Skip to content

Azure Cosmos DB

Overview

Azure Cosmos DB is Microsoft’s globally distributed, multi-model, NoSQL database service.

It’s designed for high availability, low latency, and horizontal scalability — across multiple regions if needed.

You can think of it as a serverless database platform built for modern cloud-native and distributed applications.

Key Features

  • Multi-model: Supports several data models:

  • Document (Core SQL API, MongoDB API)

  • Key-value (Table API)
  • Graph (Gremlin API)
  • Column-family (Cassandra API)
  • Global distribution: Read/write replicas in any Azure region
  • Elastic scaling of throughput (RU/s) and storage
  • Automatic indexing without schema or index management
  • Millisecond latency guarantees for reads and writes
  • 99.999% availability SLA (multi-region)

Advantages

Benefit Description
Performance at scale Single-digit millisecond read/write latency worldwide
Global availability Geo-redundancy and multi-master write capability
Flexible schema Schemaless — great for agile development
Multiple APIs Familiar data access with SQL, MongoDB, Cassandra, Gremlin, Table
Integrated with Azure Secure with RBAC, Private Link, managed identities, diagnostics
Change Feed support Event stream of data changes for real-time processing

Drawbacks / Considerations

Limitation Description
Cost Cosmos DB can be expensive, especially with high RU/s provisioning
RU-based pricing model Can be unintuitive — need to plan throughput carefully
No full SQL capabilities Not a relational DB — no joins across collections or transactions
Consistency trade-offs You must choose between strong and eventual consistency models

What is RU/s?

  • Request Unit (RU) is the currency of throughput.
  • Each operation (read, write, query) consumes RUs.
  • You provision RU/s based on expected usage, or use serverless mode for pay-per-use.

Conceptual Use Case

Use Cosmos DB when:

  • You need globally available, highly responsive data access
  • Your data is semi-structured or unstructured
  • You want to avoid managing indexes, partitions, or replication
  • You are building IoT, real-time analytics, personalization, or e-commerce apps
  • You need to serve a globally distributed user base

Example: Using Cosmos DB with SQL API in C

using Microsoft.Azure.Cosmos;

CosmosClient client = new CosmosClient("<connection-string>");
Database database = await client.CreateDatabaseIfNotExistsAsync("MyDatabase");
Container container = await database.CreateContainerIfNotExistsAsync("Products", "/category");

var product = new { id = "1", name = "Mug", category = "Kitchen" };
await container.CreateItemAsync(product, new PartitionKey("Kitchen"));

Global Distribution Example

Imagine a travel app with users in the UK, US, and Japan. With Cosmos DB, you can:

  • Replicate data to Azure regions West Europe, East US, and Japan East
  • Enable multi-region writes to keep all users fast and in sync
  • Use conflict resolution policies if concurrent updates occur

Consistency Models

Azure Cosmos DB offers 5 consistency levels:

Level Description
Strong Linearizability (most strict)
Bounded Staleness Lag allowed (e.g. 100 writes, 5 seconds)
Session Per-client consistency (default)
Consistent Prefix Reads never see out-of-order writes
Eventual Lowest latency, least consistency

You choose the trade-off between performance and data freshness.

When Not to Use Cosmos DB

Scenario Better Option
Complex relational data with joins Azure SQL Database
Large-scale data warehousing Azure Synapse Analytics, Data Lake
Small workloads needing low cost Azure Table Storage, SQLite, or Lite DB
Strict ACID transactional behaviour SQL Server or PostgreSQL

Pricing

  • You pay for:
    • RU/s throughput
    • Data storage (GB)
    • Geo-replication (per region)
    • Backup & restore
  • Serverless mode is available for bursty or low-volume apps

Use Azure Cosmos DB Pricing Calculator to plan costs.

Further Learning

Summary Table

Feature Cosmos DB Offers
Data Model Document, key-value, column, graph
API Support SQL, MongoDB, Cassandra, Gremlin, Table
Global Scale Yes, with multi-region replication
Consistency Levels 5, from strong to eventual
Indexing Automatic and customizable
Serverless Option Yes
Ideal For Low-latency apps, IoT, mobile, analytics