/graphql-weaver

A tool to combine, link and transform GraphQL schemas

Primary LanguageTypeScriptMIT LicenseMIT

graphql-weaver

npm version Build Status

A tool to combine, link and transform GraphQL schemas

Use graphql-weaver if you have multiple GraphQL servers and want to combine them into one API. Features like namespacing, links and custom transformation modules allow you to augment the API as you like.

Try it online in Apollo Launchpad

How to use

npm install --save graphql-weaver

Basic usage:

const schema: GraphQLSchema = await weaveSchemas({
    endpoints: [{
        namespace: 'model',
        typePrefix: 'Model',
        url: 'http://localhost:8080/graphql' // url to a GraphQL endpoint
    }, {
        namespace: 'local',
        schema: new GraphQLSchema(/* ... */) // use schema instance directly
    }]
})

A woven schema is an executable GraphQL schema built from several endpoints. For each endpoint, you can either specify a URL to a GraphQL server, pass an executable GraphQL schema instance, or implement the GraphQLClient interface yourself.

In its basic configuration, weaveSchemas merges the query, mutation and subscription fields of all endpoints. To avoid name collisions, you can specify the namespace and typePrefix properties like seen above. The typePrefix will be prepended to all types; namespace causes the fields of this endpoint to be wrapped in a field, to be queried via { model { aFieldOfModel } }.

Links

In the spirit of GraphQL, this tool allows you to create links between objects of different endpoints. Suppose you have a music recommendation service and a music library service. You can make the whole properties of a song available in the recommendation API without the recommendation service knowing all song properties.

const schema: GraphQLSchema = await weaveSchemas({
    endpoints: [{
        namespace: 'library',
        url: 'http://example.com/library/graphql'
    }, {
        namespace: 'recommendations',
        url: 'http://example.com/recommendations/graphql',
        fieldMetadata: {
            'Recommendation.song': { // Field song in type Recommendation
                link: {
                    field: 'library.Song', // field Song in namespace library
                    argument: 'id', // argument of library.Song
                    batchMode: false,
                }
            }
        }
     }]
});

This assumes the library schema has a field Song with argument id, and the recommendations schema has a type Recommendation with a field song which contains the song id. Then, you can query the recommendations with all song information like this:

query {
    recommendations {
        myRecommendations {
            recommendedAt
            song {
                id
                artist
                title
                year
            }
        }
    }
}

If there are many recommendations, this is ineficcient because all songs are queried independently. If the library schema supports querying multiple songs at once, you can set batchMode to true. If the library schema may return the songs in a different order than the ids its get, you need to set keyField too.

const schema: GraphQLSchema = await weaveSchemas({
    endpoints: [{
        namespace: 'library',
        url: 'http://example.com/library/graphql'
    }, {
        namespace: 'recommendations',
        url: 'http://example.com/recommendations/graphql',
        fieldMetadata: {
            'Recommendation.song': {
                link: {
                    field: 'library.allSongs',
                    argument: 'filter.ids', // allSongs has an argument filter with an array field ids
                    batchMode: true,
                    keyField: 'id' // the name of a field in Song type that contains the id
                }
            }
        }
     }]
});

In the case where you want to perfom a link from a single id to a list of many related items, include linkFieldName as an alias for the linked list of items, and the oneToMany flag. Since we are making a single request for a GraphQL list of items, batchMode is not required and should be set to false.

const schema: GraphQLSchema = await weaveSchemas({
    endpoints: [{
        namespace: 'library',
        url: 'http://example.com/library/graphql'
    }, {
        namespace: 'recommendations',
        url: 'http://example.com/recommendations/graphql',
        fieldMetadata: {
            'Artist.name': { // Field name in type Artist
                link: {
                    field: 'library.allSongs', // field allSongs in namespace library
                    argument: 'filter.artistName', // argument of library.Song
                    batchMode: false, // batchMode must be set to false
                    oneToMany: true,
                    linkFieldName: 'songsByThisArtist' // the "virtual" field that will be populated
                }
            }
        }
     }]
});

This assumes the library schema has a field allSongs which accepts a filter argument, where one of the filter properties is artistName. This also assumes that the recommendations schema has a type Artist with a field name which contains the artists name. This configuration will add new virtual field songsByThisArtist on the Artist type which will allow you to query an Artist along with their songs like this:

query {
    recommendations {
        myFavoriteArtists {
            name
            genre
            songsByThisArtist {
                id
                title
                year
            }
        }
    }
}

Joins

What if you want to sort the recommendations by the song age, or filter by artist? The recommendation service currently does not know about these fields, so it does not offer an API to sort or order by any of them. Using graphql-weaver, this problem is easily solved:

const schema: GraphQLSchema = await weaveSchemas({
    endpoints: [{
        namespace: 'library',
        url: 'http://example.com/library/graphql'
    }, {
        namespace: 'recommendations',
        url: 'http://example.com/recommendations/graphql',
        fieldMetadata: {
            'Recommendation.song': {
                link: {
                    field: 'library.allSongs',
                    argument: 'filter.ids',
                    batchMode: true, // is now required
                    keyField: 'id' // this one too
                }
            },
            'Query.myRecommendations': { // Field myRecommendations on type Query
                join: {
                    linkField: 'song', // The field name song in the type Recommendation
                }
            }
        }
     }]
});

This assumes that the library service offers a way to filter and sort songs via the orderBy and filter arguments. Using it is simple:

query {
    recommendations {
        myRecommendations(filter: { song: { artist: "Ed Sheeran" } }, orderBy: song_year_DESC) {
            recommendedAt
            song {
                id
                artist
                title
                year
            }
        }
    }
}

A note on efficiency: The list of recommendations should be relatively small (not more than a few hundred), as all recommendations need to be fetched so that their ids can be sent to the library for filtering and sorting.

Custom transformations

All four presented features (namespaces, type prefixes, links and joins) are implemented as independent modules. If you need something else, you can just write your own module:

class MyModule implements PipelineModule {
    transformExtendedSchema(schema: ExtendedSchema): ExtendedSchema {
        // do something with the schema
        return schema;
    }
    transformQuery(query: Query): Query {
        // do something with the query
        return query;
    }
}

const schema: GraphQLSchema = weaveSchemas({
    endpoints: [{
        namespace: 'library',
        url: 'http://example.com/library/graphql',
        
    }],
    pipelineConfig: {
        transformPreMergePipeline(modules: PipelineModule[], context: PreMergeModuleContext): PipelineModule[] {
            // These modules are executed for each endpoint
            return [
                ...modules,
                new MyModule()
            ]
        },
        transformPostMergePipeline(modules: PipelineModule[], context: PostMergeModuleContext): PipelineModule[] {
            // These modules are executed once for the merged schema
            return [
                ...modules,
                new MyModule()
            ]
        }
    }
});

For a simple module, see TypePrefixModule. The section Architecture below gives an overview over the pipeline architecture.

To simplify modifications to a schema, graphql-weaver ships graphql-transformer (and transformExtendedSchema). You can change types and fields as you like with a simple function:

const transformedSchema = transformSchema(originalSchema, {
    transformField(field: GraphQLNamedFieldConfig<any, any>, context) {
        // Rename a field in a type
        if (context.oldOuterType.name == 'MyType') {
            return {
                ...field,
                name: field.name + 'ButCooler'
            }
        }
        return field;
    },

    transformObjectType(type: GraphQLObjectTypeConfig<any, any>) {
        if (type.name == 'MyType') {
            return {
                ...type,
                name: 'MyCoolType'
            };
        }
        return type;
    },

    transformFields(fields: GraphQLFieldConfigMap<any, any>, context) {
        // You can even copy types on the fly and transform the copies
        const type2 = context.copyType(context.oldOuterType, {
            transformObjectType(typeConfig: GraphQLObjectTypeConfig<any, any>) {
                return {
                    ...typeConfig,
                    name: typeConfig.name + '2'
                };
            }
        });

        // This just adds a reflexive field "self" to all types, but its type does not have
        // the "self" field (because it is a copy from the original type, see above)
        // it also won't have the "cool" rename applied because the top-level transformers are not applied
        return {
            ...fields,
            self: {
                type: type2,
                resolve: (source: any) => source
            }
        }
    }
});

For more information, refer to the graphql-transformer project.

Schema error handling

By default, weaveSchemas throws a WeavingError if an endpoint schema could not be fetched or the link and join features are configured incorrectly. You can change this behavior with the errorHandling property:

weaveSchemas({
    endpoints: [ /* ... */ ],
    errorHandling: WeavingErrorHandlingMode.CONTINUE_AND_REPORT_IN_SCHEMA
})

There are four modes available:

  • THROW is the default behavior which throws all schema errors
  • CONTINUE ignores errors. If the endpoint schema cannot be created at all, it will be missing in the result config. If a join or link feature is malconfigured, this one configuration will be skipped. Use weaveSchemasExt to retrieve a list of errors.
  • CONTINUE_AND_REPORT_IN_SCHEMA behaves like CONTINUE, but errors are additionally displayed to the user via a special _errors field on the root query type.
  • CONTINUE_AND_ADD_PLACEHOLDERS is like CONTINUE_AND_REPORT_IN_SCHEMA, but namespaced endpoints that completely fail are also replaced by an object with a field _error.

Runtime error handling

Since version 0.11, graphql-weaver propagates runtime errors of endpoints transparently and in the most intuitive way.

  • If a user requested fields from two endpoints of which one failed and one succeeded, the result data of the successful endpoint will be returned and the error of the failing one will be propagated.
  • If the endpoint reported errors for some fields but returned data for other fields, graphql-weaver will pass this through exactly. The errors will be reported at the correct fields in the woven schema.
  • If the error reported by the endpoint included source location information, graphql-weaver will try to map this these source location to the woven schema. If the location can not be mapped, the field's location in the woven schema will be used, so if you get a location, you can be sure it is a location in your request to graphql-weaver.

For the most part, this means everything should work as expected. However, the mapping is more complex as it might seem, so if you find an error mapping that looks wrong, please open an issue.

Contributing

After cloning the repository, run

npm install
npm start

To run the test suite, run

npm test

To debug/run the application (or tests) in WebStorm, right-click on graphql-weaver.js (or graphql-weaver-tests.js, respectively) and choose Debug/Run.

Release workflow

  • For normal development, create a branch from master, commit and create a merge request to master.
  • To fix a bug in a previous release, find the release- branch for the corresponding version, increase the patch level in package.json and push the changes. Once the tests pass, manually trigger the deploy stage in Gitlab. You can also release a -rc.1 version before the actual release for prior testing in dependent modules.
  • To prepare a new feature release (currently, this means a new minor version), create a release-0.x branch from master. Set the version to 0.x-rc.1, push and manually trigger the deploy stage in Gitlab. Test the changes in dependent modules. Once everything is ok, change the version to 0.x and deploy again. Finally, merge the release branch into master. Do not delete the release branch as it is used for hotfixes.

Architecture

graphql-weaver takes a set of GraphQL endpoints, transforms them through pipelines, merges them, transforms the merged schema again and exposes that as its woven schema.

           +------+  +------+  +------+
Endpoints  |Schema|  |Schema|  |Schema|
           +------+  +------+  +------+

           +------+  +------+  +------+
            X    X    X    X    X    X
Pipelines    X  X      X  X      X  X
              XX        XX        XX

               +                  +
Merge           +----------------+

                     +------+
                      X    X
Pipeline               X  X
                        XX

                     +------+
Server               |Schema|
                     +------+

The merge in the middle simply merges all the fields of the Query/Mutation types. All the other features, like type prefixing, field namespacing, even resolvers, is implemented by pipeline modules.

You'll find the list of modules in src/pipeline/pipeline.ts. For a description of each module, please refer to the TypeDoc comments.

Module structure

  • graphql - general utilities for working with GraphQL schemas and queries
  • extended-schema - an implementation of storing and exposing metadata on fields, the concept being discussed on GitHub
  • graphql-client - GraphQL client library, with local and http implementations
  • pipeline - the core, being framework and modules for graphql-weaver's features
  • config - configuration parameter types for weaveSchemas
  • utils - utilities unrelated to GraphQL
  • typings - typings for thirdparty modules