Join relation with subquery

Issue type:

[x] question

Database system/driver:

[x] postgres

TypeORM version:

[x] 0.2.0

Steps to reproduce or a small repository showing the problem:

I have the following relation Movie -> Actors:

import {
    Entity,
    Column,
    ManyToMany,
    JoinTable,
} from "typeorm"
import { Actor } from "@app/entity"

@Entity("movies")
export class Movie {
    @Column({ type: "text" })
    title: string

    @ManyToMany(type => Actor, actor => actor.movies)
    @JoinTable({
        name: "movies_actors",
        joinColumn: { name: "movieId" },
        inverseJoinColumn: { name: "actorId" }
    })
    actors: Actor[]
}

with query.leftJoinAndSelect("movie.actors", "actor") I get following query:

SELECT
  "movie"."id" AS "movie_id",
  "movie"."title" AS "movie_title",
  "actors"."id" AS "actors_id",
  "actors"."name" AS "actors_name"
FROM "movies" "movie"
LEFT JOIN "movies_actors" "movie_actors"
  ON "movie_actors"."movieId" = "movie"."id"
LEFT JOIN "actors" "actors"
  ON "actors"."id" = "movie_actors"."actorId"

which works fine. If I want to apply limit/offset to actors, I have to generate query like this:

SELECT
  "movie"."id" AS "movie_id",
  "movie"."title" AS "movie_title",
  "actors"."id" AS "actors_id",
  "actors"."name" AS "actors_name"
FROM "movies" "movie"
LEFT JOIN (
  SELECT * FROM (
    SELECT *, ROW_NUMBER() OVER (PARTITION BY "movieId") AS "seqId"
    FROM "movies_actors"
  ) "movies_actors"
  WHERE "movies_actors"."seqId" BETWEEN 0 AND 5
) "movie_actors"
  ON "movie_actors"."movieId" = "movie"."id"
LEFT JOIN "actors" "actors"
  ON "actors"."id" = "movie_actors"."actorId"

I’m not sure how can I do this with query builder, if I use subquery in join, it seems I have to manually replicate what TypeORM does with respect to relationship type, join table column names, ON statements etc. in subquery. Is there any other way of doing it, where I have control over select statement inside join, but everything else stays the same?

[GaryChangCN]

Anything new ？

[onderonur]

yeah that shall work however for map many “user.photos” is just a property name, it does not care if its a relation or something else. And btw most of times its really better to execute a separate queries and map the data in terms of both performance and code, so I was thinking to deprecate leftJoinAndMapMany method, maybe it will be in the future.

So you recommend to execute separate queries and I’ve read in another issue that you usually load your entities first and execute queries for things like relation counts etc and map that data to properties of your entities. Does this mean multiple db hits to collect the desired data? Or are you talking about creating subqueries and using addSelect etc?

It might be a really silly question. But many people I know just create giant queries with a lot of joins and execute them once. I can’t say that is always performant but when there are more than one db hits, many people goes crazy. I mostly work as a front-end developer and don’t have too solid knowledge about this kind of stuff. Can you clarify this a little bit more? (Maybe with some examples)

Thanks!