| commit | 86494f11358840086b281a38cd39495b565e041f | [log] [tgz] |
|---|---|---|
| author | Bertrand Delacretaz <bdelacretaz@apache.org> | Thu Sep 10 12:15:39 2020 +0200 |
| committer | GitHub <noreply@github.com> | Thu Sep 10 12:15:39 2020 +0200 |
| tree | 55c80c7d9d39e9629d40b4fe4b32ec11c60fa293 | |
| parent | 6b73eb1ea66622a8ce44043b823f7e269a45e15a [diff] |
Add Persisted Query Hash section
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.
As I write this, work is ongoing at SLING-9550 to implement custom GraphQL Scalars
This module enables the following GraphQL “styles”
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.
Schemas are provided by SchemaProvider services:
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.
*/
@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.
The GraphQL schemas used by this module can be enhanced using schema directives (see also the Apollo docs for how those work) that select specific SlingDataFetcher services to return the appropriate data.
A default data fetcher is used for types and fields which have no such directive.
Here's a simple example, the test code has more:
# This directive maps fields to our Sling data fetchers
directive @fetcher(
name : String,
options : String = "",
source : String = ""
) on FIELD_DEFINITION
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 hava 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.
Besides Java, SlingDataFetcher scripts can be written in any scripting language that supported by the Sling instance's configuration.
Here's an example from the test code.
The schema contains the following statement:
scriptedFetcher (testing : String) : Test @fetcher(name:"scripted/example")
And here's the data fetcher code:
var result = {
boolValue: true,
resourcePath: "From the test script: " + resource.path,
testingArgument: environment.getArgument("testing"),
anotherValue: 450 + 1
};
result;
The data fetcher provider then looks for a script that handles the graphql/fetchers/scripted/example resource type with a fetcherscript name. graphql/fetchersis a prefix (hardcoded for now) and scripted/example comes from the above schema's @fetcher directive.
In that test, the /graphql/fetchers/scripted/example/fetcher.js shown above resolves with those requirements, it is executed to retrieve the requested data. That execution happens with a context consisting of the current SlingDataFetcherEnvironment under the environment key, and the current Sling Resource under the resource key, both used in this test script.
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.
/persisted request suffix; the value is configurable, via the persistedQueries.suffix parameter of the factory configuration.POSTing the query text to the endpoint where the GraphQL servlet is bound, plus the /persisted suffix.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.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.404 on a persisted query must be prepared to rePOST the query, in order to store the prepared query again.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 is already cache this saves the POST request, and if not the client gets a 404 status and has to POST the query first.
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 < Content-Length: 0
curl -v http://localhost:8080/graphql.json/persisted/e1ce2e205e1dfb3969627c6f417860cadab696e0e87b1c44de1438848661b62f > GET /graphql.json/persisted/e1ce2e205e1dfb3969627c6f417860cadab696e0e87b1c44de1438848661b62f 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" ] } ] } ] } }