dynamodb-single-table（dynamodb 单表）

DynamoDB Single Table Design

This skill covers single table design patterns for DynamoDB, the AWS-recommended approach for modeling complex relationships in a single table.

Core Philosophy

Single table design principles:

• One table to rule them all: Model all entities in a single table
• Overload keys: Use generic pk/sk for flexibility
• Optimize for access patterns: Design around how you query data
• Denormalize when needed: Duplicate data to avoid joins
• Use GSIs strategically: Add secondary indexes for alternate access patterns

Why Single Table Design?

Benefits:

• ✅ Consistent performance across all queries
• ✅ Lower costs (fewer tables to provision)
• ✅ Atomic transactions across entity types
• ✅ Simpler infrastructure management
• ✅ Better suited for serverless architectures

Trade-offs:

• ❌ More complex to design initially
• ❌ Requires understanding access patterns upfront
• ❌ Less intuitive than relational databases

Key Concepts

Generic Key Names

Use generic partition (pk) and sort (sk) keys instead of entity-specific names:

// ✅ Good - Generic and flexible
{
  pk: "USER#123",
  sk: "PROFILE",
  // ... entity data
}

// ❌ Bad - Entity-specific
{
  userId: "123",
  // ... entity data
}

Composite Keys

Build keys from multiple attributes:

// User entity
pk: "USER#userId"
sk: "PROFILE"

// User's posts
pk: "USER#userId"
sk: "POST#postId"

// Post details
pk: "POST#postId"
sk: "METADATA"

// Post comments
pk: "POST#postId"
sk: "COMMENT#commentId"

Item Collections

Group related items under the same partition key:

// All items with pk="USER#123" form an "item collection"
// Can retrieve in a single query

// User profile
{ pk: "USER#123", sk: "PROFILE", name: "John", email: "..." }

// User's posts
{ pk: "USER#123", sk: "POST#post1", title: "...", content: "..." }
{ pk: "USER#123", sk: "POST#post2", title: "...", content: "..." }

// User's subscriptions
{ pk: "USER#123", sk: "SUB#sub1", plan: "pro", ... }

Access Pattern Design

Pattern 1: Get Single Item

// Get user profile
{
  pk: "USER#123",
  sk: "PROFILE"
}

// DynamoDB operation
await client.send(new GetCommand({
  TableName: Resource.Database.name,
  Key: {
    pk: "USER#123",
    sk: "PROFILE"
  }
}));

Pattern 2: Query Item Collection

// Get all posts by user
await client.send(new QueryCommand({
  TableName: Resource.Database.name,
  KeyConditionExpression: "pk = :pk AND begins_with(sk, :sk)",
  ExpressionAttributeValues: {
    ":pk": "USER#123",
    ":sk": "POST#"
  }
}));

Pattern 3: Query with Sort

// Get user's recent posts (sorted by timestamp)
{
  pk: "USER#123",
  sk: "POST#2025-01-02T10:30:00Z#post1"  // ISO timestamp for sorting
}

await client.send(new QueryCommand({
  TableName: Resource.Database.name,
  KeyConditionExpression: "pk = :pk AND begins_with(sk, :sk)",
  ExpressionAttributeValues: {
    ":pk": "USER#123",
    ":sk": "POST#"
  },
  ScanIndexForward: false  // Descending order
}));

Pattern 4: Global Secondary Index (GSI)

// GSI for querying posts by status across all users
// GSI: gsi1pk = "POST#STATUS#published", gsi1sk = timestamp

// In SST config
const table = new sst.aws.Dynamo("Database", {
  fields: {
    pk: "string",
    sk: "string",
    gsi1pk: "string",
    gsi1sk: "string"
  },
  primaryIndex: { hashKey: "pk", rangeKey: "sk" },
  globalIndexes: {
    gsi1: { hashKey: "gsi1pk", rangeKey: "gsi1sk" }
  }
});

// Query all published posts
await client.send(new QueryCommand({
  TableName: Resource.Database.name,
  IndexName: "gsi1",
  KeyConditionExpression: "gsi1pk = :pk",
  ExpressionAttributeValues: {
    ":pk": "POST#STATUS#published"
  }
}));

Common Patterns

Pattern: User with Posts and Comments

// User profile
{
  pk: "USER#userId",
  sk: "PROFILE",
  name: "John",
  email: "john@example.com",
  createdAt: "2025-01-01T00:00:00Z"
}

// User's post
{
  pk: "USER#userId",
  sk: "POST#postId",
  title: "My Post",
  content: "...",
  createdAt: "2025-01-02T00:00:00Z"
}

// Post metadata (for reverse lookup)
{
  pk: "POST#postId",
  sk: "METADATA",
  userId: "userId",
  title: "My Post",
  content: "...",
  commentCount: 5
}

// Post comments
{
  pk: "POST#postId",
  sk: "COMMENT#2025-01-02T10:00:00Z#commentId",
  userId: "commenterId",
  text: "Great post!",
  createdAt: "2025-01-02T10:00:00Z"
}

// Commenter profile (denormalized for display)
{
  pk: "POST#postId",
  sk: "COMMENT#2025-01-02T10:00:00Z#commentId",
  userId: "commenterId",
  userName: "Jane",  // Denormalized!
  userAvatar: "https://...",  // Denormalized!
  text: "Great post!"
}

Access Patterns:

1. Get user profile: GetItem(pk="USER#userId", sk="PROFILE")
2. Get user's posts: Query(pk="USER#userId", sk begins_with "POST#")
3. Get post with comments: Query(pk="POST#postId")
4. Get recent comments: Sort by timestamp in sk

Pattern: Many-to-Many (Users and Groups)

// User membership in group
{
  pk: "USER#userId",
  sk: "GROUP#groupId",
  groupName: "Developers",  // Denormalized
  role: "admin",
  joinedAt: "2025-01-01"
}

// Group membership list
{
  pk: "GROUP#groupId",
  sk: "USER#userId",
  userName: "John",  // Denormalized
  role: "admin",
  joinedAt: "2025-01-01"
}

// Group metadata
{
  pk: "GROUP#groupId",
  sk: "METADATA",
  name: "Developers",
  description: "...",
  memberCount: 42
}

Access Patterns:

1. Get user's groups: Query(pk="USER#userId", sk begins_with "GROUP#")
2. Get group's members: Query(pk="GROUP#groupId", sk begins_with "USER#")
3. Check membership: GetItem(pk="USER#userId", sk="GROUP#groupId")

Pattern: Hierarchical Data (Folders and Files)

// Folder
{
  pk: "FOLDER#folderId",
  sk: "METADATA",
  name: "Documents",
  parentId: "parentFolderId",
  path: "/Documents"
}

// Files in folder
{
  pk: "FOLDER#folderId",
  sk: "FILE#2025-01-02#fileId",  // Sorted by date
  name: "report.pdf",
  size: 1024000,
  uploadedAt: "2025-01-02T10:00:00Z"
}

// File metadata (for direct access)
{
  pk: "FILE#fileId",
  sk: "METADATA",
  name: "report.pdf",
  folderId: "folderId",
  size: 1024000
}

Pattern: Time Series Data

// Metrics by date
{
  pk: "METRICS#resourceId",
  sk: "2025-01-02T10:00:00Z",
  cpu: 45.2,
  memory: 67.8,
  requests: 1234
}

// Query metrics for a time range
await client.send(new QueryCommand({
  TableName: Resource.Database.name,
  KeyConditionExpression: "pk = :pk AND sk BETWEEN :start AND :end",
  ExpressionAttributeValues: {
    ":pk": "METRICS#resourceId",
    ":start": "2025-01-01T00:00:00Z",
    ":end": "2025-01-02T00:00:00Z"
  }
}));

Implementation with SST

Basic Setup

// sst.config.ts
const table = new sst.aws.Dynamo("Database", {
  fields: {
    pk: "string",
    sk: "string",
    gsi1pk: "string",
    gsi1sk: "string",
    gsi2pk: "string",
    gsi2sk: "string"
  },
  primaryIndex: { hashKey: "pk", rangeKey: "sk" },
  globalIndexes: {
    gsi1: { hashKey: "gsi1pk", rangeKey: "gsi1sk" },
    gsi2: { hashKey: "gsi2pk", rangeKey: "gsi2sk" }
  },
  stream: "new-and-old-images"  // For event-driven updates
});

Type-Safe Helpers

// src/lib/db.ts
import { DynamoDBClient } from "@aws-sdk/client-dynamodb";
import { DynamoDBDocumentClient } from "@aws-sdk/lib-dynamodb";
import { Resource } from "sst";

export const dynamodb = DynamoDBDocumentClient.from(new DynamoDBClient({}), {
  marshallOptions: {
    removeUndefinedValues: true
  }
});

export const TableName = Resource.Database.name;

// Key builders
export const keys = {
  user: (userId: string) => ({
    profile: { pk: `USER#${userId}`, sk: "PROFILE" },
    post: (postId: string) => ({ pk: `USER#${userId}`, sk: `POST#${postId}` })
  }),
  post: (postId: string) => ({
    metadata: { pk: `POST#${postId}`, sk: "METADATA" },
    comment: (commentId: string, timestamp: string) => ({
      pk: `POST#${postId}`,
      sk: `COMMENT#${timestamp}#${commentId}`
    })
  })
};

CRUD Operations

import { GetCommand, PutCommand, UpdateCommand, DeleteCommand, QueryCommand } from "@aws-sdk/lib-dynamodb";
import { dynamodb, TableName, keys } from "./db";

// Create user
export async function createUser(userId: string, data: UserData) {
  await dynamodb.send(new PutCommand({
    TableName,
    Item: {
      ...keys.user(userId).profile,
      ...data,
      createdAt: new Date().toISOString()
    }
  }));
}

// Get user
export async function getUser(userId: string) {
  const result = await dynamodb.send(new GetCommand({
    TableName,
    Key: keys.user(userId).profile
  }));
  return result.Item as User | undefined;
}

// Update user
export async function updateUser(userId: string, updates: Partial<UserData>) {
  await dynamodb.send(new UpdateCommand({
    TableName,
    Key: keys.user(userId).profile,
    UpdateExpression: "SET #name = :name, #email = :email",
    ExpressionAttributeNames: {
      "#name": "name",
      "#email": "email"
    },
    ExpressionAttributeValues: {
      ":name": updates.name,
      ":email": updates.email
    }
  }));
}

// Get user's posts
export async function getUserPosts(userId: string) {
  const result = await dynamodb.send(new QueryCommand({
    TableName,
    KeyConditionExpression: "pk = :pk AND begins_with(sk, :sk)",
    ExpressionAttributeValues: {
      ":pk": `USER#${userId}`,
      ":sk": "POST#"
    }
  }));
  return result.Items as Post[];
}

Best Practices

1. Design for Access Patterns First

❌ Don't design entities first:

Users table, Posts table, Comments table...

✅ Do design access patterns first:

1. Get user profile
2. Get user's posts
3. Get post with comments
4. Get all published posts
// Then design keys to support these

2. Use Sparse Indexes

Only items with GSI keys appear in the index:

// Only published posts have gsi1pk
{
  pk: "POST#123",
  sk: "METADATA",
  status: "published",
  gsi1pk: "POST#STATUS#published",  // Only published posts have this
  gsi1sk: "2025-01-02T10:00:00Z"
}

3. Denormalize Strategically

Duplicate data to avoid secondary queries:

// Comment with user info denormalized
{
  pk: "POST#postId",
  sk: "COMMENT#commentId",
  userId: "userId",
  userName: "John",  // From users table
  userAvatar: "...",  // From users table
  text: "Great post!"
}

4. Use Transactions for Related Items

import { TransactWriteCommand } from "@aws-sdk/lib-dynamodb";

await dynamodb.send(new TransactWriteCommand({
  TransactItems: [
    {
      Put: {
        TableName,
        Item: { pk: "POST#123", sk: "METADATA", ... }
      }
    },
    {
      Update: {
        TableName,
        Key: { pk: "USER#userId", sk: "PROFILE" },
        UpdateExpression: "SET postCount = postCount + :inc",
        ExpressionAttributeValues: { ":inc": 1 }
      }
    }
  ]
}));

5. Handle Hot Partitions

Distribute writes using suffixes:

// Instead of: pk: "METRICS"
// Use: pk: "METRICS#0", "METRICS#1", ..., "METRICS#9"
const suffix = Math.floor(Math.random() * 10);
const pk = `METRICS#${suffix}`;

Common Gotchas

1. Sort Key is Required for Queries

// ❌ This won't work
Query(pk = "USER#123")

// ✅ Use begins_with
Query(pk = "USER#123" AND sk begins_with "POST#")

2. GSI Consistency is Eventually Consistent

// After writing to main table
await putItem({ pk: "USER#123", gsi1pk: "ACTIVE" });

// GSI query might not see it immediately
const result = await query({ IndexName: "gsi1", gsi1pk: "ACTIVE" });
// May not include the item yet!

3. Item Size Limit is 400KB

// Don't store large data in items
❌ { pk: "POST#123", content: "<10MB of text>" }

// Store large data in S3
✅ { pk: "POST#123", contentUrl: "s3://bucket/key" }

4. Projection of GSI Matters

// GSI with ALL projection (expensive)
globalIndexes: {
  gsi1: {
    hashKey: "gsi1pk",
    projection: "all"  // Copies all attributes
  }
}

// GSI with KEYS_ONLY (cheaper)
globalIndexes: {
  gsi1: {
    hashKey: "gsi1pk",
    projection: "keys_only"  // Only pk, sk, gsi keys
  }
}

Testing Single Table Design

// Use local DynamoDB for tests
import { DynamoDBClient } from "@aws-sdk/client-dynamodb";

const client = new DynamoDBClient({
  endpoint: "http://localhost:8000"
});

describe("User operations", () => {
  test("creates user and retrieves profile", async () => {
    await createUser("123", { name: "John", email: "john@example.com" });
    const user = await getUser("123");
    expect(user?.name).toBe("John");
  });

  test("queries user posts", async () => {
    await createPost("123", "post1", { title: "First Post" });
    await createPost("123", "post2", { title: "Second Post" });

    const posts = await getUserPosts("123");
    expect(posts).toHaveLength(2);
  });
});

Migration Strategy

If migrating from multiple tables:

1. Identify access patterns in existing code
2. Design new key structure to support patterns
3. Create migration scripts to transform data
4. Run in parallel (dual writes during transition)
5. Verify data integrity before cutover
6. Switch to single table atomically

Further Reading

• AWS DynamoDB Best Practices: https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/best-practices.html
• Alex DeBrie's "DynamoDB Book": https://www.dynamodbbook.com/
• Rick Houlihan's re:Invent talks on YouTube
• SST DynamoDB docs: https://sst.dev/docs/component/aws/dynamo