GraphQL Field Collection

In this section, we continue our GraphQL example by explaining how Ent implements GraphQL Field Collection for our GraphQL schema and solves the “N+1 Problem” in our resolvers.

Clone the code (optional)

The code for this tutorial is available under github.com/a8m/ent-graphql-example, and tagged (using Git) in each step. If you want to skip the basic setup and start with the initial version of the GraphQL server, you can clone the repository as follows:

  1. git clone git@github.com:a8m/ent-graphql-example.git
  2. cd ent-graphql-example
  3. go run ./cmd/todo/

Problem

The “N+1 problem” in GraphQL means that a server executes unnecessary database queries to get node associations (i.e. edges) when it can be avoided. The number of queries that will be potentially executed (N+1) is a factor of the number of the nodes returned by the root query, their associations, and so on recursively. Meaning, this can potentially be a very big number (much bigger than N+1).

Let’s try to explain this with the following query:

  1. query {
  2. users(first: 50) {
  3. edges {
  4. node {
  5. photos {
  6. link
  7. }
  8. posts {
  9. content
  10. comments {
  11. content
  12. }
  13. }
  14. }
  15. }
  16. }
  17. }

In the query above, we want to fetch the first 50 users with their photos and their posts, including their comments.

In the naive solution (the problematic case), a server will fetch the first 50 users in one query, then, for each user will execute a query for getting their photos (50 queries), and another query for getting their posts (50). Let’s say each user has exactly 10 posts. Therefore, for each post (of each user), the server will execute another query for getting its comments (500). That means we will have 1+50+50+500=601 queries in total.

gql-request-tree

Ent Solution

The Ent extension for field collection adds support for automatic GraphQL field collection for associations (i.e. edges) using eager loading. Meaning, if a query asks for nodes and their edges, entgql will automatically add With steps to the root query, and as a result, the client will execute a constant number of queries to the database - and it works recursively.

In the GraphQL query above, the client will execute 1 query for getting the users, 1 for getting the photos, and another 2 for getting the posts and their comments (4 in total!). This logic works both for root queries/resolvers and for the node(s) API.

Example

For the purpose of the example, we disable the automatic field collection, change the ent.Client to run in debug mode in the Todos resolver, and restart our GraphQL server:

ent.resolvers.go

  1. func (r *queryResolver) Todos(ctx context.Context, after *ent.Cursor, first *int, before *ent.Cursor, last *int, orderBy *ent.TodoOrder) (*ent.TodoConnection, error) {
  2. - return r.client.Todo.Query().
  3. + return r.client.Debug().Todo.Query().
  4. Paginate(ctx, after, first, before, last,
  5. ent.WithTodoOrder(orderBy),
  6. )
  7. }

We execute the GraphQL query from the pagination tutorial, and add the parent edge to the result:

  1. query {
  2. todos(last: 10, orderBy: {direction: DESC, field: TEXT}) {
  3. edges {
  4. node {
  5. id
  6. text
  7. parent {
  8. id
  9. }
  10. }
  11. cursor
  12. }
  13. }
  14. }

Check the process output, and you will see that the server executed 11 queries to the database. 1 for getting the last 10 todo items, and another 10 queries for getting the parent of each item:

  1. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` ORDER BY `id` ASC LIMIT 11
  2. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  3. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  4. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  5. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  6. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  7. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  8. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  9. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  10. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2
  11. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` JOIN (SELECT `todo_parent` FROM `todos` WHERE `id` = ?) AS `t1` ON `todos`.`id` = `t1`.`todo_parent` LIMIT 2

Let’s see how Ent can automatically solve our problem: when defining an Ent edge, entgql auto binds it to its usage in GraphQL and generates edge-resolvers for the nodes under the gql_edge.go file:

ent/gql_edge.go

  1. func (t *Todo) Children(ctx context.Context) ([]*Todo, error) {
  2. if fc := graphql.GetFieldContext(ctx); fc != nil && fc.Field.Alias != "" {
  3. result, err = t.NamedChildren(graphql.GetFieldContext(ctx).Field.Alias)
  4. } else {
  5. result, err = t.Edges.ChildrenOrErr()
  6. }
  7. if IsNotLoaded(err) {
  8. result, err = t.QueryChildren().All(ctx)
  9. }
  10. return result, err
  11. }

If we check the process’ output again without disabling fields collection, we will see that this time the server executed only two queries to the database. One to get the last 10 todo items, and a second for getting the parent-item of each todo-item that was returned to the first query.

  1. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority`, `todos`.`todo_parent` FROM `todos` ORDER BY `id` DESC LIMIT 11
  2. SELECT DISTINCT `todos`.`id`, `todos`.`text`, `todos`.`created_at`, `todos`.`status`, `todos`.`priority` FROM `todos` WHERE `todos`.`id` IN (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)

If you’re having trouble running this example, go to the first section, clone the code and run the example.


Well done! By using automatic field collection for our Ent schema definition, we were able to greatly improve the GraphQL query efficiency in our application. In the next section, we will learn how to make our GraphQL mutations transactional.