DP-900 · 3 Describe considerations for working with non-relational data on Azure
Cosmos DB use cases
Exam objective: Identify use cases for Azure Cosmos DB
Azure Cosmos DB is a fully managed, schema agnostic NoSQL database built for globally distributed apps that need low latency reads and writes.
Azure Cosmos DB is a fully managed, platform as a service NoSQL database. Microsoft runs the servers, patching and backups, so the application team can focus on the application itself. Its defining trait is that it is schema agnostic: items stored in the same container do not need identical properties, which suits data whose shape changes over time or differs between records.
Cosmos DB organizes data as an account that holds databases, which hold containers, which hold items. The partition key, throughput and indexing policy are all set at the container level. Add a region to the account, and Cosmos DB replicates the data there automatically, so an application can read and write close to its users anywhere in the world, typically within a few milliseconds.
Pick Cosmos DB when an application needs that combination of a flexible item shape, global reach and steady low latency. Typical examples are IoT telemetry ingestion, gaming leaderboards, retail catalogs and personalized apps. An application built around complex joins across many tables, or large scale historical analytics, fits a relational database or an analytics service better.
On the exam, look for globally distributed, schema agnostic or low latency next to a NoSQL use case.
Key points
- Cosmos DB is schema agnostic: items stored in the same container do not need to share the same set of properties.
- Data is organized as an account holding databases, which hold containers, which hold items, and the container is where you set the partition key, throughput and indexing policy.
- Adding a region to a Cosmos DB account replicates data there automatically, so an application reads and writes from the nearest region.
- Five consistency levels let you trade data freshness for availability, from Strong to Eventual, with Session as the most widely used level.
- Good fits include IoT telemetry, gaming leaderboards, retail catalogs and personalized web or mobile apps, while heavy multi table joins or large scale historical analytics fit other services better.
Exam trap
Schema agnostic does not mean undisciplined. A partition key is still required, and a poorly chosen one, with too few distinct values, limits scalability even though no fixed schema was ever defined.
Check yourself
A mobile game needs a leaderboard that player apps around the world can read and update with single digit millisecond latency. Which Azure service fits this need best?
Go deeper on Microsoft Learn
Checked against Microsoft Learn on October 1, 2026.