Cosmos DB

Azure Cosmos DB is a globally distributed, multi-model NoSQL database with guaranteed single-digit millisecond latency, automatic scaling, and five consistency levels.

Simple Explanation

Cosmos DB is a database that lives everywhere at once. You store data in one place and it instantly replicates to datacenters worldwide — users in Tokyo and New York both get <10ms reads from a local copy.

When to Use Cosmos DB

• Globally distributed apps where users are geographically spread (e-commerce, gaming, social).
• Applications requiring guaranteed single-digit millisecond latency at any scale.
• Schema-flexible data where document structure evolves over time.
• Systems that need to process millions of requests per second.

APIs (Data Models)

Consistency Levels

Cosmos DB lets you choose the trade-off between consistency and latency/availability. Stronger = slower but more accurate; weaker = faster but may read slightly stale data.

Partitioning

Cosmos DB scales horizontally by distributing data across partitions. The Partition Key determines how data is split. Choosing the wrong partition key is the #1 cause of performance problems.

Request Units (RUs)

Cosmos DB billing is based on Request Units — a normalized measure of CPU, memory, and IOPS. A 1KB point read costs 1 RU. Writes cost more (~5 RU). Configure Throughput (RU/s) manually or use Autoscale.

Commands

# Create a Cosmos DB account (NoSQL API)
az cosmosdb create \
  --resource-group rg-database \
  --name mycosmosdb-account \
  --kind GlobalDocumentDB \
  --locations regionName=eastus failoverPriority=0 isZoneRedundant=False \
  --default-consistency-level Session

# Add a second read region (global distribution)
az cosmosdb update \
  --resource-group rg-database \
  --name mycosmosdb-account \
  --locations regionName=eastus failoverPriority=0 isZoneRedundant=False \
                regionName=westeurope failoverPriority=1 isZoneRedundant=False

# Create a database
az cosmosdb sql database create \
  --account-name mycosmosdb-account \
  --resource-group rg-database \
  --name AppDB

# Create a container with partition key
az cosmosdb sql container create \
  --account-name mycosmosdb-account \
  --resource-group rg-database \
  --database-name AppDB \
  --name UserProfiles \
  --partition-key-path "/userId" \
  --throughput 400

# Enable autoscale (instead of fixed throughput)
az cosmosdb sql container throughput update \
  --account-name mycosmosdb-account \
  --resource-group rg-database \
  --database-name AppDB \
  --name UserProfiles \
  --max-throughput 4000

Hands-on

1. Create a Cosmos DB account with NoSQL API in a single region.
2. Create a database and container with /userId as partition key.
3. Use Azure Data Explorer (in Portal) to insert and query JSON documents.
4. Add a second region and observe replication in the Replicate data globally panel.
5. Try changing the consistency level to Eventual and discuss the trade-offs.

Debugging Scenario

Issue: Queries are performing 100x worse than expected.

• Enable Diagnostics in Portal → check QueryMetrics in query response headers.
• A cross-partition query (no partition key in filter) scans all partitions — extremely expensive on large containers. Add partition key to all WHERE clauses.
• Check if RU/s is throttled (429 Too Many Requests) — increase throughput or enable autoscale.
• Review partition distribution: a hot partition causes latency spikes even with sufficient total RU/s.

Interview Questions

Summary

Cosmos DB is built for global-scale, high-throughput NoSQL workloads. The multi-API support simplifies migration from MongoDB, Cassandra, and others. The main decisions are: API choice, partition key (most critical for performance), and consistency level (Session is the right default).