AWS Databases

AWS offers managed databases across all major categories: relational (RDS, Aurora), NoSQL (DynamoDB), in-memory (ElastiCache), data warehouse (Redshift), document (DocumentDB), graph (Neptune), ledger (QLDB), and time-series (Timestream).

Service Map

Service	Type	Engine	Use Case
RDS	Relational	MySQL, PostgreSQL, MariaDB, Oracle, SQL Server	General OLTP, web apps
Aurora	Relational (MySQL/PG compatible)	Aurora MySQL, Aurora PostgreSQL	High-scale, HA, serverless
DynamoDB	NoSQL (key-value, document)	DynamoDB	High-scale, low-latency
ElastiCache	In-memory	Redis, Memcached	Caching, sessions, pub/sub
Redshift	Data warehouse	Redshift (PostgreSQL-based)	Analytics, BI
DocumentDB	Document	MongoDB compatible	Semi-structured data
Neptune	Graph	Gremlin, SPARQL, openCypher	Social, fraud, knowledge graphs
QLDB	Ledger	Amazon Quantum Ledger Database	Audit trail, immutable
Timestream	Time-series	Timestream	IoT, metrics, events

Database Selection Decision Tree

What type of data?
  │
  ├── Structured relational (ACID transactions)?
  │     │
  │     ├── Need high availability + auto-scaling + serverless?
  │     │     YES → Aurora (MySQL or PostgreSQL)
  │     │     NO ↓
  │     │
  │     ├── Need managed Oracle/SQL Server?
  │     │     YES → RDS (Oracle, SQL Server)
  │     │     NO → RDS (MySQL, PostgreSQL, MariaDB)
  │     │
  │     └── Need petabyte-scale data warehouse?
  │           YES → Redshift
  │           NO ↓
  │
  ├── Semi-structured (JSON documents)?
  │     YES → DocumentDB (MongoDB compatible)
  │     NO ↓
  │
  ├── Graph (relationships, social, fraud)?
  │     YES → Neptune (Gremlin or SPARQL)
  │     NO ↓
  │
  ├── Key-value (high-scale, low-latency)?
  │     YES → DynamoDB
  │     NO ↓
  │
  ├── In-memory (caching, sessions)?
  │     YES → ElastiCache (Redis or Memcached)
  │     NO ↓
  │
  ├── Immutable audit log (financial, compliance)?
  │     YES → QLDB
  │     NO ↓
  │
  └── Time-series (IoT, metrics)?
        YES → Timestream
        NO → Consider what you're actually storing

Architecture Patterns

Read Replicas

Writer (Primary)
  │
  └──► Read Replica 1 (async, read-only)
  └──► Read Replica 2 (async, read-only)
  └──► Read Replica 3 (async, read-only)

Multi-AZ (HA)

AZ-1: Primary DB
  │
  └──► AZ-2: Standby (sync replication)
  └──► AZ-3: Read Replica 1

Cache Layer

Application
  │
  ▼
ElastiCache (Redis/Memcached)
  │ Cache miss
  ▼
RDS / DynamoDB

Aurora Serverless

Application
  │
  ▼
Aurora Serverless (auto-scales ACU)
  │
  └── Data API (HTTPS, no persistent connections)

Shared Concepts

Encryption at Rest

All AWS managed databases support encryption at rest using KMS:

• AWS managed keys (free)
• Customer managed keys (CMK) — you pay for KMS

Automated Backups

Database	Default Retention	Max Retention
RDS MySQL/PG	1 day	35 days
Aurora	1 day (continuous)	35 days
DynamoDB	Incremental forever	Infinite (PITR)
ElastiCache Redis	1 day (RDB)	35 days
Redshift	1 day	35 days

Maintenance Windows

All databases have a weekly maintenance window (30 minutes):

• Engine version upgrades
• OS patches
• Instance class changes

Set maintenance-window to off-peak hours.

References

• Homepage: https://aws.amazon.com/products/databases/
• Documentation: https://docs.aws.amazon.com/databases/
• Pricing: https://aws.amazon.com/products/databases/pricing/

Nuggets & Gotchas

• RDS and Aurora are NOT multi-region by default — you must create read replicas in another region for DR: A Multi-AZ deployment keeps a standby in the same region only. For true DR across regions, use cross-region read replicas or Aurora Global Database.
• DynamoDB has no schema — you can store anything, but you should enforce a schema at the application layer: DynamoDB will happily store {"name": "x"} in one item and {"count": 42} in another. Your application must handle schema variations.
• ElastiCache Redis has two modes — Cluster Mode Disabled (single shard, up to 5 replicas) and Cluster Mode Enabled (up to 90 shards): Cluster Mode Enabled allows horizontal scaling but requires Redis cluster clients. Most use cases work fine with Cluster Mode Disabled.
• Redshift is not meant for OLTP — it’s a columnar MPP database for analytics, not transactions: If you need sub-second queries on millions of rows with high write throughput, use RDS or DynamoDB. Redshift is for complex analytical queries on large datasets (BI, reporting).
• QLDB is immutable but not encrypted at rest by default with customer managed keys — it’s always encrypted with AWS managed keys: For financial compliance, verify your encryption requirements match QLDB’s default behavior. Use customer managed keys if you need extra control.