Databases

First PublishedFeb 17, 2026ByAtif Alam

Azure provides fully managed databases — relational and NoSQL — that handle patching, backups, high availability, and scaling.

Azure SQL

Azure SQL is the managed version of Microsoft SQL Server. It comes in three deployment options:

Option	What It Is	Best For
Azure SQL Database	Single database (serverless or provisioned)	New cloud-native apps
Elastic Pool	Multiple databases sharing resources	SaaS apps (one DB per tenant)
SQL Managed Instance	Near-complete SQL Server compatibility	Lift-and-shift of on-prem SQL Server

Creating an Azure SQL Database

# Create a server (logical container for databases)
az sql server create \
  --resource-group myapp-rg \
  --name my-sql-server \
  --admin-user sqladmin \
  --admin-password 'S3cure!Pass123'

# Create a database
az sql db create \
  --resource-group myapp-rg \
  --server my-sql-server \
  --name myappdb \
  --edition GeneralPurpose \
  --compute-model Serverless \
  --auto-pause-delay 60 \
  --min-capacity 0.5

# Allow Azure services to connect
az sql server firewall-rule create \
  --resource-group myapp-rg \
  --server my-sql-server \
  --name AllowAzure \
  --start-ip-address 0.0.0.0 \
  --end-ip-address 0.0.0.0

Service Tiers

Tier	Compute	Best For
General Purpose	Balanced (standard SSD)	Most workloads
Business Critical	High IOPS (local SSD, built-in read replica)	Low-latency, mission-critical
Hyperscale	Scale up to 100 TB, near-instant backups, up to 4 read replicas	Very large databases

Serverless Compute

Azure SQL’s serverless tier auto-scales and can pause when idle:

No activity ──► Auto-pause (60 min default) ──► $0 compute cost
                                                  │
Traffic arrives ──► Auto-resume (~1 min) ──► Scale up

You pay for storage when paused, but not compute. Great for dev/test and variable workloads.

High Availability

Tier	HA Method	Failover
General Purpose	Remote storage replicas	Managed failover (~30s)
Business Critical	Always On (local replicas, built-in read)	Automatic (~30s)
Hyperscale	Multiple compute replicas	Automatic, read replicas always available

Geo-Replication

# Create a readable geo-replica in another region
az sql db replica create \
  --resource-group myapp-rg \
  --server my-sql-server \
  --name myappdb \
  --partner-server my-sql-server-eu \
  --partner-resource-group myapp-eu-rg

Use auto-failover groups for automatic DNS-based failover between regions.

Security

Feature	How to Enable
Private endpoint	Access via private IP in your VNet
Encryption at rest	Transparent Data Encryption (TDE) — on by default
Encryption in transit	TLS — enforced by default
Azure AD authentication	Login with Entra ID instead of SQL credentials
Dynamic data masking	Mask sensitive columns from non-privileged users
Auditing	Log all database operations to Storage or Log Analytics
Advanced Threat Protection	Detect SQL injection, anomalous access

Cosmos DB

Azure Cosmos DB is a globally distributed, multi-model NoSQL database. It’s Azure’s answer to DynamoDB — but more flexible, supporting multiple APIs and data models.

APIs (Data Models)

API	Data Model	When to Use
NoSQL (native)	JSON documents	New apps, most flexible
MongoDB	MongoDB-compatible documents	Migrating from MongoDB
Cassandra	Wide-column	Migrating from Cassandra
Gremlin	Graph	Social networks, knowledge graphs
Table	Key-value	Migrating from Azure Table Storage
PostgreSQL	Relational (distributed)	Distributed PostgreSQL (Citus)

Core Concepts

Cosmos DB Account
  └── Database: myapp
       └── Container: orders        (like a table/collection)
            ├── Item: {id: "1", customer: "alice", amount: 50}
            ├── Item: {id: "2", customer: "bob", amount: 30}
            └── Partition key: /customer

Concept	What It Is
Account	Top-level resource. Choose API, regions, and consistency.
Database	Logical grouping of containers.
Container	Where data lives (like a table). Has a partition key and throughput.
Item	A single document/row (up to 2 MB).
Partition key	Determines data distribution. Choose a high-cardinality key.

Creating a Cosmos DB Container

# Create account
az cosmosdb create \
  --name mycosmosacct \
  --resource-group myapp-rg \
  --locations regionName=eastus failoverPriority=0 \
  --locations regionName=westeurope failoverPriority=1

# Create database
az cosmosdb sql database create \
  --account-name mycosmosacct \
  --resource-group myapp-rg \
  --name myapp

# Create container
az cosmosdb sql container create \
  --account-name mycosmosacct \
  --resource-group myapp-rg \
  --database-name myapp \
  --name orders \
  --partition-key-path /customer \
  --throughput 400

Throughput Models

Model	How It Works	Best For
Provisioned	Set RU/s (Request Units per second)	Predictable traffic
Autoscale	Scales between min and max RU/s	Variable traffic
Serverless	Pay per request (no provisioned RU/s)	Dev/test, low/intermittent traffic

Request Units (RU): A single point read (1 KB item by ID + partition key) costs 1 RU. Writes cost ~5-10 RU. Complex queries cost more.

Global Distribution

Cosmos DB’s killer feature — multi-region, active-active replication with single-digit ms latency:

Write in East US ──► replicate ──► West Europe (read replica)
                                ──► Southeast Asia (read replica)

Enable multi-region writes for active-active (writes in any region).

Consistency Levels

Level	Guarantee	Latency	Use Case
Strong	Linearizable (latest write always)	Highest	Financial transactions
Bounded staleness	Reads lag by at most N versions or T time	High	Leaderboards, metrics
Session (default)	Read-your-own-writes within a session	Moderate	Most applications
Consistent prefix	Reads are in order but may lag	Low	Chat, activity feeds
Eventual	No ordering guarantee	Lowest	Counters, non-critical

Python SDK Example

from azure.cosmos import CosmosClient

client = CosmosClient("https://mycosmosacct.documents.azure.com", credential)
database = client.get_database_client("myapp")
container = database.get_container_client("orders")

# Create
container.upsert_item({
    "id": "order-001",
    "customer": "alice",
    "amount": 49.99,
    "status": "shipped"
})

# Read (point read — 1 RU)
item = container.read_item("order-001", partition_key="alice")

# Query
results = container.query_items(
    "SELECT * FROM orders o WHERE o.customer = @customer AND o.amount > @min",
    parameters=[
        {"name": "@customer", "value": "alice"},
        {"name": "@min", "value": 20}
    ],
    partition_key="alice"
)

Azure Database for PostgreSQL / MySQL

Managed PostgreSQL and MySQL — equivalent to AWS RDS.

Deployment Options

Option	What It Is	Best For
Flexible Server (recommended)	Zone-redundant HA, burstable compute, stop/start	Most workloads
Single Server (legacy)	Basic managed PostgreSQL/MySQL	Existing deployments

Creating a PostgreSQL Flexible Server

az postgres flexible-server create \
  --resource-group myapp-rg \
  --name my-postgres \
  --admin-user pgadmin \
  --admin-password 'S3cure!Pass123' \
  --sku-name Standard_D2s_v3 \
  --tier GeneralPurpose \
  --storage-size 128 \
  --version 16 \
  --high-availability ZoneRedundant

Key Features

Feature	Detail
HA	Zone-redundant (standby in another AZ, automatic failover)
Read replicas	Up to 10 replicas (async, cross-region supported)
Backups	Automatic daily, 7–35 day retention, point-in-time restore
Stop/Start	Pause dev servers to save costs
VNet integration	Private access via VNet (no public IP)
Entra ID auth	Authenticate with managed identities

Choosing a Database

Scenario	Choose
SQL Server migration / .NET ecosystem	Azure SQL
PostgreSQL or MySQL workload	Azure Database for PostgreSQL/MySQL
Global distribution, multi-model NoSQL	Cosmos DB
Key-value at massive scale, simple access	Cosmos DB (NoSQL API)
Document database (MongoDB compatible)	Cosmos DB (MongoDB API)
Graph queries	Cosmos DB (Gremlin API)
In-memory caching	Azure Cache for Redis
Data warehouse / analytics	Azure Synapse Analytics

Key Takeaways

• Azure SQL is managed SQL Server. Use the serverless tier for variable workloads (auto-pause saves costs). Business Critical tier for low-latency production.
• Cosmos DB is globally distributed NoSQL with multiple APIs. Choose the right consistency level and partition key. Pay per RU.
• Azure Database for PostgreSQL/MySQL (Flexible Server) is the equivalent of AWS RDS — zone-redundant HA, read replicas, point-in-time restore.
• Use Private Endpoints to access databases over private IPs. Use Entra ID authentication where supported.
• Always enable encryption at rest, automated backups, and restrict access via NSGs or Private Endpoints.