graphql skill

GraphQL

Default to REST. GraphQL trades away things API Platform gives you for free over REST: HTTP cache semantics (ETag, Cache-Control, invalidation), one URL per resource, simple CDN/proxy caching, and predictable per-operation cost. A single GraphQL query can fan out into arbitrarily deep/expensive resolution. Reach for GraphQL when it is a hard client requirement (e.g. a Relay/Apollo frontend, or clients that genuinely need to select fields and avoid round-trips) — not as a default. The same resource class can serve both; you don't have to choose globally.

Enabling GraphQL

Install api-platform/graphql (composer require api-platform/graphql), then it's on. A /graphql endpoint and the GraphiQL playground (/graphql/graphiql) appear. Disable globally or per resource as needed:

# config/packages/api_platform.yaml
api_platform:
    graphql:
        enabled: true
        graphiql:
            enabled: true

Declaring GraphQL operations

GraphQL operations live in graphQlOperations and are separate classes from the REST ones, under ApiPlatform\Metadata\GraphQl\. A resource with no graphQlOperations still gets a default set (item query, collection query, create / update / delete mutations) once GraphQL is enabled.

use ApiPlatform\Metadata\ApiResource;
use ApiPlatform\Metadata\GraphQl\DeleteMutation;
use ApiPlatform\Metadata\GraphQl\Mutation;
use ApiPlatform\Metadata\GraphQl\Query;
use ApiPlatform\Metadata\GraphQl\QueryCollection;

#[ApiResource(graphQlOperations: [
    new Query(),
    new QueryCollection(),
    new Mutation(name: 'create'),
    new Mutation(name: 'update'),
    new DeleteMutation(name: 'delete'),
])]
class Book {}

Mutation and DeleteMutation require a name — it becomes the GraphQL field name (createBook, updateBook, deleteBook). Query and QueryCollection only need a name when you declare more than one of the same kind (e.g. a custom query alongside the default).

Security on GraphQL operations

Same ExpressionLanguage as REST (see operations), set per GraphQl operation:

new Query(security: "is_granted('ROLE_USER')")
new Mutation(name: 'update', security: "object.getOwner() == user")
new Mutation(name: 'update', securityPostDenormalize: "object.getOwner() == user")

Because one GraphQL query can traverse relations, securing only the top-level operation is not enough — guard the related resources' operations too, or a nested field becomes an unguarded read path.

Custom resolvers

Use a resolver when a query/mutation needs logic beyond fetch-by-id. Resolvers are services implementing one of:

• QueryItemResolverInterface — __invoke(?object $item, array $context): object
• QueryCollectionResolverInterface — __invoke(iterable $collection, array $context): iterable
• MutationResolverInterface — __invoke(?object $item, array $context): ?object

Query arguments arrive in $context['args'].

use ApiPlatform\GraphQl\Resolver\QueryItemResolverInterface;

final class BookResolver implements QueryItemResolverInterface
{
    public function __invoke(?object $item, array $context): object
    {
        // $item is the fetched Book (or null if read: false); enrich or replace it
        return $item;
    }
}

With Symfony autoconfiguration the resolver is wired automatically. Without it, tag the service api_platform.graphql.query_resolver (or ..._mutation_resolver). Then reference it by class name on the operation, and set read: false when the resolver should fetch the data itself rather than receiving a hydrated item:

new Query(name: 'recommended', resolver: BookResolver::class, read: false)
new Query(
    name: 'search',
    resolver: BookResolver::class,
    args: [
        'query' => ['type' => 'String!', 'description' => 'Full-text search'],
        'limit' => ['type' => 'Int'],
    ],
)

args overrides the auto-generated argument set — define it when the query takes parameters that aren't resource fields.

Relations and the N+1 trap

GraphQL embeds related resources by selecting nested fields:

{
  book(id: "/books/1") {
    title
    author { name }
  }
}

Relations resolve through the same providers as REST. A deeply nested query can trigger many small queries (the classic N+1). API Platform mitigates common cases, but verify against real queries and add Doctrine joins / a custom collection provider where a hot path fans out. This open-ended cost is the main reason REST is the safer default for cache-sensitive APIs.

Pagination: Relay cursor connections

Collection queries return Relay-style cursor connections by default (edges { node { ... } cursor }, pageInfo, totalCount), driven by first/ after/last/before arguments:

{
  books(first: 10, after: "endCursor") {
    totalCount
    edges { node { title } cursor }
    pageInfo { endCursor hasNextPage }
  }
}

To use simple page-based pagination instead, set paginationType: 'page' on the resource or the QueryCollection:

use ApiPlatform\Metadata\GraphQl\QueryCollection;

#[ApiResource(graphQlOperations: [
    new QueryCollection(paginationType: 'page'),
])]
class Book {}

All the pagination* controls from pagination (items per page, max, partial) apply to GraphQL collections too.

Real-time subscriptions

When a resource has mercure: true, GraphQL subscription operations push updates through the Mercure hub — see mercure.

Laravel

GraphQL is supported on Laravel. The operation classes (Query, QueryCollection, Mutation, DeleteMutation), security expressions, Relay connections, paginationType, custom resolvers and the N+1 caveats are all the same. Differences:

• Install composer require api-platform/graphql, then enable it in config/api-platform.php under graphql ('enabled' => true) — not YAML. Depth/ complexity limits and graphiql live in the same config block.
• Resolvers implement the same QueryItemResolverInterface / QueryCollectionResolverInterface / MutationResolverInterface and are referenced by class-string on the operation; Laravel's container resolves them — there are no api_platform.graphql.*_resolver tags to apply.
• Real-time subscription operations depend on Mercure, which has no Laravel integration (see mercure), so GraphQL subscriptions are effectively Symfony-only.

Checklist

• GraphQL chosen for a real client requirement, not as a REST default
• Every Mutation/DeleteMutation has a name
• security set on nested resources, not just the entry-point operation
• Custom resolvers tagged (or autoconfigured) and referenced by class name
• read: false set when the resolver fetches its own data
• Deep/nested queries checked for N+1; joins added on hot paths
• paginationType: 'page' set only if the client doesn't want Relay connections