Azure Cosmos DB

Azure Cosmos DB is a globally distributed, multi-model database with guaranteed single-digit millisecond latency at the 99th percentile. It supports document (NoSQL), key-value, graph, and column-family data models with five tunable consistency levels.

Core Concepts

• Account — top-level resource, defines global regions and consistency
• Database — a namespace for containers
• Container — equivalent to a collection/table, holds items
• Partition Key — determines data distribution; critical for performance
• Request Unit (RU) — normalized cost of database operations
• Consistency Level — Strong, Bounded Staleness, Session, Consistent Prefix, Eventual

Account and Database Setup

# Create a Cosmos DB account with global replication
az cosmosdb create \
  --name my-app-cosmos \
  --resource-group my-app-rg \
  --kind GlobalDocumentDB \
  --default-consistency-level Session \
  --locations regionName=eastus failoverPriority=0 \
  --locations regionName=westeurope failoverPriority=1 \
  --enable-automatic-failover true

# Create a database with shared throughput
az cosmosdb sql database create \
  --account-name my-app-cosmos \
  --resource-group my-app-rg \
  --name app-db \
  --throughput 400

# Create a container with partition key and autoscale
az cosmosdb sql container create \
  --account-name my-app-cosmos \
  --resource-group my-app-rg \
  --database-name app-db \
  --name orders \
  --partition-key-path /customerId \
  --max-throughput 4000 \
  --idx '{"indexingMode":"consistent","automatic":true,"includedPaths":[{"path":"/*"}],"excludedPaths":[{"path":"/payload/*"}]}'

CRUD Operations

# Initialize client and perform CRUD
from azure.cosmos import CosmosClient, PartitionKey

client = CosmosClient(
    url="https://my-app-cosmos.documents.azure.com:443/",
    credential="your-key-here"
)
database = client.get_database_client("app-db")
container = database.get_container_client("orders")

# Create an item
order = {
    "id": "order-001",
    "customerId": "customer-123",
    "items": [
        {"name": "Widget", "qty": 2, "price": 29.99},
        {"name": "Gadget", "qty": 1, "price": 49.99}
    ],
    "total": 109.97,
    "status": "pending",
    "createdAt": "2024-01-15T10:30:00Z"
}
container.create_item(body=order)

# Read an item (requires partition key)
item = container.read_item(item="order-001", partition_key="customer-123")
print(f"Order: {item['status']}, Total: ${item['total']}")

# Replace (full update)
item['status'] = 'shipped'
item['shippedAt'] = '2024-01-16T14:00:00Z'
container.replace_item(item=item['id'], body=item)

# Partial update with patch operations
container.patch_item(
    item="order-001",
    partition_key="customer-123",
    patch_operations=[
        {"op": "set", "path": "/status", "value": "delivered"},
        {"op": "add", "path": "/deliveredAt", "value": "2024-01-17T09:00:00Z"},
        {"op": "incr", "path": "/updateCount", "value": 1}
    ]
)

# Delete an item
container.delete_item(item="order-001", partition_key="customer-123")

Querying

# SQL queries on Cosmos DB
# Query orders for a customer
orders = container.query_items(
    query="SELECT * FROM c WHERE c.customerId = @customerId AND c.status = @status",
    parameters=[
        {"name": "@customerId", "value": "customer-123"},
        {"name": "@status", "value": "pending"}
    ],
    partition_key="customer-123"
)
for order in orders:
    print(f"{order['id']}: ${order['total']}")

# Cross-partition query (more expensive, use sparingly)
all_pending = container.query_items(
    query="SELECT c.id, c.customerId, c.total FROM c WHERE c.status = 'pending' ORDER BY c.total DESC",
    enable_cross_partition_query=True,
    max_item_count=50
)

# Aggregation query
result = container.query_items(
    query="SELECT VALUE COUNT(1) FROM c WHERE c.status = 'shipped'",
    enable_cross_partition_query=True
)
count = list(result)[0]

Consistency Levels

# Update default consistency level
az cosmosdb update \
  --name my-app-cosmos \
  --resource-group my-app-rg \
  --default-consistency-level BoundedStaleness \
  --max-staleness-prefix 100 \
  --max-interval 5

Level	Guarantee	RU Cost	Use Case
Strong	Linearizable reads	Highest	Financial transactions
Bounded Staleness	Reads lag by ≤K versions or T time	High	Leaderboards, counters
Session	Read-your-writes per session	Medium	Default — most apps
Consistent Prefix	Reads never see out-of-order writes	Low	Social feeds
Eventual	No ordering guarantee	Lowest	Non-critical analytics

Change Feed

# Process change feed for event-driven architecture
from azure.cosmos import CosmosClient

container = CosmosClient(url, credential).get_database_client("app-db").get_container_client("orders")

# Read changes from beginning
change_feed = container.query_items_change_feed(
    is_start_from_beginning=True,
    partition_key_range_id="0"
)
for change in change_feed:
    print(f"Changed item: {change['id']}, status: {change.get('status')}")

Global Distribution

# Add a read region
az cosmosdb update \
  --name my-app-cosmos \
  --resource-group my-app-rg \
  --locations regionName=eastus failoverPriority=0 \
  --locations regionName=westeurope failoverPriority=1 \
  --locations regionName=southeastasia failoverPriority=2

# Enable multi-region writes
az cosmosdb update \
  --name my-app-cosmos \
  --resource-group my-app-rg \
  --enable-multiple-write-locations true

Throughput Management

# Enable autoscale on a container
az cosmosdb sql container throughput migrate \
  --account-name my-app-cosmos \
  --resource-group my-app-rg \
  --database-name app-db \
  --name orders \
  --throughput-type autoscale

# Check current throughput and usage
az cosmosdb sql container throughput show \
  --account-name my-app-cosmos \
  --resource-group my-app-rg \
  --database-name app-db \
  --name orders

Best Practices

• Choose partition key carefully — it determines scalability and query performance
• Use Session consistency for most applications (best balance of performance and guarantees)
• Use autoscale throughput for variable workloads to avoid over-provisioning
• Query within a single partition whenever possible to minimize RU consumption
• Use the change feed for event-driven patterns instead of polling
• Enable automatic failover for production accounts
• Exclude large payload paths from indexing to save RUs on writes
• Use point reads (by id + partition key) instead of queries when possible — 1 RU