Graphql

GraphQL API Design and Implementation Rules

1. Schema Design Principles

SDL-first vs Code-first

• SDL-first (recommended for most projects): Define the schema in .graphqls files, then implement resolvers to match
  • Encourages contract-first design — schema is the API contract
  • Better for team collaboration — frontend and backend agree on schema first
  • Tools: Spring for GraphQL, Apollo Server
• Code-first: Generate the schema from code annotations or DSL
  • Useful when schema closely mirrors domain model
  • Risk of coupling schema to implementation details
  • Tools: Netflix DGS (annotation-based), GraphQL Kotlin, TypeGraphQL

Schema Organization

src/main/resources/graphql/
  schema.graphqls          # Root schema (Query, Mutation, Subscription)
  types/
    user.graphqls          # User-related types
    order.graphqls         # Order-related types
  inputs/
    user-input.graphqls    # User input types
  enums/
    status.graphqls        # Enum definitions
  interfaces/
    node.graphqls          # Shared interfaces

Naming Conventions

• Types: PascalCase (User, OrderItem)
• Fields: camelCase (firstName, createdAt)
• Enums: PascalCase type, SCREAMING_SNAKE_CASE values (OrderStatus.IN_PROGRESS)
• Mutations: verb + noun (createUser, updateOrder, deleteComment)
• Queries: noun for single (user), plural for list (users)
• Input types: suffix with Input (CreateUserInput, UpdateOrderInput)
• Payload types: suffix with Payload (CreateUserPayload)

---

2. Type System

Object Types

type User {
  id: ID!
  email: String!
  name: String!
  avatar: String
  orders(first: Int, after: String): OrderConnection!
  createdAt: DateTime!
}

• Use ID! for identifiers — never Int! or String!
• Mark required fields with ! (non-null)
• Use nullable fields (String) for optional data
• Add arguments to fields for filtering/pagination

Input Types

input CreateUserInput {
  email: String!
  name: String!
  avatar: String
}

input UpdateUserInput {
  email: String
  name: String
  avatar: String
}

• Separate input types for create and update operations
• Create inputs: required fields are non-null
• Update inputs: all fields nullable (partial update)
• Never reuse output types as input types

Interface Types

interface Node {
  id: ID!
}

interface Timestamped {
  createdAt: DateTime!
  updatedAt: DateTime!
}

type User implements Node & Timestamped {
  id: ID!
  email: String!
  createdAt: DateTime!
  updatedAt: DateTime!
}

• Use interfaces for shared field contracts
• Node interface is standard for Relay-compatible APIs

Union Types

union SearchResult = User | Order | Product

type Query {
  search(term: String!): [SearchResult!]!
}

• Use unions when return types share no common fields
• Prefer interfaces when types share fields
• Always handle all union members in the client

Enum Types

enum OrderStatus {
  PENDING
  CONFIRMED
  SHIPPED
  DELIVERED
  CANCELLED
}

• Use enums for fixed sets of values
• Use SCREAMING_SNAKE_CASE for values
• Add descriptions for non-obvious values

Custom Scalar Types

scalar DateTime
scalar URL
scalar JSON
scalar BigDecimal

• Define custom scalars for domain-specific types
• Always provide serialization/deserialization logic
• Prefer specific scalars over String for validation

---

3. Resolver Patterns and DataLoader

Resolver Structure

Query.user(id) → UserResolver
  User.orders → OrderResolver (field resolver)
    Order.items → OrderItemResolver (field resolver)

• Keep resolvers thin — delegate business logic to service layer
• One resolver class per type or domain area
• Field resolvers load data lazily — only when requested

See references/advanced-patterns.md for N+1 problem explanation, DataLoader implementation example, and mutation error patterns.

DataLoader Rules

• Always use DataLoader for field resolvers that load related entities
• DataLoader batches requests within a single request context
• DataLoader also provides per-request caching
• Monitor batch sizes — unexpectedly large batches may indicate issues

Resolver Best Practices

• Never call repositories directly from resolvers — use a service layer
• Handle null values explicitly in resolvers
• Use @Secured or custom directives for field-level authorization
• Log resolver errors with correlation IDs for debugging

---

4. Pagination

Cursor-based Pagination (Relay Connection Spec)

type Query {
  users(first: Int, after: String, last: Int, before: String): UserConnection!
}

type UserConnection {
  edges: [UserEdge!]!
  pageInfo: PageInfo!
  totalCount: Int!
}

type UserEdge {
  node: User!
  cursor: String!
}

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

Pagination Guidelines

• Prefer cursor-based over offset-based pagination
  • Cursor-based: stable under concurrent inserts/deletes
  • Offset-based: simple but breaks with data changes
• Cursors should be opaque strings (base64-encoded IDs)
• Always include totalCount when feasible
• Set reasonable defaults and maximums for first/last (e.g., default 20, max 100)
• For simple use cases, offset-based pagination is acceptable:

type Query {
  users(page: Int = 1, size: Int = 20): UserPage!
}

type UserPage {
  content: [User!]!
  totalElements: Int!
  totalPages: Int!
  hasNext: Boolean!
}

---

5. Error Handling

GraphQL Error Response Structure

{
  "data": null,
  "errors": [
    {
      "message": "User not found",
      "locations": [{ "line": 2, "column": 3 }],
      "path": ["user"],
      "extensions": {
        "code": "NOT_FOUND",
        "classification": "DataFetchingException"
      }
    }
  ]
}

Error Handling Patterns

• Use extensions.code for machine-readable error codes

• Define a consistent set of error codes across the API:

  • NOT_FOUND, VALIDATION_ERROR, UNAUTHORIZED, FORBIDDEN, INTERNAL_ERROR, CONFLICT, RATE_LIMITED

• Return partial data when possible — GraphQL supports partial responses

• Never expose internal details (stack traces, SQL) in error messages

• Use union return types for mutations to provide typed error responses (see references/advanced-patterns.md for union-based mutation error pattern examples)

---

6. Query Complexity and Depth Limiting

Depth Limiting

• Set a maximum query depth (recommended: 7-10 for general APIs)
• Reject queries exceeding the depth limit before execution

# This nested query could be malicious if unbounded:
{
  user {
    friends {
      friends {
        friends {
          friends { ... }  # Excessive depth
        }
      }
    }
  }
}

Complexity Analysis

• Assign complexity scores to fields based on cost:
  • Scalar fields: 0-1
  • Object fields: 1
  • List fields: multiply by expected/requested count
  • Database-heavy fields: higher weight
• Set a maximum total complexity per query (e.g., 1000)
• Return the complexity cost in response extensions for transparency

For DGS framework configuration example, see references/advanced-patterns.md.

---

7. Authentication and Authorization

Authentication

• Authenticate at the transport layer (HTTP headers, cookies)
• Pass the authenticated user context to resolvers via GraphQL context
• Do NOT put authentication logic inside individual resolvers

Field-Level Authorization

type User {
  id: ID!
  name: String!                  # Public
  email: String! @auth(role: SELF)     # Only the user themselves
  salary: BigDecimal @auth(role: ADMIN) # Admin only
}

• Use custom directives or schema annotations for declarative authorization
• Implement authorization in a middleware/interceptor layer, not in business logic
• Return null for unauthorized nullable fields; throw for non-null fields
• Log authorization failures for security monitoring

Authorization Patterns

• Directive-based: @auth(role: ADMIN) on schema fields
• Resolver-based: Check permissions in the resolver or service layer
• DataLoader-based: Filter results in DataLoader based on context

---

8. Subscriptions (Real-time)

Subscription Design

type Subscription {
  orderStatusChanged(orderId: ID!): OrderStatusEvent!
  newMessage(channelId: ID!): Message!
}

type OrderStatusEvent {
  order: Order!
  previousStatus: OrderStatus!
  newStatus: OrderStatus!
  changedAt: DateTime!
}

Implementation Guidelines

• Use WebSocket (graphql-ws protocol) for subscriptions
• Keep subscription payloads small — include only changed data
• Always require authentication for subscriptions
• Set connection timeouts and maximum subscription limits per client
• Use server-sent events (SSE) as an alternative for simpler use cases
• Consider using a message broker (Redis Pub/Sub, Kafka) for scalability

---

9. Federation (Apollo Federation)

When to Use Federation

• Multiple teams own different parts of the graph
• Microservice architecture where each service owns its domain types
• Need to compose a unified schema from independent subgraphs

For subgraph schema examples, see references/advanced-patterns.md.

Federation Guidelines

• Each entity has exactly one owning subgraph
• Use @key directive to define entity identity
• Use @external, @requires, @provides for cross-subgraph fields
• Keep the gateway/router stateless — all logic in subgraphs
• Monitor subgraph latency — gateway adds overhead
• Version subgraphs independently; use schema registry for compatibility

---

10. Security

For the full security reference, see the OWASP GraphQL Cheat Sheet.

Introspection

• Disable introspection in production — it exposes the full schema
• Enable only in development/staging environments

Query Restrictions

• Set maximum query depth (see Section 6)
• Set maximum query complexity (see Section 6)
• Limit batch queries (array of operations in single request)
• Set request size limits at the HTTP layer
• Implement rate limiting per client/IP

Input Validation

• Validate all input arguments at the schema level (non-null, enums)
• Add custom validation in resolvers for business rules
• Sanitize string inputs to prevent injection attacks
• Use custom scalars with built-in validation (e.g., Email, URL)

Persisted Queries

• Use automatic persisted queries (APQ) or pre-registered queries in production
• APQ: client sends a hash; server looks up the query
• Prevents arbitrary query execution in sensitive environments
• Reduces bandwidth by not sending full query strings

Additional Security Measures

• Enable CORS with strict origin policies
• Use HTTPS for all GraphQL endpoints
• Log and monitor query patterns for anomaly detection
• Set timeouts for resolver execution to prevent resource exhaustion

---

11. GraphQL vs REST Selection Criteria

Prefer GraphQL When

• Clients need flexible data fetching (mobile vs web different needs)
• Multiple related entities are fetched in a single request
• API serves diverse clients with varying data requirements
• Real-time features (subscriptions) are needed
• Rapid frontend iteration without backend changes

Prefer REST When

• Simple CRUD operations on well-defined resources
• File upload/download is a primary use case
• Caching at the HTTP layer (CDN, browser) is critical
• Team has limited GraphQL experience
• API is primarily server-to-server communication
• OpenAPI/Swagger tooling ecosystem is a requirement

Hybrid Approach

• Use GraphQL for client-facing APIs with complex data requirements
• Use REST for internal service-to-service communication
• Use REST for file handling, health checks, and webhook endpoints
• A single application can expose both GraphQL and REST endpoints

---

12. Framework Reference

For detailed framework-specific implementation patterns with Spring for GraphQL and Netflix DGS, see references/spring-graphql-dgs.md.

For advanced schema design patterns, naming conventions, and mutation design guidelines, see references/schema-design-patterns.md.