Heads up: Composite Types

We will soon release a new feature, that is relevant for ERD generation: Composite Types. They are pretty much like Models, but represent Embedded Documents for MongoDB for example:

This is what the underlying data looks like: https://docs.atlas.mongodb.com/sample-data/sample-airbnb/

And this the schema:

generator client {
  provider        = "prisma-client-js"
  previewFeatures = ["mongoDb"]
}

datasource db {
  provider = "mongodb"
  url      = "mongodb+srv://us...godb.net/sample_airbnb"
}

type ListingsAndReviewsAddress {
  country         String
  country_code    String
  government_area String
  location        ListingsAndReviewsAddressLocation
  market          String
  street          String
  suburb          String
}

type ListingsAndReviewsAddressLocation {
  coordinates       Float[]
  is_location_exact Boolean
  type              String
}

type ListingsAndReviewsAvailability {
  availability_30  Int
  availability_365 Int
  availability_60  Int
  availability_90  Int
}

type ListingsAndReviewsHost {
  host_about                String
  host_has_profile_pic      Boolean
  host_id                   String
  host_identity_verified    Boolean
  host_is_superhost         Boolean
  host_listings_count       Int
  host_location             String
  host_name                 String
  host_neighbourhood        String
  host_picture_url          String
  host_response_rate        Int?
  host_response_time        String?
  host_thumbnail_url        String
  host_total_listings_count Int
  host_url                  String
  host_verifications        String[]
}

type ListingsAndReviewsImages {
  medium_url     String
  picture_url    String
  thumbnail_url  String
  xl_picture_url String
}

type ListingsAndReviewsReviewScores {
  review_scores_accuracy      Int?
  review_scores_checkin       Int?
  review_scores_cleanliness   Int?
  review_scores_communication Int?
  review_scores_location      Int?
  review_scores_rating        Int?
  review_scores_value         Int?
}

type ListingsAndReviewsReviews {
  id            String   @map("_id")
  comments      String?
  date          DateTime @db.Date
  listing_id    String
  reviewer_id   String
  reviewer_name String
}

model listingsAndReviews {
  id                    String                         @id @map("_id")
  access                String
  accommodates          Int
  address               ListingsAndReviewsAddress
  amenities             String[]
  availability          ListingsAndReviewsAvailability
  bathrooms             Decimal?
  bed_type              String
  bedrooms              Int?
  beds                  Int?
  calendar_last_scraped DateTime                       @db.Date
  cancellation_policy   String
  cleaning_fee          Decimal?
  description           String
  extra_people          Decimal
  first_review          DateTime?                      @db.Date
  guests_included       Decimal
  host                  ListingsAndReviewsHost
  house_rules           String
  images                ListingsAndReviewsImages
  interaction           String
  last_review           DateTime?                      @db.Date
  last_scraped          DateTime                       @db.Date
  listing_url           String
  maximum_nights        String
  minimum_nights        String
  monthly_price         Decimal?
  name                  String
  neighborhood_overview String
  notes                 String
  number_of_reviews     Int
  price                 Decimal
  property_type         String
  review_scores         ListingsAndReviewsReviewScores
  reviews               ListingsAndReviewsReviews[]
  reviews_per_month     Int?
  room_type             String
  security_deposit      Decimal?
  space                 String
  summary               String
  transit               String
  weekly_price          Decimal?

  @@index([property_type, room_type, beds], map: "property_type_1_room_type_1_beds_1")
  @@index([name], map: "name_1")
}

So we have one model, that then uses quite a few type in place where there could also be a Json field.

It probably makes sense to display these composite types as normal “tables” via “relations” - 1-1 or 1-n (lists) are possible. They don’t have a primary key and can not include relations, and their usage is also not via @relation but more like a type.

Let me know if you have any questions.

[janpio]

I wanted to visualize some of our schemas with composite types, so I spent a few minutes and implemented a first quick and dirty version of this: https://github.com/keonik/prisma-erd-generator/pull/71

The schemas posted above, in the same order as they appear:

sample-airbnb:

sample_mflix:

sample_restaurants:

sample_supplies:

sample_training: (This one is buggy somehow)

sample_weatherdata:

They are all missing the normal relations between models because these can not be introspected in MongoDB and have to be added by hand - which I have not done for these yet. Still nice results I think.

[keonik]

It looks pretty similar to model at first glance, just doesn’t have the explicit @relation attached. Thanks for the heads up! If I get some time I’ll play around with what is here to see if I can get it rendered correctly