LEFT JOINS and aggregation in a single Prisma query

Question

I have a database with multiple tables that frequently need to be queried with LEFT JOIN so that results contain aggregated data from other tables. Snippet from my Prisma schema:

model posts {
  id                Int      @id @unique @default(autoincrement())
  user_id           Int
  movie_id          Int      @unique
  title             String   @db.Text
  description       String?  @db.Text
  tags              Json?
  created_at        DateTime @default(now()) @db.DateTime(0)
  image             String?  @default("https://picsum.photos/400/600/?blur=10") @db.VarChar(256)
  year              Int
  submitted_by      String   @db.Text
  tmdb_rating       Decimal? @default(0.0) @db.Decimal(3, 1)
  tmdb_rating_count Int?     @default(0)
}

model ratings {
  id         Int       @unique @default(autoincrement()) @db.UnsignedInt
  entry_id   Int       @db.UnsignedInt
  user_id    Int       @db.UnsignedInt
  rating     Int       @default(0) @db.UnsignedTinyInt
  created_at DateTime  @default(now()) @db.DateTime(0)
  updated_at DateTime? @db.DateTime(0)

  @@id([entry_id, user_id])
}

If I wanted to return the average rating when querying posts, I could use a query like:

SELECT 
    p.*, ROUND(AVG(rt.rating), 1) AS user_rating
FROM
    posts AS p
        LEFT JOIN
    ratings AS rt ON rt.entry_id = p.id
GROUP BY p.id;

I'm not exactly sure how/whether I can achieve something similar with Prisma, because as it stands right now, it seems like this would require two separate queries, which isn't optimal because there is sometimes the need for 2 or 3 joins or SELECTs from other tables.

How can I make a query/model/something in Prisma to achieve the above?

Partite

Answer 1

Yes, this is possible with Prisma!. For making this work, you need to specify on your "schema.prisma" file how are models related with each other. That way, code generation will set the possible queries/operations.

Change it to this:

model Post {
  id              Int      @id @unique @default(autoincrement()) @map("id")
  userId          Int      @map("user_id")
  movieId         Int      @unique @map("movie_id")
  title           String   @map("title") @db.Text
  description     String?  @map("description") @db.Text
  tags            Json?    @map("tags")
  createdAt       DateTime @default(now()) @map("created_at") @db.DateTime(0)
  image           String?  @default("https://picsum.photos/400/600/?blur=10") @map("image") @db.VarChar(256)
  year            Int      @map("year")
  submittedBy     String   @map("submitted_by") @db.Text
  tmdbRating      Decimal? @default(0.0) @map("tmdb_rating") @db.Decimal(3, 1)
  tmdbRatingCount Int?     @default(0) @map("tmdb_rating_count")
  ratings         Rating[]

  @@map("posts")
}

model Rating {
  id        Int       @unique @default(autoincrement()) @map("id") @db.UnsignedInt
  userId    Int       @map("user_id") @db.UnsignedInt
  rating    Int       @default(0) @map("rating") @db.UnsignedTinyInt
  entryId   Int
  entry     Post      @relation(fields: [entryId], references: [id])
  createdAt DateTime  @default(now()) @map("created_a") @db.DateTime(0)
  updatedAt DateTime? @map("updated_a") @db.DateTime(0)

  @@id([entryId, userId])
  @@map("ratings")
}

Note: Please follow the naming conventions (singular form, PascalCase). I made those changes for you at the schema above. @@map allows you to set the name you use on your db tables.

Then, after generating the client, you will get access to the relational operations.

    // All posts with ratings data
    const postsWithRatings = await prisma.post.findMany({
        include: {
            // Here you can keep including data from other models
            ratings: true
        },
        // you can also "select" specific properties
    });

    // Calculate on your API
    const ratedPosts = postsWithRatings.map( post => {
        const ratingsCount = post.ratings.length;
        const ratingsTotal = post.ratings.reduce((acc, b) => acc + b.rating, 0)
        return {
            ...post,
            userRating: ratingsTotal / ratingsCount
        }
    })

    // OR...


    // Get avg from db
    const averages = await prisma.rating.groupBy({
        by: ["entryId"],
        _avg: {
            rating: true
        },
        orderBy: {
            entryId: "desc"
        }
    })
    //  Get just posts
    const posts = await prisma.post.findMany({
        orderBy: {
            id: "desc"
        }
    });
    // then match the ratings with posts
    const mappedRatings = posts.map( (post, idx) => {
        return {
            ...post,
            userRating: averages[idx]._avg.rating
        }
    })

You could also create a class with a method for making this easier. But I strongly recommend you to implement GraphQL on your API. That way, you can add a virtual field inside your post type. Any time a post is requested alone or in a list, the average will be calculated. In that same way, you would have the flexibility to request data from other models and the "JOINS" will get handled for you automatically.

Last but not least, if you ever want to do a lot of queries at the same time, you can take advantage of the Prisma transactions.

Whiney

I think the question is hoping to literally solve the problem, as in generate 1 SQL statement that does both. Two queries "works" but isn't a solution. – Taffeta

Answer 2

Despite the accepted answer, the actual answer is: No.

For actual performant joins to work, they must solve an issue that's been open for about a year and a half as of writing this response: https://github.com/prisma/prisma/issues/5184

Currently, there is no way to join tables together. Queries that include relations only include the relational data by using separate queries.

Caloric

This is no longer accurate as of Feb '24: prisma.io/blog/… – Trueblue