dynamodb-single-table（dynamodb 单表）

DynamoDB Single-Table Design Pipeline

This skill orchestrates a 3-step pipeline for designing and implementing a DynamoDB single-table design. Use it when the user wants the full workflow or needs guidance on which step to take next.

Pipeline Overview

  Step 1                    Step 2                    Step 3
  ┌──────────────────┐      ┌──────────────────┐      ┌──────────────────┐
  │ Access Patterns   │ ──>  │ Table Design      │ ──>  │ Query Interfaces  │
  │                   │      │                   │      │                   │
  │ Elicit entities   │      │ Design PK/SK/GSI  │      │ Generate method   │
  │ & access patterns │      │ Map AP -> Query   │      │ signatures & types│
  │ Output: .md file  │      │ Output: .md file  │      │ Output: .md file  │
  └──────────────────┘      └──────────────────┘      └──────────────────┘
  Skill:                    Skill:                    Skill:
  dynamodb-access-patterns  dynamodb-table-design     dynamodb-query-interfaces

Step 0: Detect Pipeline State

Before routing, check if the user's project already has pipeline artifacts. Search for files matching these patterns:

• **/access-patterns.md or **/access_patterns.md
• **/table-design.md or **/table_design.md
• **/query-interfaces.md or **/query_interfaces.md

Also check common locations:

• docs/dynamodb/
• docs/
• Project root

Based on what exists:

Tell the user what you found:

I found existing DynamoDB design artifacts in your project:

  [x] Access Patterns: docs/dynamodb/access-patterns.md
  [x] Table Design:    docs/dynamodb/table-design.md
  [ ] Query Interfaces: (not found)

You're ready for Step 3. Should I generate the query interfaces from
your table design?

How to Route

Core Concepts (Quick Reference)

What is Single-Table Design?

All entity types live in ONE DynamoDB table. You pre-join related data into item collections (items sharing a partition key) so they can be fetched in a single Query call. This eliminates the need for joins and reduces network round-trips.

When to Use Single-Table Design

• You need sub-30ms response times at any scale
• You have well-defined access patterns that won't change frequently
• You need to fetch multiple entity types in a single request
• You're building OLTP workloads (not analytics)

When NOT to Use Single-Table Design

• Your access patterns are still evolving rapidly (early-stage startup)
• You use GraphQL with per-type resolvers (the resolver model negates single-query benefits)
• You need flexible ad-hoc queries (consider a relational database instead)

Key Principles

1. Access patterns first. Never design keys without knowing your queries.
2. Pre-join into item collections. Related data shares a partition key.
3. One table, generic key names. Use pk/sk (not userId/orderId).
4. GSIs for alternate access patterns. Reuse before creating new ones (max 20).
5. Composite sort keys for multi-field sorting. Encode hierarchy in the SK.
6. No Scans. Every query must use a partition key. If you need Scan, redesign.
7. Transactions for multi-item writes. Ensure consistency across item collections.

Artifacts Produced

By the end of the pipeline, the user will have 3 files:

These files serve as the specification for the implementation phase.