Cosmos DB Architecture

Develop for Azure Storage

Shahzad Mian

Content developer, DataCamp

Azure Cosmos DB

The problem

What is Cosmos DB?

 

• Multi-model, globally distributed NoSQL database.
• NoSQL - flexible, folder-like storage.
• Supports multiple APIs and data types.
• Like a flexible restaurant menu - choose your API flavor:
  • SQL (Core) for JSON documents.
  • MongoDB for Mongo developers.
  • Cassandra for wide-column data.
  • Gremlin for graph relationships.
  • Table API for key-value storage.

Cosmos DB for developers

• Built with developers in mind.
• Automatic indexing - no schema management.
• Fast queries.
• Five consistency levels: from Strong to Eventual.
• Trade-off between performance vs consistency.
  • Strong - everyone hears the same message instantly.
  • Eventual - everyone catches up later.

 

Scaling and Throughput (RU/s)

• Request units per second (RU/s) = performance currency of Cosmos DB.
• Each read, write, or query spends some RU/s.
• Choose fixed provisioning or autoscale.

Scaling and Throughput (RU/s)

• Request units per second (RU/s) = performance currency of Cosmos DB.
• Each read, write, or query spends some RU/s.
• Choose fixed provisioning or autoscale.
• Autoscale adjusts automatically with demand.

Elastic scale and Partitioning

• Cosmos DB scales through partitioning.
• Each container uses a partition key.
• The key groups related data together.
• Example: Cipher Coffee uses /customerId for orders.
• All a customer's orders live in the same partition when using /customerId as the partition key.
• Queries stay fast even as data grows.
• Hot partition: Uneven partition load slowing down performance.

Global distribution

• Replicate data to any Azure region easily.
• Just tick checkboxes, no setup hassle.
• If one region fails, traffic reroutes automatically.
• No downtime, no data loss.
• Like a backup cafe ready to serve customers instantly.

Monitoring and cost optimization

• Track performance in the Azure Portal.
• View metrics like RU/s, latency, and request charge.
• Like a car dashboard - see speed and fuel use.
• Cipher Coffee engineers analyzed these metrics.
• Reduced costs by adjusting throughput dynamically.

Let's practice!

Develop for Azure Storage