Skip to content

qudini/gom

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

GOM

GOM (GraphQL-Object Mapping) batches your resolvers in order to reduce your data queries, while still allowing them to be customised depending on the arguments and selection.

See qudini-reactive for an example of how it can be used with Spring Boot and Spring WebFlux.

Overview

Originally, BatchLoaders are "DataFetchingEnvironment-agnostic" by design, meaning they take only keys to allow fetching the corresponding values from your data source, for example:

BatchLoader<Integer, Article> articleBatchLoader = new BatchLoader<Integer, Article>() {
    
    @Override
    public CompletionStage<List<Article>> load(List<Integer> keys) {
        return articleService.getArticles(keys);
    }
    
};

GOM "enhances" those keys by passing instances of DataLoaderKey instead (internal class):

class DataLoaderKey {
    
    // the source (DataFetchingEnvironment#getSource)
    private Object source;
    
    // the arguments (DataFetchingEnvironment#getArguments)
    private Object arguments;
    
    // the selection (DataFetchingEnvironment#getSelectionSet)
    private Object selection;
    
    // the context (DataFetchingEnvironment#getGraphQlContext)
    private Object context;
    
}

This allows a BatchLoader to "group the keys by arguments and selection", and call your resolvers as many times as there are distinct arguments and selection, but each time will all the sources.

GOM will prepare DataFetchers and DataLoaders for you, so that you just need to decorate your RuntimeWiring and DataLoaderRegistry instances.

Installation

Maven Central

<dependency>
    <groupId>com.qudini</groupId>
    <artifactId>qudini-gom</artifactId>
    <version>${qudini-gom.version}</version>
</dependency>

Usage

Resolvers

To create a resolver, just annotate a class with @com.qudini.gom.TypeResolver while passing the GraphQL type this resolver handles:

@TypeResolver("Article")
public class ArticleResolver {
    
}

Now, say your Article GraphQL type has a blog: Blog! field, just annotate one of ArticleResolver's methods with @com.qudini.gom.FieldResolver while passing the field name this method handles:

@TypeResolver("Article")
public class ArticleResolver {
    
    @FieldResolver("blog")
    public ... getArticleBlog(...) {
        ...
    }
    
}

A resolver's method takes one, two or three parameters, in this specific order:

  1. the source, mandatory,
  2. the arguments, optional (see the Arguments section).
  3. the selection, optional (see the Selection section).

Special case for types that have no source (e.g. Query): their resolver methods will only accept arguments and/or selection.

A resolver method can return anything (more details in the Converters section).

The following resolvers are all valid ones:

@TypeResolver("Query")
public class QueryResolver {
    
    @FieldResolver("articles")
    public List<Article> listArticles() {
        return articleService.find();
    }
    
}

@TypeResolver("Query")
public class QueryResolver {
    
    @FieldResolver("articles")
    public List<Article> listArticles(Arguments arguments) {
        return articleService.find(arguments);
    }
    
}

@TypeResolver("Query")
public class QueryResolver {
    
    @FieldResolver("articles")
    public List<Article> listArticles(Selection selection) {
        return articleService.find(selection);
    }
    
}

@TypeResolver("Query")
public class QueryResolver {
    
    @FieldResolver("articles")
    public List<Article> listArticles(Arguments arguments, Selection selection) {
        return articleService.find(arguments, selection);
    }
    
}

@TypeResolver("Article")
public class ArticleResolver {
    
    @FieldResolver("blog")
    public Blog getArticleBlog(Article article) {
        return blogService.findByArticle(article);
    }
    
}

@TypeResolver("Article")
public class ArticleResolver {
    
    @FieldResolver("blog")
    public Blog getArticleBlog(Article article, Arguments arguments) {
        return blogService.findByArticle(article, arguments);
    }
    
}

@TypeResolver("Article")
public class ArticleResolver {
    
    @FieldResolver("blog")
    public Blog getArticleBlog(Article article, Selection selection) {
        return blogService.findByArticle(article, selection);
    }
    
}

@TypeResolver("Article")
public class ArticleResolver {
    
    @FieldResolver("blog")
    public Blog getArticleBlog(Article article, Arguments arguments, Selection selection) {
        return blogService.findByArticle(article, arguments, selection);
    }
    
}

The above resolvers will have one DataFetcher implemented for each declared method.

@Batched

To have a DataLoader implemented instead of only a DataFetcher, use the @com.qudini.gom.Batched annotation in addition to the @FieldResolver one:

@TypeResolver("Article")
public class ArticleResolver {
    
    @Batched
    @FieldResolver("blog")
    public ... getArticleBlog(...) {
        ...
    }
    
}

What changes:

  • the method now takes a Set<Source> instead of Source directly,
  • the method now needs to return Map<Source, Result> instead of Result directly (MappedBatchLoader is used, see java-dataloader).

What doesn't change:

  • the arguments and selection are still available as parameters.

For example:

@TypeResolver("Article")
public class ArticleResolver {
    
    @Batched
    @FieldResolver("blog")
    public Map<Article, Blog> getArticleBlog(Set<Article> articles, Arguments arguments, Selection selection) {
        return blogService.findByArticles(articles, arguments, selection);
    }
    
    @Batched
    @FieldResolver("comments")
    public Map<Article, List<Comment>> getArticleComments(Set<Article> articles) {
        return commentService.findByArticles(articles);
    }
    
}

Important note: as @Batched resolvers take a Set<Source>, the Source class has to implement equals/hashCode carefully (i.e. not leave it to the default Object's, as it is per-instance implemented).

Arguments

When requesting the arguments as a parameter of your resolvers, you will receive an instance of graphql.gom.Arguments. This is basically an abstraction of the value returned by DataFetchingEnvironment#getArguments (Map<String, Object>). It provides three main methods:

  • <T> T get(String name): to be used when an argument is mandatory according to the GraphQL schema (will fail if the returned value is null or absent).
  • <T> Optional<T> getOptional(String name): to be used when an argument is optional according to the GraphQL schema (an empty Optional will be returned if the value is null or absent).
  • <T> Optional<Optional<T>> getNullable(String name): to be used when you need to make a difference between "no argument given" and "argument given but explicitly set to null". It will return an empty Optional if the argument hasn't been found at all, an Optional of an empty Optional if the argument got found but was null, and an Optional of an Optional of the value otherwise. This is mostly useful when one of your mutation can "optionally update an optional data field":
    • Empty Optional? Do nothing.
    • Optional of an empty Optional? Update the data field by emptying its value (e.g. setting null).
    • Optional of a valued Optional? Update the data field with the wrapped value.

You will also find:

  • <T extends Enum<T>> T getEnum(String name, Class<T> clazz) and <T extends Enum<T>> List<T> getEnumArray(String name, Class<T> clazz) (plus the Optional and Nullable variants) to deserialise directly into an Enum.
  • Arguments getInput(String name) and List<Arguments> getInputArray(String name) (plus the Optional and Nullable variants) when dealing with GraphQL inputs.

Selection

When requesting the selection as a parameter of your resolvers, you will receive an instance of graphql.gom.Selection, which exposes the following methods:

  • boolean contains(String field): returns true if the given field is part of the selection.
  • Stream<String> stream(): streams the selected fields.
  • Selection subSelection(String prefix): returns a new selection with the fields starting with the given prefix (those matching fields will have the given prefix removed, see the @Depth section).

For example, given the following query:

article {
    id
    title
}

The selection injected in the resolver of article will behave the following way:

  • selection.contains("id"): true
  • selection.stream().anyMatch("title"::equals): true
  • selection.contains("foobar"): false
@Depth

By default, the selection will have a depth of 1, meaning the nested selections won't be taken into account. For example, given the following query:

article {
    id
    title
    comments {
        content
    }
}
@FieldResolver("article")
public Article getArticle(Selection selection) {...}

Those above selection will only contain id, title and comments. A resolver on comments though, would receive the content field in its selection.

To access comments/content in this parent selection directly, annotate the parameter with @com.qudini.gom.Depth:

@FieldResolver("article")
public Article getArticle(@Depth(2) Selection selection) {...}

Those above selection will then contain id, title, comments, as well as comments/content. You can then use selection.subSelection("comments/") to receive a selection with the children of comments/ only (i.e. content in this case).

This can end up particularly useful when doing cursor-based pagination:

articles(first: 10) {
    edges {
        node {
            title
        }
    }
}
@FieldResolver("articles")
public Article getPaginatedArticles(@Depth(3) Selection selection) {
    Selection nodeSelection = selection.subSelection("edges/node/");
    nodeSelection.contains("title"); // true
    ...
}

Gom

Once you've implemented your resolvers, you then need to create is an instance of graphql.gom.Gom. You should create only one instance of it, on server startup (i.e. not per GraphQL query).

Gom gom = Gom
    .newGom()
    .resolvers(myResolverInstances)
    .converters(myConvertersInstance)
    .build();

.resolvers(myResolverInstances)

This is where you pass the instances of your resolvers so that they get registered, e.g.:

.resolvers(asList(
    new QueryResolver(),
    new BlogResolver(),
    new ArticleResolver(),
    new CommentResolver()
))

.converters(myConvertersInstance)

All your resolver methods will return CompletableFutures behind the scene (if they don't, the returned value will be wrapped into CompletableFuture#completedFuture automatically). GOM's converters are useful if the returned value is "future-capable" but just needs a "conversion".

For example, say you're using Project Reactor's reactive types Mono and Flux, you can provide a converter for both of these types so that the returned value becomes "graphql-java compliant". A converter takes the instance it needs to convert plus the GraphQL query context, and returns the converted value. In this case:

Converters converters = Converters
    .newConverters()
    .converter(Mono.class, (mono, context) -> mono.toFuture())
    .converter(Flux.class, (flux, context) -> flux.collectList().toFuture())
    .build(); 

You will then have to pass these converters to Gom#converters so that they get successfully registered.

The context parameter (of type graphql.GraphQLContext) looks a bit superfluous, but it can actually be pretty powerful, especially in cases like the above one: when building your ExecutionInput, if you make the GraphQL context hold an instance of Project Reactor's current reactor.util.context.Context thanks to ExecutionInput.Builder#graphQLContext(...), you can pass it through to your resolvers transparently, which will then allow using Spring Security annotations on your resolvers for example:

.converter(Mono.class, (mono, context) -> mono.contextWrite(context.get(REACTOR_CONTEXT_KEY)).toFuture())
.converter(Flux.class, (flux, context) -> flux.collectList())

Gom#decorateRuntimeWiringBuilder and Gom#decorateDataLoaderRegistry

Once you have your Gom instance created (either stored as a singleton or as a bean in a dependency-injection-aware architecture), when you create your graphql-java's RuntimeWiring (on server startup), just call:

gom.decorateRuntimeWiringBuilder(runtimeWiringBuilder);

When you create your java-dataloader's DataLoaderRegistry (usually at query runtime, depending on how you want the cache to behave), just call:

gom.decorateDataLoaderRegistry(dataLoaderRegistry);

You're now good to go!

Other utilities

According to the GraphQL Cursor Connections Specification, predefined types have been made available:

  • com.qudini.gom.paging.Connection
  • com.qudini.gom.paging.Edge
  • com.qudini.gom.paging.PageArguments
  • com.qudini.gom.paging.PageInfo

Example

This example has been implemented as a proof, see src/test/java/graphql/gom/example.

Database model

Say you have the following database model:

public class Blog {
    private int id;
    private String name;
    private List<Article> articles;
    // accessors
    // equals/hashCode on id
}

public class Article {
    private int id;
    private String title;
    private Blog blog;
    private List<Comment> comments;
    // accessors
    // equals/hashCode on id
}

public class Comment {
    private int id;
    private String content;
    private Article article;
    // accessors
    // equals/hashCode on id
}

As you can see:

  • one Blog has many Articles,
  • one Article has one Blog and many Comments,
  • one Comment has one Article.

GraphQL schema

Here could be your GraphQL schema:

type Query {
    # Lists all the blogs:
    blogs: [Blog!]!
}

type Blog {
    # Gets a blog's id:
    id: Int!
    # Gets a blog's name:
    name: String!
    # Lists all articles of a blog:
    articles: [Article!]!
}

type Article {
    # Gets an article's id:
    id: Int!
    # Gets an article's title:
    title: String!
    # Gets an article's blog:
    blog: Blog!
    # Lists all comments of an article, possibly filtered by content:
    comments(containing: String): [Comment!]!
}

type Comment {
    # Gets a comment's id:
    id: Int!
    # Gets a comment's content:
    content: String!
    # Gets a comment's article:
    article: Article!
}

GraphQL query

A consumer calls your GraphQL endpoint with the following query (intentionally overfetching):

query {
    blogs {
        name
        articles {
            title
            blog {
                name
            }
            allComments: comments {
                content
                article {
                    title
                }
            }
            commentsContainingFoobar: comments(containing: "foobar") {
                 content
                 article {
                     title
                 }
             }
        }
    }
}

For the following examples, let's imagine your DB has two blogs, two articles per blog, and two comments per article. For each article, one of the comment contains "foobar" while the other doesn't.

Resolving using DataFetchers

A naive implementation could be to implement the resolution of this GraphQL schema via DataFetchers. GOM allows you to do it in a Java-friendly way:

@TypeResolver("Query")
public class QueryResolver {
    
    private BlogService blogService;
    
    @FieldResolver("blogs")
    public List<Blog> getBlogs() {
        return blogService.findAll();
    }
    
}

@TypeResolver("Blog")
public class BlogResolver {
    
    private ArticleService articleService;
    
    @FieldResolver("articles")
    public List<Article> getArticles(Blog blog) {
        return articleService.findManyByBlog(blog);
    }
    
}

@TypeResolver("Article")
public class ArticleResolver {
    
    private BlogService blogService;
    
    private CommentService commentService;
    
    @FieldResolver("blog")
    public Blog getBlog(Article article) {
        return blogService.findOneByArticle(article);
    }
    
    @FieldResolver("comments")
    public List<Comment> getComments(Article article, Arguments arguments) {
        return commentService.findManyByArticle(article, arguments.getOptional("containing"));
    }
    
}

@TypeResolver("Comment")
public class CommentResolver {
    
    private ArticleService articleService;
    
    @FieldResolver("article")
    public Article getArticle(Comment comment) {
        return articleService.findOneByComment(comment);
    }
    
}

Let's see how many times the above methods were run:

  • QueryResolver#getBlogs: 1 time (entry point, returning 2 blogs),
  • BlogResolver#getArticles: 2 times (2 blogs, returning 4 articles),
  • ArticleResolver#getBlog: 4 times (4 articles, returning 4 blogs),
  • ArticleResolver#getComments: 8 times (4 articles but two calls per article - one filtering by "foobar" and another one without filtering - returning respectively 4 and 8 comments, so 12 comments in total),
  • CommentResolver#getArticle: 12 times (12 comments, returning 12 articles).

If resolvers are actually doing IOs to a database for example, this can quite quickly go crazy, especially if you have more data than in this small example.

Resolving using DataLoaders

Let's improve this query resolution by using DataLoaders instead:

@TypeResolver("Query")
public class QueryResolver {
    
    private BlogService blogService;
    
    @FieldResolver("blogs")
    public List<Blog> getBlogs() {
        return blogService.findAll();
    }
    
}

@TypeResolver("Blog")
public class BlogResolver {
    
    private ArticleService articleService;
    
    @Batched
    @FieldResolver("articles")
    public Map<Blog, List<Article>> getArticles(Set<Blog> blogs) {
        return articleService.findManyByBlogs(blogs);
    }
    
}

@TypeResolver("Article")
public class ArticleResolver {
    
    private BlogService blogService;
    
    private CommentService commentService;
    
    @Batched
    @FieldResolver("blog")
    public Map<Article, Blog> getBlog(Set<Article> articles) {
        return blogService.findManyByArticles(articles);
    }
    
    @Batched
    @FieldResolver("comments")
    public Map<Article, List<Comment>> getComments(Set<Article> articles, Arguments arguments) {
        return commentService.findManyByArticles(articles, arguments.getOptional("containing"));
    }
    
}

@TypeResolver("Comment")
public class CommentResolver {
    
    private ArticleService articleService;
    
    @Batched
    @FieldResolver("article")
    public Map<Comment, Article> getArticle(Set<Comment> comments) {
        return articleService.findManyByComments(comments);
    }
    
}

Let's compare how many times the above methods were now run:

  • QueryResolver#getBlogs: 1 time (entry point, returning 2 blogs),
  • BlogResolver#getArticles: 1 time (takes 2 blogs, returns 4 articles),
  • ArticleResolver#getGetBlog: 1 time (takes 4 articles, returns 4 blogs),
  • ArticleResolver#getComments: 2 times (takes 4 articles but still two calls - one filtering by "foobar" and another one without filtering - returning respectively 4 and 8 comments, so 12 comments in total),
  • CommentResolver#getArticle: 1 time (takes 12 comments, returns 12 articles).

See? Ready for the load!