Azure Cosmos DB: Global NoSQL

Azure Cosmos DB is Microsoft's globally distributed, multi-model NoSQL database service. It is designed to deliver single-digit millisecond read and write latency at any scale, anywhere in the world. Unlike traditional databases constrained to a single region, Cosmos DB can transparently replicate data to dozens of Azure regions simultaneously.

Cosmos DB supports multiple database APIs through a single service. You can use the native NoSQL API (JSON documents), the MongoDB API for apps built on MongoDB drivers, the Apache Cassandra API for wide-column workloads, the Gremlin API for graph data, and the Table API for key-value data. Each API exposes Cosmos DB's underlying engine through a familiar interface.

The Cosmos DB resource hierarchy consists of three levels. An account is the top-level resource that defines the replication region list and consistency level. Inside an account, you create databases as logical namespaces. Inside each database, you create containers — the unit that holds items (documents) and where throughput is provisioned.

A partition key is a property on every item that Cosmos DB uses to distribute data across physical partitions. Choosing the right partition key is critical: it should have high cardinality (many distinct values), spread write traffic evenly, and appear in the majority of your queries. Poor partition key choices lead to hot partitions that throttle throughput and degrade performance.

Cosmos DB charges for throughput in Request Units per second (RU/s). One RU is roughly the cost of reading a 1 KB document by its ID. Writes, queries, and deletes cost more RUs depending on item size and index complexity. You can provision RU/s at the container level or share a pool across all containers in a database, and switch to autoscale mode to let Cosmos DB scale throughput automatically.

Cosmos DB's most powerful feature is its ability to replicate data globally with a few clicks. You add regions to your account, and Cosmos DB synchronises all writes across every region automatically. Each region can serve both reads and writes simultaneously in a multi-write (multi-master) configuration, minimising latency for geographically dispersed users.

Cosmos DB offers five tunable consistency levels as a spectrum from strongest to weakest: Strong (linearizable reads, highest latency), Bounded Staleness (reads lag behind writes by a bounded window), Session (default — consistent reads for your own writes within a session), Consistent Prefix (reads never see out-of-order writes), and Eventual (fastest reads, weakest guarantees). Choose based on your application's tolerance for stale data.

By default, Cosmos DB automatically indexes every property in every item you insert, without requiring you to define a schema upfront. This enables flexible ad-hoc queries without index maintenance. You can customise the indexing policy to exclude paths that are never queried (reducing storage cost) or add composite indexes for ORDER BY queries across multiple properties.

The change feed is a sorted log of all create and update operations on a Cosmos DB container. Applications subscribe to the change feed to trigger downstream processing — for example, updating a search index, sending notifications, or writing to a materialised view. Change feed is the foundation for event-driven and CQRS patterns in Cosmos DB architectures.

Cosmos DB supports Time-to-Live (TTL), a property you set on a container or individual items to automatically delete them after a specified number of seconds. This is ideal for session data, cache entries, audit logs with regulatory retention windows, or any data with a known expiration. TTL deletions are performed as background tasks and do not consume provisioned RU/s.

Cosmos DB offers two billing modes. Provisioned throughput bills you per RU/s-hour regardless of actual usage, suited to steady predictable workloads. Serverless mode bills per RU consumed and GB stored, with no minimum cost, making it cost-effective for development, prototyping, and traffic spikes. Serverless Cosmos DB is currently limited to a single region.

Test your understanding of Microsoft Azure Fundamentals (AZ-900) concepts from this lesson.

In this lesson you learned: Cosmos DB is a globally distributed, multi-model NoSQL service, partition keys determine how data is sharded and must be chosen carefully, and five consistency levels let you balance latency and data freshness. Next up we explore Azure's managed services for open-source database engines.