references/architecture_patterns.md
Azure Architecture Patterns
Reference guide for selecting the right Azure architecture pattern based on application requirements.
Table of Contents
- Pattern Selection Matrix
- Pattern 1: App Service Web Application
- Pattern 2: Microservices on AKS
- Pattern 3: Serverless Event-Driven
- Pattern 4: Data Pipeline
- Pattern 5: Multi-Region Active-Active
- Well-Architected Framework Alignment
Pattern Selection Matrix
| Pattern | Best For | Users | Monthly Cost | Complexity | |---------|----------|-------|--------------|------------| | App Service Web | MVPs, SaaS, APIs | <100K | $50-500 | Low | | Microservices on AKS | Complex platforms, multi-team | Any | $500-5000 | High | | Serverless Event-Driven | Event processing, webhooks, APIs | <1M | $20-500 | Low-Medium | | Data Pipeline | Analytics, ETL, ML | Any | $200-3000 | Medium-High | | Multi-Region Active-Active | Global apps, 99.99% uptime | >100K | 1.5-2x single | High |
Pattern 1: App Service Web Application
Architecture
┌──────────────┐
│ Azure Front │
│ Door │
│ (CDN + WAF) │
└──────┬───────┘
│
┌──────▼───────┐
│ App Service │
│ (Linux P1v3)│
│ + Slots │
└──┬───────┬───┘
│ │
┌────────▼──┐ ┌──▼────────┐
│ Azure SQL │ │ Blob │
│ Serverless │ │ Storage │
└────────────┘ └───────────┘
│
┌────────▼──────────┐
│ Key Vault │
│ (secrets, certs) │
└───────────────────┘
Services
| Service | Purpose | Configuration | |---------|---------|---------------| | Azure Front Door | Global CDN, WAF, SSL | Standard or Premium tier, custom domain | | App Service | Web application hosting | Linux P1v3 (production), B1 (dev) | | Azure SQL Database | Relational database | Serverless GP_S_Gen5_2 with auto-pause | | Blob Storage | Static assets, uploads | Hot tier with lifecycle policies | | Key Vault | Secrets management | RBAC authorization, soft-delete enabled | | Application Insights | Monitoring and APM | Workspace-based, connected to Log Analytics | | Entra ID | Authentication | Easy Auth or MSAL library |
Deployment Strategy
- Deployment slots: staging slot for zero-downtime deploys, swap to production after validation
- Auto-scale: CPU-based rules, 1-10 instances in production
- Health checks:
/healthendpoint monitored by App Service and Front Door
Cost Estimate
| Component | Dev | Production | |-----------|-----|-----------| | App Service | $13 (B1) | $75 (P1v3) | | Azure SQL | $5 (Basic) | $40-120 (Serverless GP) | | Front Door | $0 (disabled) | $35-55 | | Blob Storage | $1 | $5-15 | | Key Vault | $0.03 | $1-5 | | Application Insights | $0 (free tier) | $5-20 | | Total | ~$19 | ~$160-290 |
Pattern 2: Microservices on AKS
Architecture
┌──────────────┐
│ Azure Front │
│ Door │
└──────┬───────┘
│
┌──────▼───────┐
│ API Mgmt │
│ (gateway) │
└──────┬───────┘
│
┌────────────▼────────────┐
│ AKS Cluster │
│ ┌───────┐ ┌───────┐ │
│ │ svc-A │ │ svc-B │ │
│ └───┬───┘ └───┬───┘ │
│ │ │ │
│ ┌───▼─────────▼───┐ │
│ │ Service Bus │ │
│ │ (async msgs) │ │
│ └─────────────────┘ │
└─────────────────────────┘
│ │
┌────────▼──┐ ┌──▼────────┐
│ Cosmos DB │ │ ACR │
│ (data) │ │ (images) │
└────────────┘ └───────────┘
Services
| Service | Purpose | Configuration | |---------|---------|---------------| | AKS | Container orchestration | 3 node pools: system (D2s_v5), app (D4s_v5), jobs (spot) | | API Management | API gateway, rate limiting | Standard v2 or Consumption tier | | Cosmos DB | Multi-model database | Session consistency, autoscale RU/s | | Service Bus | Async messaging | Standard tier, topics for pub/sub | | Container Registry | Docker image storage | Basic (dev), Standard (prod) | | Key Vault | Secrets for pods | CSI driver + workload identity | | Azure Monitor | Cluster and app observability | Container Insights + App Insights |
AKS Best Practices
Node Pools:
- System pool: 2-3 nodes, D2s_v5, taints for system pods only
- App pool: 2-10 nodes (autoscaler), D4s_v5, for application workloads
- Jobs pool: spot instances, for batch processing and CI runners
Networking:
- Azure CNI for VNet-native pod networking
- Network policies (Azure or Calico) for pod-to-pod isolation
- Ingress via NGINX Ingress Controller or Application Gateway Ingress Controller (AGIC)
Security:
- Workload Identity for pod-to-Azure service auth (replaces pod identity)
- Azure Policy for Kubernetes (OPA Gatekeeper)
- Defender for Containers for runtime threat detection
- Private cluster for production (API server not exposed to internet)
Deployment:
- Helm charts for application packaging
- Flux or ArgoCD for GitOps
- Horizontal Pod Autoscaler (HPA) + KEDA for event-driven scaling
Cost Estimate
| Component | Dev | Production | |-----------|-----|-----------| | AKS nodes (system) | $60 (1x D2s_v5) | $180 (3x D2s_v5) | | AKS nodes (app) | $120 (1x D4s_v5) | $360 (3x D4s_v5) | | API Management | $0 (Consumption) | $175 (Standard v2) | | Cosmos DB | $25 (serverless) | $100-400 (autoscale) | | Service Bus | $10 | $10-50 | | Container Registry | $5 | $20 | | Monitoring | $0 | $50-100 | | Total | ~$220 | ~$900-1300 |
Pattern 3: Serverless Event-Driven
Architecture
┌──────────┐ ┌──────────┐ ┌──────────┐
│ HTTP │ │ Blob │ │ Timer │
│ Trigger │ │ Trigger │ │ Trigger │
└────┬─────┘ └────┬─────┘ └────┬─────┘
│ │ │
└────────┬───────┘─────────┬───────┘
│ │
┌──────▼───────┐ ┌──────▼───────┐
│ Azure │ │ Azure │
│ Functions │ │ Functions │
│ (handlers) │ │ (workers) │
└──┬────┬──────┘ └──────┬───────┘
│ │ │
┌─────────▼┐ ┌─▼──────────┐ ┌─▼──────────┐
│ Event │ │ Service │ │ Cosmos DB │
│ Grid │ │ Bus Queue │ │ (data) │
│ (fanout) │ │ (reliable) │ │ │
└──────────┘ └────────────┘ └────────────┘
Services
| Service | Purpose | Configuration | |---------|---------|---------------| | Azure Functions | Event handlers, APIs | Consumption plan (dev), Premium (prod) | | Event Grid | Event routing and fan-out | System + custom topics | | Service Bus | Reliable messaging with DLQ | Basic or Standard, queues + topics | | Cosmos DB | Low-latency data store | Serverless (dev), autoscale (prod) | | Blob Storage | File processing triggers | Lifecycle policies | | Application Insights | Function monitoring | Sampling at 5-10% for high volume |
Durable Functions Patterns
Use Durable Functions for orchestration instead of building custom state machines:
| Pattern | Use Case | Example | |---------|----------|---------| | Function chaining | Sequential steps | Order: validate -> charge -> fulfill -> notify | | Fan-out/fan-in | Parallel processing | Process all images in a batch, aggregate results | | Async HTTP APIs | Long-running operations | Start job, poll for status, return result | | Monitor | Periodic polling | Check external API until condition met | | Human interaction | Approval workflows | Send approval email, wait for response with timeout |
Cost Estimate
| Component | Dev | Production | |-----------|-----|-----------| | Functions (Consumption) | $0 (1M free) | $5-30 | | Event Grid | $0 | $0-5 | | Service Bus | $0 (Basic) | $10-30 | | Cosmos DB | $0 (serverless free tier) | $25-150 | | Blob Storage | $1 | $5-15 | | Application Insights | $0 | $5-15 | | Total | ~$1 | ~$50-245 |
Pattern 4: Data Pipeline
Architecture
┌──────────┐ ┌──────────┐
│ IoT/Apps │ │ Batch │
│ (events) │ │ (files) │
└────┬─────┘ └────┬─────┘
│ │
┌────▼─────┐ ┌────▼─────┐
│ Event │ │ Data │
│ Hubs │ │ Factory │
└────┬─────┘ └────┬─────┘
│ │
└────────┬───────┘
│
┌────────▼────────┐
│ Data Lake │
│ Storage Gen2 │
│ (raw/curated) │
└────────┬────────┘
│
┌────────▼────────┐
│ Synapse │
│ Analytics │
│ (SQL + Spark) │
└────────┬────────┘
│
┌────────▼────────┐
│ Power BI │
│ (dashboards) │
└─────────────────┘
Services
| Service | Purpose | Configuration | |---------|---------|---------------| | Event Hubs | Real-time event ingestion | Standard, 2-8 partitions | | Data Factory | Batch ETL orchestration | Managed, 90+ connectors | | Data Lake Storage Gen2 | Raw and curated data lake | HNS enabled, lifecycle policies | | Synapse Analytics | SQL and Spark analytics | Serverless SQL pool (pay-per-query) | | Azure Functions | Lightweight processing | Triggered by Event Hubs or Blob | | Power BI | Business intelligence | Pro ($10/user/month) |
Data Lake Organization
data-lake/
├── raw/ # Landing zone — immutable source data
│ ├── source-system-a/
│ │ └── YYYY/MM/DD/ # Date-partitioned
│ └── source-system-b/
├── curated/ # Cleaned, validated, business-ready
│ ├── dimension/
│ └── fact/
├── sandbox/ # Ad-hoc exploration
└── archive/ # Cold storage (lifecycle policy target)
Cost Estimate
| Component | Dev | Production | |-----------|-----|-----------| | Event Hubs (1 TU) | $22 | $44-176 | | Data Factory | $0 (free tier) | $50-200 | | Data Lake Storage | $5 | $20-80 | | Synapse Serverless SQL | $5 | $50-300 | | Azure Functions | $0 | $5-20 | | Power BI Pro | $10/user | $10/user | | Total | ~$42 | ~$180-800 |
Pattern 5: Multi-Region Active-Active
Architecture
┌──────────────┐
│ Azure Front │
│ Door (Global│
│ LB + WAF) │
└──┬────────┬──┘
│ │
┌──────────▼──┐ ┌──▼──────────┐
│ Region 1 │ │ Region 2 │
│ (East US) │ │ (West EU) │
│ │ │ │
│ App Service │ │ App Service │
│ + SQL │ │ + SQL │
│ + Redis │ │ + Redis │
└──────┬──────┘ └──────┬──────┘
│ │
┌──────▼───────────────▼──────┐
│ Cosmos DB │
│ (multi-region writes) │
│ Session consistency │
└─────────────────────────────┘
Multi-Region Design Decisions
| Decision | Recommendation | Rationale | |----------|---------------|-----------| | Global load balancer | Front Door Premium | Built-in WAF, CDN, health probes, fastest failover | | Database replication | Cosmos DB multi-write or SQL failover groups | Cosmos for global writes, SQL for relational needs | | Session state | Azure Cache for Redis (per region) | Local sessions, avoid cross-region latency | | Static content | Front Door CDN | Edge-cached, no origin required | | DNS strategy | Front Door handles routing | No separate Traffic Manager needed | | Failover | Automatic (Front Door health probes) | 10-30 second detection, automatic reroute |
Azure SQL Failover Groups vs Cosmos DB Multi-Region
| Feature | SQL Failover Groups | Cosmos DB Multi-Region | |---------|-------------------|----------------------| | Replication | Async (RPO ~5s) | Sync or async (configurable) | | Write region | Single primary | Multi-write capable | | Failover | Automatic or manual (60s grace) | Automatic | | Consistency | Strong (single writer) | 5 levels (session recommended) | | Cost | 2x compute (active-passive) | Per-region RU/s charge | | Best for | Relational data, transactions | Document data, global low-latency |
Cost Impact
Multi-region typically costs 1.5-2x single region:
- Compute: 2x (running in both regions)
- Database: 1.5-2x (replication, multi-write)
- Networking: Additional cross-region data transfer (~$0.02-0.05/GB)
- Front Door Premium: ~$100-200/month
Well-Architected Framework Alignment
Every architecture pattern should address all five pillars of the Azure Well-Architected Framework.
Reliability
- Deploy across Availability Zones (zone-redundant App Service, AKS, SQL)
- Enable health probes at every layer
- Implement retry policies with exponential backoff (Polly for .NET, tenacity for Python)
- Define RPO/RTO and test disaster recovery quarterly
- Use Azure Chaos Studio for fault injection testing
Security
- Entra ID for all human and service authentication
- Managed Identity for all Azure service-to-service communication
- Key Vault for secrets, certificates, and encryption keys — no secrets in code or config
- Private Endpoints for all PaaS services in production
- Microsoft Defender for Cloud for threat detection and compliance
Cost Optimization
- Use serverless and consumption-based services where possible
- Auto-pause Azure SQL in dev/test (serverless tier)
- Spot VMs for fault-tolerant AKS node pools
- Reserved Instances for steady-state production workloads (1-year = 35% savings)
- Azure Advisor cost recommendations — review weekly
- Set budgets and alerts at subscription and resource group level
Operational Excellence
- Bicep for all infrastructure (no manual portal deployments)
- GitOps for AKS (Flux or ArgoCD)
- Deployment slots or blue-green for zero-downtime deploys
- Centralized logging in Log Analytics with standardized KQL queries
- Azure DevOps or GitHub Actions for CI/CD with workload identity federation
Performance Efficiency
- Application Insights for distributed tracing and performance profiling
- Azure Cache for Redis for session state and hot-path caching
- Front Door for edge caching and global acceleration
- Autoscale rules on compute (CPU, memory, HTTP queue length)
- Load testing with Azure Load Testing before production launch