clients for certain types of users.

import SchemaBuilder from '@pothos/core';
import { PrismaClient, Prisma } from '@prisma/client';
import PrismaPlugin from '@pothos/plugin-prisma';
import type PrismaTypes from '@pothos/plugin-prisma/generated';

const prisma = new PrismaClient({});

const readOnlyPrisma = new PrismaClient({
  datasources: {
    db: {
      url: process.env.READ_ONLY_REPLICA_URL,
    },
  },
});

const builder = new SchemaBuilder<{
  Context: { user: { isAdmin: boolean } };
  PrismaTypes: PrismaTypes;
}>({
  plugins: [PrismaPlugin],
  prisma: {
    client: (ctx) => (ctx.user.isAdmin ? prisma : readOnlyPrisma),
    // Because the prisma client is loaded dynamically, we need to explicitly provide the some information about the prisma schema
    dmmf: Prisma.dmmf,
    // use where clause from prismaRelatedConnection for totalCount (will true by default in next major version)
    filterConnectionTotalCount: true,
  },
});

Creating types with `builder.prismaObject`

builder.prismaObject takes 2 arguments:

name: The name of the prisma model this new type represents
options: options for the type being created, this is very similar to the options for any other object type

builder.prismaObject('User', {
  // Optional name for the object, defaults to the name of the prisma model
  name: 'PostAuthor',
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
  }),
});

builder.prismaObject('Post', {
  fields: (t) => ({
    id: t.exposeID('id'),
    title: t.exposeString('title'),
  }),
});

So far, this is just creating some simple object types. They work just like any other object type in Pothos. The main advantage of this is that we get the type information without using object refs, or needing imports from prisma client.

Adding prisma fields to non-prisma objects (including Query and Mutation)

There is a new t.prismaField method which can be used to define fields that resolve to your prisma types:

builder.queryType({
  fields: (t) => ({
    me: t.prismaField({
      type: 'User',
      resolve: async (query, root, args, ctx, info) =>
        prisma.user.findUniqueOrThrow({
          ...query,
          where: { id: ctx.userId },
        }),
    }),
  }),
});

This method works just like th normal t.field method with a couple of differences:

The type option must contain the name of the prisma model (eg. User or [User] for a list field).
The resolve function has a new first argument query which should be spread into query prisma query. This will be used to load data for nested relationships.

You do not need to use this method, and the builder.prismaObject method returns an object ref than can be used like any other object ref (with t.field), but using t.prismaField will allow you to take advantage of more efficient queries.

The query object will contain an object with include or select options to pre-load data needed to resolve nested parts of the current query. The included/selected fields are based on which fields are being queried, and the options provided when defining those fields and types.

Adding relations

You can add fields for relations using the t.relation method:

builder.queryType({
  fields: (t) => ({
    me: t.prismaField({
      type: 'User',
      resolve: async (query, root, args, ctx, info) =>
        prisma.user.findUniqueOrThrow({
          ...query,
          where: { id: ctx.userId },
        }),
    }),
  }),
});

builder.prismaObject('User', {
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    posts: t.relation('posts'),
  }),
});

builder.prismaObject('Post', {
  fields: (t) => ({
    id: t.exposeID('id'),
    title: t.exposeString('title'),
    author: t.relation('author'),
  }),
});

t.relation defines a field that can be pre-loaded by a parent resolver. This will create something like { include: { author: true }} that will be passed as part of the query argument of a prismaField resolver. If the parent is another relation field, the includes will become nested, and the full relation chain will be passed to the prismaField that started the chain.

For example the query:

query {
  me {
    posts {
      author {
        id
      }
    }
  }
}

the me prismaField would receive something like the following as its query parameter:

{
  include: {
    posts: {
      include: {
        author: true;
      }
    }
  }
}

This will work perfectly for the majority of queries. There are a number of edge cases that make it impossible to resolve everything in a single query. When this happens Pothos will automatically construct an additional query to ensure that everything is still loaded correctly, and split into as few efficient queries as possible. This process is described in more detail below

Fallback queries

There are some cases where data can not be pre-loaded by a prisma field. In these cases, pothos will issue a findUnique query for the parent of any fields that were not pre-loaded, and select the missing relations so those fields can be resolved with the correct data. These queries should be very efficient, are batched by pothos to combine requirements for multiple fields into one query, and batched by Prisma to combine multiple queries (in an n+1 situation) to a single sql query.

The following are some edge cases that could cause an additional query to be necessary:

The parent object was not loaded through a field defined with t.prismaField, or t.relation
The root prismaField did not correctly spread the query arguments in is prisma call.
The query selects multiple fields that use the same relation with different filters, sorting, or limits
The query contains multiple aliases for the same relation field with different arguments in a way that results in different query options for the relation.
A relation field has a query that is incompatible with the default includes of the parent object

All of the above should be relatively uncommon in normal usage, but the plugin ensures that these types of edge cases are automatically handled when they do occur.

Filters, Sorting, and arguments

So far we have been describing very simple queries without any arguments, filtering, or sorting. For t.prismaField definitions, you can add arguments to your field like normal, and pass them into your prisma query as needed. For t.relation the flow is slightly different because we are not making a prisma query directly. We do this by adding a query option to our field options. Query can either be a query object, or a method that returns a query object based on the field arguments.

builder.prismaObject('User', {
  fields: (t) => ({
    id: t.exposeID('id'),
    posts: t.relation('posts', {
      // We can define arguments like any other field
      args: {
        oldestFirst: t.arg.boolean(),
      },
      // Then we can generate our query conditions based on the arguments
      query: (args, context) => ({
        orderBy: {
          createdAt: args.oldestFirst ? 'asc' : 'desc',
        },
      }),
    }),
  }),
});

The returned query object will be added to the include section of the query argument that gets passed into the first argument of the parent t.prismaField, and can include things like where, skip, take, abd orderBy. The query function will be passed the arguments for the field, and the context for the current request. Because it is used for pre-loading data, and solving n+1 issues, it can not be passed the parent object because it may not be loaded yet.

builder.prismaObject('User', {
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    posts: t.relation('posts', {
      // We can define arguments like any other field
      args: {
        oldestFirst: t.arg.boolean(),
      },
      // Then we can generate our query conditions based on the arguments
      query: (args, context) => ({
        orderBy: {
          createdAt: args.oldestFirst ? 'asc' : 'desc',
        },
      }),
    }),
  }),
});

relationCount

Prisma supports querying for relation counts which allow including counts for relations along side other includes. Before prisma 4.2.0, this does not support any filters on the counts, but can give a total count for a relation. Starting from prisma 4.2.0, filters on relation count are available under the filteredRelationCount preview feature flag.

builder.prismaObject('User', {
  fields: (t) => ({
    id: t.exposeID('id'),
    postCount: t.relationCount('posts', {
      where: {
        published: true,
      },
    }),
  }),
});

Includes on types

In some cases, you may want to always pre-load certain relations. This can be helpful for defining fields directly on type where the underlying data may come from a related table.

builder.prismaObject('User', {
  // This will always include the profile when a user object is loaded.  Deeply nested relations can
  // also be included this way.
  include: {
    profile: true,
  },
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    bio: t.string({
      // The profile relation will always be loaded, and user will now be typed to include the
      // profile field so you can return the bio from the nested profile relation.
      resolve: (user) => user.profile.bio,
    }),
  }),
});

Select mode for types

By default, the prisma plugin will use include when including relations, or generating fallback queries. This means we are always loading all columns of a table when loading it in a t.prismaField or a t.relation. This is usually what we want, but in some cases, you may want to select specific columns instead. This can be useful if you have tables with either a very large number of columns, or specific columns with large payloads you want to avoid loading.

To do this, you can add a select instead of an include to your prismaObject:

builder.prismaObject('User', {
  select: {
    id: true,
  },
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
  }),
});

The t.expose* and t.relation methods will all automatically add selections for the exposed fields WHEN THEY ARE QUERIED, ensuring that only the requested columns will be loaded from the database.

In addition to the t.expose and t.relation, you can also add custom selections to other fields:

builder.prismaObject('User', {
  select: {
    id: true,
  },
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    bio: t.string({
      // This will select user.profile.bio when the the `bio` field is queried
      select: {
        profile: {
          select: {
            bio: true,
          },
        },
      },
      resolve: (user) => user.profile.bio,
    }),
  }),
});

Using arguments or context in your selections

The following is a slightly contrived example, but shows how arguments can be used when creating a selection for a field:

const PostDraft = builder.prismaObject('Post', {
  fields: (t) => ({
    title: t.exposeString('title'),
    commentFromDate: t.string({
      args: {
        date: t.arg({ type: 'Date', required: true }),
      },
      select: (args) => ({
        comments: {
          take: 1,
          where: {
            createdAt: {
              gt: args.date,
            },
          },
        },
      }),
      resolve: (post) => post.comments[0]?.content,
    }),
  }),
});

Extending prisma objects

The normal builder.objectField(s) methods can be used to extend prisma objects, but do not support using selections, or exposing fields not in the default selection. To use these features, you can use

builder.prismaObjectField or builder.prismaObjectFields instead.

Type variants

The prisma plugin supports defining multiple GraphQL types based on the same prisma model. Additional types are called variants. You will always need to have a "Primary" variant (defined as described above). Additional variants can be defined by providing a variant option instead of a name option when creating the type:

const Viewer = builder.prismaObject('User', {
  variant: 'Viewer',
  fields: (t) => ({
    id: t.exposeID('id'),
});

You can define variant fields that reference one variant from another:

const Viewer = builder.prismaObject('User', {
  variant: 'Viewer',
  fields: (t) => ({
    id: t.exposeID('id'),
    // Using the model name ('User') will reference the primary variant
    user: t.variant('User'),
  });
});

const User = builder.prismaNode('User', {
  id: {
    resolve: (user) => user.id,
  },
  fields: (t) => ({
    // To reference another variant, use the returned object Ref instead of the model name:
    viewer: t.variant(Viewer, {
      // return null for viewer if the parent User is not the current user
      isNull: (user, args, ctx) => user.id !== ctx.user.id,
    }),
    email: t.exposeString('email'),
  }),
});

You can also use variants when defining relations by providing a type option:

const PostDraft = builder.prismaNode('Post', {
  variant: 'PostDraft'
  // This set's what database field to use for the nodes id field
  id: { field: 'id' },
  // fields work just like they do for builder.prismaObject
  fields: (t) => ({
    title: t.exposeString('title'),
    author: t.relation('author'),
  }),
});

const Viewer = builder.prismaObject('User', {
  variant: 'Viewer',
  fields: (t) => ({
    id: t.exposeID('id'),
    drafts: t.relation('posts', {
      // This will cause this relation to use the PostDraft variant rather than the default Post variant
      type: PostDraft,
      query: { where: { draft: true } },
    }),
  });
});

You may run into circular reference issues if you use 2 prisma object refs to reference each other. To avoid this, you can split out the field definition for one of the relationships using builder.prismaObjectField

const Viewer = builder.prismaObject('User', {
  variant: 'Viewer',
  fields: (t) => ({
    id: t.exposeID('id'),
    user: t.variant(User),
  });
});

const User = builder.prismaNode('User', {
  interfaces: [Named],
  id: {
    resolve: (user) => user.id,
  },
  fields: (t) => ({
    email: t.exposeString('email'),
  }),
});

// Viewer references the `User` ref in its field definiton,
// referencing the `User` in fields would cause a circular type issue
builder.prismaObjectField(Viewer, 'user', t.variant(User));

This same workaround applies when defining relations using variants.

Selecting fields from a nested GraphQL field

By default, the nestedSelection function will return selections based on the type of the current field. nestedSelection can also be used to get a selection from a field nested deeper inside other fields. This is useful if the field returns a type that is not a prismaObject, but a field nested inside the returned type is.

const PostRef = builder.prismaObject('Post', {
  fields: (t) => ({
    title: t.exposeString('title'),
    content: t.exposeString('content'),
    author: t.relation('author'),
  }),
});

const PostPreview = builder.objectRef<Post>('PostPreview').implement({
  fields: (t) => ({
    post: t.field({
      type: PostRef,
      resolve: (post) => post,
    }),
    preview: t.string({
      nullable: true,
      resolve: (post) => post.content?.slice(10),
    }),
  }),
});

builder.prismaObject('User', {
  fields: (t) => ({
    id: t.exposeID('id'),
    postPreviews: t.field({
      select: (args, ctx, nestedSelection) => ({
        posts: nestedSelection(
          {
            // limit the number of postPreviews to load
            take: 2,
          },
          // Look at the selections in postPreviews.post to determine what relations/fields to select
          ['post'],
        ),
      }),
      type: [PostPreview],
      resolve: (user) => user.posts,
    }),
  }),
});

Indirect relations (eg. Join tables)

If you want to define a GraphQL field that directly exposes data from a nested relationship (many to many relations using a custom join table is a common example of this) you can use the nestedSelection function passed to select.

Given a prisma schema like the following:

model Post {
  id        Int         @id @default(autoincrement())
  title     String
  content   String
  media     PostMedia[]
}

model Media {
  id           Int         @id @default(autoincrement())
  url          String
  posts        PostMedia[]
  uploadedBy   User        @relation(fields: [uploadedById], references: [id])
  uploadedById Int
}

model PostMedia {
  id      Int   @id @default(autoincrement())
  post    Post  @relation(fields: [postId], references: [id])
  media   Media @relation(fields: [mediaId], references: [id])
  postId  Int
  mediaId Int
}

You can define a media field that can pre-load the correct relations based on the graphql query:

const PostDraft = builder.prismaObject('Post', {
  fields: (t) => ({
    title: t.exposeString('title'),
    media: t.field({
      select: (args, ctx, nestedSelection) => ({
        media: {
          select: {
            // This will look at what fields are queried on Media
            // and automatically select uploadedBy if that relation is requested
            media: nestedSelection(
              // This arument is the default query for the media relation
              // It could be something like: `{ select: { id: true } }` instead
              true,
            ),
          },
        },
      }),
      type: [Media],
      resolve: (post) => post.media.map(({ media }) => media),
    }),
  }),
});

const Media = builder.prismaObject('Media', {
  select: {
    id: true,
  },
  fields: (t) => ({
    url: t.exposeString('url'),
    uploadedBy: t.relation('uploadedBy'),
  }),
});

Optimized queries without `t.prismaField`

In some cases, it may be useful to get an optimized query for fields where you can't use t.prismaField.

This may be required for combining with other plugins, or because your query does not directly return a PrismaObject. In these cases, you can use the queryFromInfo helper. An example of this might be a mutation that wraps the prisma object in a result type.

const Post = builder.prismaObject('Post', {...});

builder.objectRef<{
  success: boolean;
  post?: Post
  }>('CreatePostResult').implement({
  fields: (t) => ({
    success: t.boolean(),
    post: t.field({
      type: Post,
      nullable:
      resolve: (result) => result.post,
    }),
  }),
});

builder.mutationField(
  'createPost',
  {
    args: (t) => ({
      title: t.string({ required: true }),
      ...
    }),
  },
  {
    resolve: async (parent, args, context, info) => {
      if (!validateCreatePostArgs(args)) {
        return {
          success: false,
        }
      }

      const post = prisma.city.create({
        ...queryFromInfo({
          context,
          info,
          // nested path where the selections for this type can be found
          path: ['post']
          // optionally you can pass a custom initial selection, genereally you wouldn't need this
          // but if the field at `path` is not selected, the initial selection set may be empty
          select: {
            comments: true,
          },
        }),
        data: {
          title: args.input.title,
          ...
        },
      });

      return {
        success: true,
        post,
      }
    },
  },
);

Relay integration

This plugin has extensive integration with the relay plugin, which makes creating nodes and connections very easy.

`prismaNode`

The prismaNode method works just like the prismaObject method with a couple of small differences:

there is a new id option that mirrors the id option from node method of the relay plugin, and must contain a resolve function that returns the id from an instance of the node. Rather than defining a resolver for the id field, you can set the field option to the name of a unique column or index.

builder.prismaNode('Post', {
  // This set's what database field to use for the nodes id field
  id: { field: 'id' },
  // fields work just like they do for builder.prismaObject
  fields: (t) => ({
    title: t.exposeString('title'),
    author: t.relation('author'),
  }),
});

If you need to customize how ids are formatted, you can add a resolver for the id, and provide a findUnique option that can be used to load the node by it's id. This is generally not necissary.

builder.prismaNode('Post', {
  id: { resolve: (post) => String(post.id) },
  // The return value will be passed as the `where` of a `prisma.post.findUnique`
  findUnique: (id) => ({ id: Number.parseInt(id, 10) }),
  fields: (t) => ({
    title: t.exposeString('title'),
    author: t.relation('author'),
  }),
});

`prismaConnection`

The prismaConnection method on a field builder can be used to create a relay connection field that also pre-loads all the data nested inside that connection.

builder.queryType({
  fields: (t) => ({
    posts: t.prismaConnection(
      {
        type: 'Post',
        cursor: 'id',
        resolve: (query, parent, args, context, info) => prisma.post.findMany({ ...query }),
      }),
      {}, // optional options for the Connection type
      {}, // optional options for the Edge type),
    ),
  }),
});

options

type: the name of the prisma model being connected to
cursor: a @unique column of the model being connected to. This is used as the cursor option passed to prisma.
defaultSize: (default: 20) The default page size to use if first and last are not provided.
maxSize: (default: 100) The maximum number of nodes returned for a connection.
resolve: Like the resolver for prismaField, the first argument is a query object that should be spread into your prisma query. The resolve function should return an array of nodes for the connection. The query will contain the correct take, skip, and cursor options based on the connection arguments (before, after, first, last), along with include options for nested selections.
totalCount: A function for loading the total count for the connection. This will add a totalCount field to the connection object. The totalCount method will receive (connection, args, context, info) as arguments

The created connection queries currently support the following combinations of connection arguments:

first, last, or before
first and before
last and after

Queries for other combinations are not as useful, and generally requiring loading all records between 2 cursors, or between a cursor and the end of the set. Generating query options for these cases is more complex and likely very inefficient, so they will currently throw an Error indicating the argument combinations are not supported.

`relatedConnection`

The relatedConnection method can be used to create a relay connection field based on a relation of the current model.

builder.prismaNode('User', {
  id: { field: 'id' },
  fields: (t) => ({
    // Connections can be very simple to define
    simplePosts: t.relatedConnection('posts', {
      cursor: 'id',
    }),
    // Or they can include custom arguments, and other options
    posts: t.relatedConnection(
      'posts',
      {
        cursor: 'id',
        args: {
          oldestFirst: t.arg.boolean(),
        },
        query: (args, context) => ({
          orderBy: {
            createdAt: args.oldestFirst ? 'asc' : 'desc',
          },
        }),
      },
      {}, // optional options for the Connection type
      {}, // optional options for the Edge type),
    ),
  }),
});

options

cursor: a @unique column of the model being connected to. This is used as the cursor option passed to prisma.
defaultSize: (default: 20) The default page size to use if first and last are not provided.
maxSize: (default: 100) The maximum number of nodes returned for a connection.
query: A method that accepts the args and context for the connection field, and returns filtering and sorting logic that will be merged into the query for the relation.
totalCount: when set to true, this will add a totalCount field to the connection object. see relationCount above for more details.

Indirect relations as connections

Creating connections from indirect relations is a little more involved, but can be acheived using prismaConnectionHelpers with a normal t.connection field.

// Create a prisma object for the node type of your connection
const Media = builder.prismaObject('Media', {
  select: {
    id: true,
  },
  fields: (t) => ({
    url: t.exposeString('url'),
  }),
});

// Create connection helpers for the media type.  This will allow you
// to use the normal t.connection with a prisma type
const mediaConnectionHelpers = prismaConnectionHelpers(
  builder,
  'PostMedia', // this should be the the join table
  {
    cursor: 'id',
    select: (nodeSelection) => ({
      // select the relation to the media node using the nodeSelection function
      media: nodeSelection({
        // optionally specify fields to select by default for the node
        select: {
          id: true,
          posts: true,
        },
      }),
    }),
    // resolve the node from the edge
    resolveNode: (postMedia) => postMedia.media,
    // additional/optional options
    maxSize: 100,
    defaultSize: 20,
  },
);

builder.prismaObjectField('Post', 'mediaConnection', (t) =>
  t.connection({
    // The type for the Node
    type: Media,
    // since we are not using t.relatedConnection we need to manually
    // include the selections for our connection
    select: (args, ctx, nestedSelection) => ({
      media: mediaConnectionHelpers.getQuery(args, ctx, nestedSelection),
    }),
    resolve: (post, args, ctx) =>
      // This helper takes a list of nodes and formats them for the connection
      mediaConnectionHelpers.resolve(
        // map results to the list of edges
        post.media,
        args,
        ctx,
      ),
  }),
);

The above example assumes that you are paginating a relation to a join table, where the pagination args are applied based on the relation to that join table, but the nodes themselves are nested deeper.

prismaConnectionHelpers can also be used to manually create a connection where the edge and connections share the same model, and pagination happens directly on a relation to nodes type (even if that relation is nested).

const commentConnectionHelpers = prismaConnectionHelpers(builder, 'Comment', {
  cursor: 'id',
});

const SelectPost = builder.prismaObject('Post', {
  fields: (t) => ({
    title: t.exposeString('title'),
    comments: t.connection({
      type: commentConnectionHelpers.ref,
      select: (args, ctx, nestedSelection) => ({
        comments: commentConnectionHelpers.getQuery(args, ctx, nestedSelection),
      }),
      resolve: (parent, args, ctx) => commentConnectionHelpers.resolve(parent.comments, args, ctx),
    }),
  }),
});

You can create reusable connection objects by using builder.connectionObject.

These connection objects can be used with t.prismaConnection, t.relatedConnection, or t.connection

Shared edges can also be created using t.edgeObject

const CommentConnection = builder.connectionObject({
  type: Comment,
  // or
  type: commentConnectionHelpers.ref,
  name: 'CommentConnection',
});

builder.prismaObject('Post', {
  fields: (t) => ({
    id: t.exposeID('id'),
    ...
    commentsConnection: t.relatedConnection(
      'comments',
      { cursor: 'id' },
      // The connection object ref can be passed in place of the connection objectoptions
      CommentConnection
    ),
  }),
});

Extending connection edges

In some cases you may want to expose some data from an indirect connection on the edge object.

const mediaConnectionHelpers = prismaConnectionHelpers(builder, 'PostMedia', {
  cursor: 'id',
  select: (nodeSelection) => ({
    // select the relation to the media node using the nodeSelection function
    media: nodeSelection({}),
    // Select additional fields from the join table
    createdAt: true,
  }),
  // resolve the node from the edge
  resolveNode: (postMedia) => postMedia.media,
});

builder.prismaObjectFields('Post', (t) => ({
  manualMediaConnection: t.connection(
    {
      type: Media,
      select: (args, ctx, nestedSelection) => ({
        media: mediaConnectionHelpers.getQuery(args, ctx, nestedSelection),
        select: {
          media: nestedSelection({}, ['edges', 'node']),
        },
      }),

      resolve: (post, args, ctx) =>
        mediaConnectionHelpers.resolve(
          post.media.map(({ media }) => media),
          args,
          ctx,
        ),
    },
    {},
    // options for the edge object
    {
      // define the additional fields on the edge object
      fields: (edge) => ({
        createdAt: edge.field({
          type: 'DateTime',
          // the parent shape for edge fields is inferred from the connections resolve function
          resolve: (media) => media.createdAt,
        }),
      }),
    },
  ),
}));

Relay Utils

`parsePrismaCursor` and `formatPrismaCursor`

These functions can be used to manually parse and format cursors that are compatible with prisma connections.

Parsing a cursor will return the value from the column used for the cursor (often the id), this value may be an array or object when a compound index is used as the cursor. Similarly, to format a cursor, you must provide the column(s) that make up the cursor.

Using Prisma without a plugin

Using prisma without a plugin is relatively straight forward using the builder.objectRef method.

The easiest way to create types backed by prisma looks something like:

import { Post, PrismaClient, User } from '@prisma/client';

const db = new PrismaClient();
const UserObject = builder.objectRef<User>('User');
const PostObject = builder.objectRef<Post>('Post');

UserObject.implement({
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    posts: t.field({
      type: [PostObject],
      resolve: (user) =>
        db.post.findMany({
          where: { authorId: user.id },
        }),
    }),
  }),
});

PostObject.implement({
  fields: (t) => ({
    id: t.exposeID('id'),
    title: t.exposeString('title'),
    author: t.field({
      type: UserObject,
      resolve: (post) => db.user.findUniqueOrThrow({ where: { id: post.authorId } }),
    }),
  }),
});

builder.queryType({
  fields: (t) => ({
    me: t.field({
      type: UserObject,
      resolve: (root, args, ctx) => db.user.findUniqueOrThrow({ where: { id: ctx.userId } }),
    }),
  }),
});

This sets up User, and Post objects with a few fields, and a me query that returns the current user. There are a few things to note in this setup:

We split up the builder.objectRef and the implement calls, rather than calling builder.objectRef(...).implement(...). This prevents typescript from getting tripped up by the circular references between posts and users.
We use findUniqueOrThrow because those fields are not nullable. Using findUnique, prisma will return a null if the object is not found. An alternative is to mark these fields as nullable.
The refs to our object types are called UserObject and PostObject, this is because User and Post are the names of the types imported from prisma. We could instead alias the types when we import them so we can name the refs to our GraphQL types after the models.

This setup is fairly simple, but it is easy to see the n+1 issues we might run into. Prisma helps with this by batching queries together, but there are also things we can do in our implementation to improve things.

One thing we could do if we know we will usually be loading the author any time we load a post is to make the author part of shape required for a post:

const UserObject = builder.objectRef<User>('User');
// We add the author here in the objectRef
const PostObject = builder.objectRef<Post & { author: User }>('Post');

UserObject.implement({
  fields: (t) => ({
    id: t.exposeID('id'),
    email: t.exposeString('email'),
    posts: t.field({
      type: [PostObject],
      resolve: (user) =>
        db.post.findMany({
          // We now need to include the author when we query for posts
          include: {
            author: true,
          },
          where: { authorId: user.id },
        }),
    }),
  }),
});

PostObject.implement({
  fields: (t) => ({
    id: t.exposeID('id'),
    title: t.exposeString('title'),
    author: t.field({
      type: UserObject,
      // Now we can just return the author from the post instead of querying for it
      resolve: (post) => post.author,
    }),
  }),
});

We may not always want to query for the author though, so we could make the author optional and fall back to using a query if it was not provided by the parent resolver:

const PostObject = builder.objectRef<Post & { author?: User }>('Post');

PostObject.implement({
  fields: (t) => ({
    id: t.exposeID('id'),
    title: t.exposeString('title'),
    author: t.field({
      type: UserObject,
      resolve: (post) =>
        post.author ?? db.user.findUnique({ rejectOnNotFound: true, where: { id: post.authorId } }),
    }),
  }),
});

With this setup, a parent resolver has the option to include the author, but we have a fallback incase it does not.

There are other patterns like data loaders than can be used to reduce n+1 issues, and make your graph more efficient, but they are too complex to describe here.

Mocks plugin Relay plugin

Prisma Plugin for Pothos

Features

Example

Install

Setup

Add a the pothos generator to your prisma schema

Set up the builder

Creating types with builder.prismaObject

Adding prisma fields to non-prisma objects (including Query and Mutation)

Adding relations

Fallback queries

Filters, Sorting, and arguments

relationCount

Includes on types

Select mode for types

Using arguments or context in your selections

Extending prisma objects

Type variants

Selecting fields from a nested GraphQL field

Indirect relations (eg. Join tables)

Optimized queries without t.prismaField

Relay integration

prismaNode

prismaConnection

options

relatedConnection

options

Indirect relations as connections

Sharing Connections objects

Extending connection edges

Relay Utils

parsePrismaCursor and formatPrismaCursor

Using Prisma without a plugin

Add a the `pothos` generator to your prisma schema

Creating types with `builder.prismaObject`

Optimized queries without `t.prismaField`

`prismaNode`

`prismaConnection`

`relatedConnection`

`parsePrismaCursor` and `formatPrismaCursor`