Apache Sling GraphQL Core

This module is one of several which provide GraphQL support for Apache Sling.

This module allows for running GraphQL queries in Sling, using dynamically built GraphQL schemas and OSGi services for data fetchers (aka "resolvers") which provide the data.

To take advantage of Sling's flexibility, it allows for running GraphQL queries in three different modes, using client or server-side queries and optionally being bound to the current Sling Resource.

Server-side queries are implemented as a Sling Script Engine.

The current version uses the graphql-java library but that's only used internally. The corresponding OSGi bundles must be active in your Sling instance but there's no need to use their APIs directly.

The GraphQL sample website provides usage examples and demonstrates using GraphQL queries (and Handlebars templates) on both the server and client sides. It's getting a bit old and as of June 2021 doesn't demonstrate the latest features.

As usual, the truth is in the tests. If something's missing from this page you can probably find the details in this module's extensive test suite.

Supported GraphQL endpoint styles

This module enables the following GraphQL "styles"

• The traditional GraphQL endpoint style, where the clients supply requests to a single URL. It is easy to define multiple such endpoints with different settings, which can be useful to provide different "views" of your content.
• A Resource-based GraphQL endpoints style where every Sling Resource can be a GraphQL endpoint (using specific request selectors and extensions) where queries are executed in the context of that Resource. This is an experimental idea at this point but it's built into the design so doesn't require more efforts to support. That style supports both server-side "prepared GraphQL queries" and the more traditional client-supplied queries.

The GraphQL requests hit a Sling resource in all cases, there's no need for path-mounted servlets which are not desirable.

See also the caching section later in this file.

Configuring the GraphQL Servlet

Here's an excerpt from an OSGi feature model file which uses the GraphQL Servlet provided by this module to serve .json requests for resources which have the samples/graphql resource type:

"configurations":{
  "org.apache.sling.graphql.core.GraphQLServlet~default" : {
    "sling.servlet.resourceTypes" : "samples/graphql",
    "sling.servlet.extensions": "json",
    "sling.servlet.methods": [ "GET", "POST" ]
  },
  "org.apache.sling.servlets.get.DefaultGetServlet" : {
    "aliases" : [ "json:rawjson" ]
  },

The rawjson selector is configured to provide Sling's default JSON output.

See the GraphQLServlet class for more info.

Resource-specific GraphQL schemas

Schemas are provided by SchemaProvider services:

@ProviderType
public interface SchemaProvider {
  
    /** Get a GraphQL Schema definition for the given resource and optional selectors
     *
     *  @param r The Resource to which the schema applies
     *  @param selectors Optional set of Request Selectors that can influence the schema selection
     *  @return a GraphQL schema that can be annotated to define the data fetchers to use, see
     *      this module's documentation. Can return null if a schema cannot be provided, in which
     *      case a different provider should be used.
     *  @throws java.io.IOException if the schema cannot be retrieved
     */
    @Nullable
    String getSchema(@NotNull Resource r, @Nullable String [] selectors) throws IOException;
}

The default provider makes an internal Sling request with for the current Resource with a .GQLschema extension.

This allows the Sling script/servlet resolution mechanism and its script engines to be used to generate schemas dynamically, taking request selectors into account.

Unless you have specific needs not covered by this mechanism, there's no need to implement your own SchemaProvider services.

Built-in GraphQL Schema Directives

Since version 0.0.10 of this module, a number of GraphQL schema directives are built-in to support specific features. As of that version, the @fetcher, @resolver and @connection directives described below can be used directly, without having to declare them explicitly in the schema with the directive statement that was required before.

Declaring these directives explicitly is still supported for backwards compatibility with existing schemas, but not needed anymore.

SlingDataFetcher selection using the @fetcher directive

The following built-in @fetcher directive is defined by this module:

    # This directive maps fields to our Sling data fetchers
    directive @fetcher(
        name : String!,
        options : String = "",
        source : String = ""
    ) on FIELD_DEFINITION

It allows for selecting a specific SlingDataFetcher service to return the appropriate data, as in the examples below.

Fileds which do not have such a directive will be retrieved using the default data fetcher.

Here are a few examples, the test code has more of them:

    type Query {
      withTestingSelector : TestData @fetcher(name:"test/pipe")
    }

    type TestData {
      farenheit: Int @fetcher(name:"test/pipe" options:"farenheit")
    }

The names of those SlingDataFetcher services are in the form

<namespace>/<name>

The sling/ namespace is reserved for SlingDataFetcher services which have Java package names that start with org.apache.sling.

The <options> and <source> arguments of the directive can be used by the SlingDataFetcher services to influence their behavior.

SlingTypeResolver selection using the @resolver directive

The following built-in @resolver directive is defined by this module:

    # This directive maps the corresponding type resolver to a given Union
    directive @resolver(
        name: String!, 
        options: String = "", 
        source: String = ""
    ) on UNION | INTERFACE

A Union or Interface type can provide a @resolver directive, to select a specific SlingTypeResolver service to return the appropriate GraphQL object type.

Here's a simple example, the test code has more:

union TestUnion @resolver(name : "test/resolver", source : "TestUnion") = Type_1 | Type_2 | Type_3 | Type_4

The names of those SlingTypeResolver services are in the form

<namespace>/<name>

The sling/ namespace is reserved for SlingTypeResolver services which have Java package names that start with org.apache.sling.

The <options> and <source> arguments of the directive can be used by the SlingTypeResolver services to influence their behavior.

Result Set Pagination using the @connection and @fetcher directives

This module implements support for the Relay Cursor Connections specification, via the built-in @connection directive, coupled with a @fetcher directive. The built-in @connection directive has the following definition:

    directive @connection(
      for: String!
    ) on FIELD_DEFINITION

To allow schemas to be ehanced with pagination support, like in this example:

    type Query {
        paginatedHumans (after : String, limit : Int) : HumanConnection @connection(for: "Human") @fetcher(name:"humans/connection")
    }

    type Human {
        id: ID!
        name: String!
        address: String
    }

Using this directive as in the above example adds the following types to the schema to provide paginated output that follows the Relay spec:

    type PageInfo {
        startCursor : String
        endCursor : String
        hasPreviousPage : Boolean
        hasNextPage : Boolean
    }

    type HumanEdge {
        cursor: String
        node: Human
    }

    type HumanConnection {
        edges : [HumanEdge]
        pageInfo : PageInfo
    }

How to implement a SlingDataFetcher that provides a paginated result set

The GenericConnection class, together with the org.apache.sling.graphql.api.pagination API provide support for paginated results. With this utility class, you just need to supply an Iterator on your data, a function to generate a string that represents the cursor for a given object, and optional parameters to control the page start and length.

The QueryDataFetcherComponent provides a usage example:

    @Override
    public Object get(SlingDataFetcherEnvironment env) throws Exception {
      // fake test data simulating a query
      final List<Resource> data = new ArrayList<>();
      data.add(env.getCurrentResource());
      data.add(env.getCurrentResource().getParent());
      data.add(env.getCurrentResource().getParent().getParent());

      // Define how to build a unique cursor that points to one of our data objects
      final Function<Resource, String> cursorStringProvider = r -> r.getPath();

      // return a GenericConnection that the library will introspect and serialize
      return new GenericConnection.Builder<>(data.iterator(), cursorStringProvider)
        .withLimit(5)
        .build();
    }

The above data fetcher code produces the following output, with the GenericConnection helper taking care of the pagination logic and of generating the required data. This follows the Relay Connections specification, which some GraphQL clients should support out of the box.

    {
      "data": {
        "oneSchemaQuery": {
          "pageInfo": {
            "startCursor": "L2NvbnRlbnQvZ3JhcGhxbC9vbmU=",
            "endCursor": "L2NvbnRlbnQ=",
            "hasPreviousPage": false,
            "hasNextPage": false
          },
          "edges": [
            {
              "cursor": "L2NvbnRlbnQvZ3JhcGhxbC9vbmU=",
              "node": {
                "path": "/content/graphql/one",
                "resourceType": "graphql/test/one"
              }
            },
            {
              "cursor": "L2NvbnRlbnQvZ3JhcGhxbA==",
              "node": {
                "path": "/content/graphql",
                "resourceType": "graphql/test/root"
              }
            },
            {
              "cursor": "L2NvbnRlbnQ=",
              "node": {
                "path": "/content",
                "resourceType": "sling:OrderedFolder"
              }
            }
          ]
        }
      }
    }

Usage of this GenericConnection helper is optional, although recommended for ease of use and consistency. As long as the SlingDataFetcher provides a result that implements the org.apache.sling.graphql.api.pagination.Connection, the output will be according to the Relay spec.

Lazy Loading of field values

The org.apache.sling.graphql.helpers.lazyloading package provides helpers for lazy loading field values.

Using this pattern, for example:

    public class ExpensiveObject {
      private final LazyLoadingField<String> lazyName;

      ExpensiveObject(String name) {
        lazyName = new LazyLoadingField<>(() -> {
          // Not really expensive but that's the idea
          return name.toUpperCase();
        });
      }

      public String getExpensiveName() {
        return lazyName.get();
      }
    }

The expensiveName is only computed if its get method is called. This avoids executing expensive computations for fields that are not used in the GraphQL result set.

A similar helper is provided for Maps with lazy loaded values.

Caching: Persisted queries API

No matter how you decide to create your Sling GraphQL endpoints, you have the option to allow GraphQL clients to use persisted queries.

After preparing a query with a POST request, it can be executed with a GET request that can be cached by HTTP caches or a CDN.

This is required as POST queries are usually not cached, and if using GET with the query as a parameter there's a concrete risk of the parameter becoming too large for HTTP services and intermediates.

How to use persisted queries?

1. An instance of the GraphQL servlet has to be configured; by default, the servlet will enable the persisted queries API on the /persisted request suffix; the value is configurable, via the persistedQueries.suffix parameter of the factory configuration.
2. A client prepares a persisted query in advance by POSTing the query text to the endpoint where the GraphQL servlet is bound, plus the /persisted suffix.
3. The servlet will respond with a 201 Created status; the response's Location header will then instruct the client where it can then execute the persisted query, via a GET request.
4. The responses for a GET requests to a persisted query will contain appropriate HTTP Cache headers, allowing front-end HTTP caches (e.g. CDNs) to cache the JSON responses.
5. There's no guarantee on how long a persisted query is stored. A client that gets a 404 on a persisted query must be prepared to rePOST the query, in order to store the prepared query again.

Persisted Query Hash

The hash that's part of the persisted URL is computed on the POSTed GraphQL query by the active GraphQLCacheProvider service. By default, this is the SimpleGraphQLCacheProvider which computes it as follows:

MessageDigest digest = MessageDigest.getInstance("SHA-256");
byte[] hash = digest.digest(query.getBytes(StandardCharsets.UTF_8));

and encodes it in hex to build the persisted query's path.

This means that, if desired, an optimistic client can compute the hash itself and try a GET to the persisted/<hash> URL without doing a POST first. If the query already exists in the cache this saves the POST request, and if not the client gets a 404 status and has to POST the query first.

Example HTTP interactions with persisted queries enabled

1. Storing a query

   curl -v 'http://localhost:8080/graphql.json/persisted' \
     -H 'Content-Type: application/json' \
     --data-binary '{"query":"{\n  navigation {\n    search\n    sections {\n      path\n      name\n    }\n  }\n  article(withText: \"virtual\") {\n    path\n    title\n    seeAlso {\n      path\n      title\n      tags\n    }\n  }\n}\n","variables":null}' \
     --compressed
   > POST /graphql.json/persisted HTTP/1.1
   > Host: localhost:8080
   > User-Agent: curl/7.64.1
   > Accept: */*
   > Accept-Encoding: deflate, gzip
   > Content-Type: application/json
   > Content-Length: 236
   >
   * upload completely sent off: 236 out of 236 bytes
   < HTTP/1.1 201 Created
   < Date: Mon, 31 Aug 2020 16:33:48 GMT
   < X-Content-Type-Options: nosniff
   < X-Frame-Options: SAMEORIGIN
   < Location: http://localhost:8080/graphql.json/persisted/e1ce2e205e1dfb3969627c6f417860cadab696e0e87b1c44de1438848661b62f.json
   < Content-Length: 0

2. Running a persisted query

curl -v http://localhost:8080/graphql.json/persisted/e1ce2e205e1dfb3969627c6f417860cadab696e0e87b1c44de1438848661b62f.json
> GET /graphql.json/persisted/e1ce2e205e1dfb3969627c6f417860cadab696e0e87b1c44de1438848661b62f.json HTTP/1.1
> Host: localhost:8080
> User-Agent: curl/7.64.1
> Accept: */*
>
< HTTP/1.1 200 OK
< Date: Mon, 31 Aug 2020 16:35:18 GMT
< X-Content-Type-Options: nosniff
< X-Frame-Options: SAMEORIGIN
< Cache-Control: max-age=60
< Content-Type: application/json;charset=utf-8
< Transfer-Encoding: chunked
<

{
  "data": {
    "navigation": {
      "search": "/content/search",
      "sections": [
        {
          "path": "/content/articles/travel",
          "name": "Travel"
        },
        {
          "path": "/content/articles/music",
          "name": "Music"
        }
      ]
    }
    "article": [
      {
        "path": "/content/articles/travel/precious-kunze-on-the-bandwidth-of-virtual-nobis-id-aka-usb",
        "title": "Travel - Precious Kunze on the bandwidth of virtual 'nobis id' (aka USB)",
        "seeAlso": [
          {
            "path": "/content/articles/travel/solon-davis-on-the-card-of-primary-reiciendis-omnis-aka-sql",
            "title": "Travel - Solon Davis on the card of primary 'reiciendis omnis' (aka SQL)",
            "tags": [
              "bandwidth",
              "protocol"
            ]
          }
        ]
      }
    ]
  }
}

Planned Extensions / Wishlist

Selector-driven prepared queries (planned)

Described in SLING-10540: prepared GraphQL queries hidden behind URL selectors, so that an HTTP GET request to /content/mypage.A.full.json executes the GraphQL query previously stored under the A.full name.

Schema Aggregator (planned)

An initial spec, without code so far, is available at sling-whiteboard:sling-org-apache-sling-graphql-schema for a schema aggregator that allows OSGi bundles to contribute partial GraphQL schemas to an overall schema.

This will allow bundles to contribute specific sets of types to a schema, along with the code that implements their retrieval and other operations.

Object Query Service (whishlist)

The Object Query service runs queries against the Sling Resource tree and returns POJOs in a way that's optimized for the GraphQL Core to consume.

Probably something along those lines:

new Query(
  """
  select Folder
  from /tmp, /conf
  where Folder.title contains 'sling'
  and Folder.lastModified < 1w
  """)
.getIterator();

which returns an Iterator optimized for this module's pagination features.

The objects that this Iterator supplies can be built from Sling Models, using the lazy loading helpers provided by this module. This would probably need an extension to Sling Models where the appropriate Model can be found for a name like Folder in the above example. The Model class might be annotated in a way that allows it to supply XPath query elements for expressions like Folder.title.

The Query might have additional options such as withXpathGenerator, withObjectMapper for edge cases where the built-in logic is not sufficient.