Purpose-Built Databases
AWS offers many databases because one engine does not fit every access pattern. Pick by data shape and how you read it, not by habit.
| Type | Service | Typical use |
|---|---|---|
| Relational (OLTP) | RDS, Aurora | Orders, accounts, ERP, SQL joins, strict schema |
| Key-value / document | DynamoDB | Shopping carts, sessions, IoT state, high scale APIs |
| In-memory | ElastiCache, MemoryDB | Caching, leaderboards, real-time counters |
| Document (MongoDB-compatible) | DocumentDB | Content catalogs, user profiles, MongoDB apps |
| Wide-column (Cassandra-compatible) | Keyspaces | Serverless Cassandra for time-ordered and high-write workloads |
| Graph | Neptune | Social graphs, fraud rings, recommendations, knowledge graphs |
| Time series | Timestream | IoT and application metrics over time |
| Search and analytics on logs | OpenSearch Service | Full-text search, log analytics, dashboards |
| Ledger | Amazon QLDB (check current status) or Aurora/DynamoDB with audit design | Immutable, verifiable history |
| Data warehouse (OLAP) | Redshift | Complex analytical SQL over huge data |
| Lake queries | Athena on S3 | Ad-hoc SQL on files |
Amazon DocumentDB
- MongoDB-compatible managed document database with a Aurora-like storage layer (6 copies across 3 AZs, up to 15 replicas)
- Use the MongoDB drivers with a few differences. Not every MongoDB feature is supported
- Choose it to lift MongoDB apps with little change. Choose DynamoDB for serverless key access at huge scale
Amazon Keyspaces
- Serverless, Apache Cassandra-compatible (CQL). Scales automatically, no clusters
- Use when you already use Cassandra or need wide-column modeling and CQL drivers
Amazon Neptune
A graph database for highly connected data. It supports Gremlin and openCypher (property graphs) and SPARQL (RDF). Relationship queries that need many joins in SQL become simple traversals.
- Use for fraud detection, recommendations, social networks, network topology, identity graphs
- Neptune Analytics supports graph analytics and vector search for GenAI use
Amazon Timestream
- Serverless time series database with automatic tiering between memory and magnetic storage and built-in time functions
- Use for IoT sensor data, DevOps metrics, clickstream over time
- Query with SQL (
ago(1h),bin(time, 5m)). Check the current availability and recommended engine options (Timestream for LiveAnalytics and for InfluxDB) before choosing
Amazon OpenSearch Service
- Managed OpenSearch (and legacy Elasticsearch) clusters or serverless collections
- Use for full-text search, autocomplete, log and security analytics, vector search for AI, dashboards
- Not a primary database. Keep the source of truth in DynamoDB, RDS or S3 and index a copy, for example with DynamoDB Streams to OpenSearch or Firehose
- Common pattern: Kinesis Firehose to OpenSearch for logs
Amazon Redshift
- Columnar, massively parallel data warehouse for analytics over terabytes to petabytes
- Serverless or provisioned. Redshift Spectrum queries S3 data. Zero-ETL integrations load from Aurora and DynamoDB
- Not for high-rate single-row OLTP. Use it for dashboards and reports
Decision Guide
| Question | Answer |
|---|---|
| Need joins, foreign keys, ACID SQL? | Aurora / RDS |
| Known key lookups, serverless, scale without limit? | DynamoDB |
| Sub-millisecond reads of hot data? | ElastiCache / DAX |
| Existing MongoDB app? | DocumentDB |
| Existing Cassandra app? | Keyspaces |
| Queries follow relationships (friends of friends)? | Neptune |
| Data points with timestamps, trends and downsampling? | Timestream |
| Search box, relevance ranking, log search? | OpenSearch |
| BI over huge historical data? | Redshift or Athena + S3 |
| Ad-hoc SQL on files, pay per query? | Athena |
Polyglot Persistence Example
Practice Tasks
- Pick a database for five scenarios and defend the choice in one sentence each
- Index a DynamoDB table into OpenSearch with a Stream-triggered Lambda
- Run a Gremlin or openCypher query on a small Neptune graph
- Store sensor readings in a time series database and query hourly averages
- Query a CSV in S3 with Athena
Conclusion
Choose the database by data shape and access pattern: Aurora or RDS for relational, DynamoDB for key access at scale, ElastiCache for speed, DocumentDB for MongoDB, Keyspaces for Cassandra, Neptune for graphs, Timestream for time series, OpenSearch for search and Redshift or Athena for analytics. Combine several when needed and check service status before adopting niche ones.

