Choosing a database: DynamoDB, caches and warehouses

Why more than one kind of database?

A relational database is a great default: tables, joins, transactions and SQL. But some jobs fit other shapes better. A shopping cart read millions of times a minute, a leaderboard that changes every second, or a report over three years of sales each strain a single relational database in different ways. Cloud providers offer purpose-built databases for each shape.

DynamoDB: the key ideas

Amazon DynamoDB is a fully managed, serverless key-value and document database. There are no servers or versions to manage, and reads and writes take single-digit milliseconds at almost any scale. In exchange, you design the table around how you will query it.

• Item: one record, like a row, holding attributes. An item can be up to 400 KB.
• Partition key: the attribute DynamoDB hashes to decide where an item is stored. Every query must give one exact partition key value.
• Sort key (optional): orders items that share a partition key, and lets you ask for ranges, such as orders between two dates or keys that begin with "2026-03".
• Primary key: the partition key, or partition key plus sort key. It must be unique; writing the same key again replaces the item.
• Query reads one partition key efficiently. Scan reads the whole table and is slow and costly on big tables. Add a Global Secondary Index (GSI) when you need another access pattern.

With this design, "all orders for C-1001 in March 2026, newest first" is a single fast Query. Putting the order ID after the date in the sort key keeps keys unique when a customer orders twice on the same day.

Capacity modes and consistency

On-demand mode charges per request and scales by itself, which suits new or spiky apps. Provisioned mode lets you set read and write capacity units (optionally with auto scaling) and is cheaper for steady, predictable traffic. One write capacity unit (WCU) is one write per second of up to 1 KB. One read capacity unit (RCU) is one strongly consistent read per second of up to 4 KB, or two eventually consistent reads. Eventually consistent reads (the default) may return data a moment old; strongly consistent reads always return the latest write.

Global tables replicate a table to several regions, and every region can accept writes. DynamoDB also offers TTL (automatic deletion of expired items), Streams (a change feed for triggers) and point-in-time recovery.

Caches and warehouses

Amazon ElastiCache runs Valkey, Redis OSS or Memcached in memory. The common pattern is cache-aside: the app checks the cache first; on a miss it reads the database and stores the result with a TTL (time to live). This cuts database load and response times for data that many users read, like a product page. Amazon Redshift is a columnar data warehouse: it stores each column together so queries like "total sales by state for three years" scan only the columns they need.

Equivalents on other clouds

• Azure: Azure SQL Database and Azure Database for PostgreSQL/MySQL (relational), Cosmos DB (key-value and document), Azure Managed Redis (cache), Azure Synapse Analytics and Microsoft Fabric (warehouse).
• Google Cloud: Cloud SQL, AlloyDB and Spanner (relational), Firestore and Bigtable (NoSQL), Memorystore (cache), BigQuery (warehouse).