harsh040506/graphql-api-design

GraphQL API Design

Production-quality patterns for GraphQL schema design, resolver implementation, and performance.

---

Schema Design Principles

Start From the Consumer, Not the Data Model

GraphQL is about the data consumers need, not the database schema. The two should often diverge.

# Bad: Leaking database schema into the API
type orders_table {
  order_id: Int!
  customer_fk: Int!
  order_status_enum: String!
  created_timestamp: Int!   # Unix epoch — hard for consumers
}

# Good: Consumer-friendly API
type Order {
  id: ID!
  status: OrderStatus!
  customer: Customer!         # Resolved via join or DataLoader, not FK
  createdAt: DateTime!        # ISO 8601, not epoch
  lineItems: [LineItem!]!
  totalAmount: Money!         # Custom scalar, not raw integer
}

enum OrderStatus {
  PENDING
  PROCESSING
  SHIPPED
  DELIVERED
  CANCELLED
}

Naming Conventions

Types:        PascalCase           — User, OrderLineItem, ShippingAddress
Fields:       camelCase            — firstName, createdAt, totalAmountCents
Enums:        PascalCase type      — OrderStatus, PaymentMethod
Enum values:  ALL_CAPS_UNDERSCORE  — PENDING, ON_HOLD, PARTIALLY_REFUNDED
Mutations:    verb + Noun          — createOrder, updateUser, cancelSubscription
Inputs:       PascalCase + Input   — CreateOrderInput, UpdateUserInput

Non-Null Discipline

# Use ! (non-null) when the value is guaranteed and meaningful
type User {
  id: ID!              # Always present — use !
  email: String!       # Always present — use !
  firstName: String!   # Always present — use !
  avatar: String       # Optional — may be null, no !
  deletedAt: DateTime  # Null = not deleted, no !
}

# Return null === "doesn't exist"
# Throw error === "something went wrong trying to get it"
# Never return null to mean "error occurred"

---

Complete Schema Template

# Custom scalars
scalar DateTime    # ISO 8601 string
scalar UUID        # UUID v4 string
scalar JSON        # Arbitrary JSON value (use sparingly)

# Pagination (Relay Connection Spec — widely supported)
type PageInfo {
  hasNextPage: Boolean!
  hasPreviousPage: Boolean!
  startCursor: String
  endCursor: String
}

# Generic connection types (repeat per type)
type OrderConnection {
  edges: [OrderEdge!]!
  pageInfo: PageInfo!
  totalCount: Int!
}

type OrderEdge {
  node: Order!
  cursor: String!
}

# Core types
type Order {
  id: ID!
  status: OrderStatus!
  customer: Customer!
  lineItems: [LineItem!]!
  subtotalCents: Int!
  taxCents: Int!
  totalCents: Int!
  shippingAddress: Address!
  createdAt: DateTime!
  updatedAt: DateTime!
  cancelledAt: DateTime
  cancelReason: String
}

enum OrderStatus {
  PENDING
  PROCESSING
  SHIPPED
  DELIVERED
  CANCELLED
}

# Queries
type Query {
  order(id: ID!): Order                      # null = not found; error = system failure
  orders(
    first: Int
    after: String
    last: Int
    before: String
    filter: OrderFilter
    orderBy: OrderOrderBy
  ): OrderConnection!
  
  currentUser: User                          # null = not authenticated
}

# Mutations
type Mutation {
  createOrder(input: CreateOrderInput!): CreateOrderPayload!
  cancelOrder(input: CancelOrderInput!): CancelOrderPayload!
}

# Mutation inputs
input CreateOrderInput {
  items: [OrderItemInput!]!
  shippingAddressId: ID!
  paymentMethodId: ID!
}

input OrderItemInput {
  productId: ID!
  quantity: Int!
}

# Mutation payloads (always return the affected resource + errors)
type CreateOrderPayload {
  order: Order              # null if creation failed
  userErrors: [UserError!]! # Empty array = success
}

type UserError {
  message: String!
  field: [String!]!         # Path to the field that failed (e.g., ["items", "0", "quantity"])
  code: String!             # Machine-readable code
}

# Filters and ordering
input OrderFilter {
  status: OrderStatus
  createdAfter: DateTime
  createdBefore: DateTime
  customerId: ID
}

enum OrderOrderBy {
  CREATED_AT_ASC
  CREATED_AT_DESC
  TOTAL_CENTS_ASC
  TOTAL_CENTS_DESC
}

# Subscriptions (for real-time updates)
type Subscription {
  orderStatusChanged(orderId: ID!): OrderStatusChangedPayload!
}

type OrderStatusChangedPayload {
  order: Order!
  previousStatus: OrderStatus!
}

---

Resolver Implementation

Basic Resolver Structure

// TypeScript resolver with DataLoader for N+1 prevention
import DataLoader from 'dataloader';

// Resolver map matches schema shape
const resolvers = {
  Query: {
    order: async (_parent, { id }, ctx) => {
      // Authorization check before data access
      if (!ctx.user) throw new AuthenticationError('Must be logged in');
      
      const order = await ctx.db.orders.findById(id);
      
      // Return null for not found — GraphQL spec allows this
      if (!order) return null;
      
      // Authorization: can this user see this order?
      if (order.customerId !== ctx.user.id && !ctx.user.isAdmin) {
        throw new ForbiddenError('Not authorized to view this order');
      }
      
      return order;
    },
    
    orders: async (_parent, args, ctx) => {
      if (!ctx.user) throw new AuthenticationError('Must be logged in');
      
      const { first = 20, after, filter, orderBy } = args;
      
      // Relay cursor pagination
      const [items, totalCount] = await Promise.all([
        ctx.db.orders.findMany({
          where: buildWhereClause(filter, ctx.user),
          orderBy: buildOrderBy(orderBy),
          cursor: after ? decodeCursor(after) : undefined,
          take: first + 1, // Request one extra to check hasNextPage
        }),
        ctx.db.orders.count({ where: buildWhereClause(filter, ctx.user) }),
      ]);
      
      const hasNextPage = items.length > first;
      const edges = items.slice(0, first).map(item => ({
        node: item,
        cursor: encodeCursor(item),
      }));
      
      return {
        edges,
        totalCount,
        pageInfo: {
          hasNextPage,
          hasPreviousPage: !!after,
          startCursor: edges[0]?.cursor ?? null,
          endCursor: edges[edges.length - 1]?.cursor ?? null,
        },
      };
    },
  },
  
  // Object type resolver — handles nested fields
  Order: {
    // customer will be called for every Order in a list — use DataLoader!
    customer: async (order, _args, ctx) => {
      return ctx.loaders.user.load(order.customerId);
    },
    
    lineItems: async (order, _args, ctx) => {
      return ctx.loaders.lineItemsByOrderId.load(order.id);
    },
  },
  
  Mutation: {
    createOrder: async (_parent, { input }, ctx) => {
      if (!ctx.user) throw new AuthenticationError('Must be logged in');
      
      try {
        const order = await ctx.db.orders.create({
          data: {
            customerId: ctx.user.id,
            items: input.items,
            shippingAddressId: input.shippingAddressId,
          },
        });
        
        return { order, userErrors: [] };
      } catch (error) {
        // Return domain errors as userErrors, not thrown errors
        if (error instanceof ValidationError) {
          return {
            order: null,
            userErrors: error.fieldErrors.map(fe => ({
              message: fe.message,
              field: fe.path,
              code: fe.code,
            })),
          };
        }
        // System errors: throw (will be masked to generic message in production)
        throw error;
      }
    },
  },
};

DataLoader — Fixing N+1 Queries

N+1 is the #1 GraphQL performance problem. Every nested field resolver runs once per parent item, causing one DB query per item in a list.

// Without DataLoader: 1 query for list + N queries for each customer = N+1
// With DataLoader: 1 query for list + 1 batched query for all customers = 2

function createLoaders(db: DatabaseClient) {
  return {
    // Batch load users by ID
    user: new DataLoader<string, User | null>(async (userIds) => {
      const users = await db.users.findMany({
        where: { id: { in: [...userIds] } },
      });
      const userMap = new Map(users.map(u => [u.id, u]));
      // Must return array in same order as input keys
      return userIds.map(id => userMap.get(id) ?? null);
    }),
    
    // Batch load line items by order ID (one-to-many)
    lineItemsByOrderId: new DataLoader<string, LineItem[]>(async (orderIds) => {
      const items = await db.lineItems.findMany({
        where: { orderId: { in: [...orderIds] } },
      });
      // Group by order ID
      const grouped = groupBy(items, item => item.orderId);
      return orderIds.map(id => grouped.get(id) ?? []);
    }),
  };
}

// Create loaders once per request (NOT per server start — must be per-request)
app.use('/graphql', (req, res, next) => {
  req.loaders = createLoaders(db);
  next();
});

---

Security

Depth and Complexity Limits

GraphQL allows deeply nested queries that can cause expensive joins:

# Malicious query — exponential explosion
{ user { orders { customer { orders { customer { orders { ... } } } } } } }

import depthLimit from 'graphql-depth-limit';
import { createComplexityLimitRule } from 'graphql-validation-complexity';

const server = new ApolloServer({
  validationRules: [
    depthLimit(8),                      // Max query depth
    createComplexityLimitRule(1000, {   // Max query complexity score
      onCost: (cost) => logger.info({ cost }, 'Query complexity'),
    }),
  ],
});

Persisted Queries (Defense Against Arbitrary Queries in Production)

// Only allow pre-approved queries in production
// Client sends hash, server looks up the full query
const server = new ApolloServer({
  persistedQueries: {
    cache: new KeyValueCache(), // Redis in production
    ttl: 900, // 15 minutes
  },
});

Field-Level Authorization

// In resolvers, check permissions per field
const resolvers = {
  User: {
    // Only expose email to the user themselves or admins
    email: (user, _args, ctx) => {
      if (ctx.user?.id === user.id || ctx.user?.isAdmin) {
        return user.email;
      }
      return null; // Or throw ForbiddenError
    },
    
    // Salary only visible to HR
    salary: (user, _args, ctx) => {
      if (!ctx.user?.roles.includes('hr')) {
        throw new ForbiddenError('Not authorized');
      }
      return user.salary;
    },
  },
};

---

Performance Patterns

Request Caching with DataLoader + Redis

// Cache individual entity results in Redis across requests
const userLoader = new DataLoader<string, User>(
  async (ids) => {
    // Check Redis first
    const cacheKeys = ids.map(id => `user:${id}`);
    const cached = await redis.mget(...cacheKeys);
    
    const missingIds: string[] = [];
    const results: Array<User | null> = cached.map((value, i) => {
      if (value !== null) return JSON.parse(value);
      missingIds.push(ids[i]);
      return null;
    });
    
    if (missingIds.length > 0) {
      const dbUsers = await db.users.findMany({ where: { id: { in: missingIds } } });
      const dbMap = new Map(dbUsers.map(u => [u.id, u]));
      
      // Fill in the gaps and cache
      const pipeline = redis.pipeline();
      for (let i = 0; i < ids.length; i++) {
        if (results[i] === null) {
          const user = dbMap.get(ids[i]) ?? null;
          results[i] = user;
          if (user) pipeline.setex(`user:${ids[i]}`, 300, JSON.stringify(user));
        }
      }
      await pipeline.exec();
    }
    
    return results;
  },
  { cache: false } // DataLoader in-request cache is separate from Redis
);

---

GraphQL vs REST Decision Guide

| Scenario | Prefer | |----------|--------| | Public API, many third-party consumers | REST (simpler to document, familiar) | | Internal API, multiple frontends (web, mobile, desktop) with different data needs | GraphQL | | Simple CRUD, few consumers | REST | | Complex, interrelated domain model | GraphQL | | Real-time subscriptions needed | GraphQL (subscriptions built-in) | | File uploads | REST (GraphQL file uploads are awkward) | | API gateway / proxy | REST or gRPC | | BFF (Backend for Frontend) | GraphQL |

Deeper Reference

For production GraphQL schema templates and resolver implementation patterns, see:

• references/schema-patterns.md — relay-compatible schema patterns, Union/Interface design, input validation, and schema-first development workflows
• references/resolver-patterns.md — DataLoader implementations, N+1 elimination strategies, subscription resolver patterns, and authorization middleware for resolvers