/architecture-design

Architecture Design - Service Selection Reference

Compute Services

Decision Tree

Need full OS control?
├─ YES → Virtual Machines or VM Scale Sets
└─ NO  → Is it containerized?
          ├─ YES → Need Kubernetes orchestration?
          │        ├─ YES → Azure Kubernetes Service (AKS)
          │        └─ NO  → Container Apps
          └─ NO  → Is it event-driven/short-lived?
                   ├─ YES → Azure Functions or Logic Apps
                   └─ NO  → Azure App Service

**Recommendation Priority:** PaaS > Containers > IaaS

Service Comparison

Data Storage Services

Relational Databases

NoSQL Databases

Files & Blobs

Messaging & Integration

Networking Services

Monitoring Services

Service Selection Guidelines

PaaS vs IaaS Decision Matrix

Database Selection Decision Tree

What type of data?

RELATIONAL (tables, SQL, ACID):
├─ Need SQL Server features? → Azure SQL Database/Managed Instance
├─ Want open-source? → PostgreSQL or MySQL
└─ Simple queries, budget-conscious? → Azure SQL (Basic/Standard tier)

NOSQL:
├─ Global distribution needed? → Cosmos DB
├─ Simple key-value? → Table Storage
├─ In-memory cache? → Redis Cache
└─ Document store with flexibility? → Cosmos DB (MongoDB API)

FILES/OBJECTS:
├─ Unstructured data (images, videos)? → Blob Storage
├─ File shares (SMB)? → Azure Files
└─ Big data analytics? → Data Lake Gen2

Messaging Service Selection

What's your messaging pattern?

COMMANDS (one receiver):
├─ Simple, cost-effective? → Storage Queues
└─ Enterprise features (FIFO, transactions)? → Service Bus Queues

EVENTS (multiple receivers):
├─ High-throughput streaming? → Event Hubs
├─ Reactive, serverless? → Event Grid
└─ Point-to-point with reliability? → Service Bus Topics

WORKFLOWS:
├─ Visual designer, many connectors? → Logic Apps
└─ Code-first, complex patterns? → Durable Functions

Common Combinations

Web Application Stack

Azure Front Door (CDN + WAF)
└─ App Service (Web + API)
   ├─ Redis Cache (Session/Data)
   ├─ Azure SQL (Primary data)
   └─ Blob Storage (Media files)

Microservices Stack

API Management (Gateway)
├─ Container Apps / AKS (Services)
├─ Service Bus (Inter-service messaging)
├─ Cosmos DB (Service A data)
└─ Azure SQL (Service B data)

Data Platform Stack

Data Factory (ETL orchestration)
├─ Data Lake Gen2 (Raw data)
├─ Databricks / Synapse (Processing)
└─ Power BI (Visualization)

IoT Platform Stack

IoT Hub (Device management)
└─ Event Hubs (Telemetry ingestion)
   ├─ Stream Analytics (Real-time)
   ├─ Functions (Processing)
   └─ Data Lake (Cold path)

Regional Considerations

Multi-Region Patterns

Active-Active:

• Both regions handle traffic simultaneously
• Highest availability (99.999%+)
• Highest cost (duplicate infrastructure)
• Use: Mission-critical, global apps

Active-Passive:

• Primary region active, secondary on standby
• Good availability (99.99%+)
• Moderate cost (standby infrastructure)
• Use: Business-critical apps with DR requirements

Single Region + Availability Zones:

• Multiple zones in one region
• Good availability (99.99%)
• Lowest cost
• Use: Most production workloads

Paired Regions

Always design multi-region using Azure paired regions for automatic geo-replication benefits:

Cost Optimization Patterns

Right-Sizing Strategy

Reserved Instances Savings

Auto-Scaling Configuration

Production App Service Example:
- Min instances: 2 (HA requirement)
- Max instances: 10 (cost control)
- Scale out: CPU > 70% for 5 minutes
- Scale in: CPU < 30% for 10 minutes
- Cool-down: 5 minutes (scale out), 10 minutes (scale in)

Result: Balanced between availability and cost

Security Defaults

Every architecture must include:

Managed Identities - No credentials in code Private Endpoints - No public internet exposure for PaaS HTTPS Only - TLS 1.2+ enforced NSGs - Network security groups with least privilege Key Vault - All secrets, certificates, keys Azure AD - Authentication and RBAC Encryption - At rest (default) and in transit

High Availability Defaults

Every production architecture must include:

Availability Zones - Minimum for production Multiple Instances - At least 2 for stateless services Health Checks - Configured on all services Auto-Scaling - Defined rules based on metrics Backup Strategy - RPO/RTO defined and tested Monitoring - Application Insights + Log Analytics Alerts - Critical scenarios with proper routing