references/service_selection.md
Azure Service Selection Guide
Quick reference for choosing the right Azure service based on workload requirements.
Table of Contents
- Compute Services
- Database Services
- Storage Services
- Messaging and Events
- Networking
- Security and Identity
- Monitoring and Observability
Compute Services
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | Event-driven, short tasks (<10 min) | Azure Functions (Consumption) | | Event-driven, longer tasks (<30 min) | Azure Functions (Premium) | | Containerized apps, simple deployment | Azure Container Apps | | Full Kubernetes control | AKS | | Traditional web apps (PaaS) | App Service | | GPU, HPC, custom OS | Virtual Machines | | Batch processing | Azure Batch | | Simple container from source | App Service (container) |
Azure Functions vs Container Apps vs AKS vs App Service
| Feature | Functions | Container Apps | AKS | App Service | |---------|-----------|---------------|-----|-------------| | Scale to zero | Yes (Consumption) | Yes | No (min 1 node) | No | | Kubernetes | No | Built on K8s (abstracted) | Full K8s | No | | Cold start | 1-5s (Consumption) | 0-2s | N/A | N/A | | Max execution time | 10 min (Consumption), 30 min (Premium) | Unlimited | Unlimited | Unlimited | | Languages | C#, JS, Python, Java, Go, Rust, PowerShell | Any container | Any container | .NET, Node, Python, Java, PHP, Ruby | | Pricing model | Per-execution | Per vCPU-second | Per node | Per plan | | Best for | Event handlers, APIs, scheduled jobs | Microservices, APIs | Complex platforms, multi-team | Web apps, APIs, mobile backends | | Operational complexity | Low | Low-Medium | High | Low | | Dapr integration | No | Built-in | Manual | No | | KEDA autoscaling | No | Built-in | Manual install | No |
Opinionated recommendation:
- Start with App Service for web apps and APIs — simplest operational model.
- Use Container Apps for microservices — serverless containers without Kubernetes complexity.
- Use AKS only when you need full Kubernetes API access (custom operators, service mesh, multi-cluster).
- Use Functions for event-driven glue (queue processing, webhooks, scheduled jobs).
VM Size Selection
| Workload | Series | Example | vCPUs | RAM | Use Case | |----------|--------|---------|-------|-----|----------| | General purpose | Dv5/Dsv5 | Standard_D4s_v5 | 4 | 16 GB | Web servers, small databases | | Memory optimized | Ev5/Esv5 | Standard_E8s_v5 | 8 | 64 GB | Databases, caching, analytics | | Compute optimized | Fv2/Fsv2 | Standard_F8s_v2 | 8 | 16 GB | Batch processing, ML inference | | Storage optimized | Lsv3 | Standard_L8s_v3 | 8 | 64 GB | Data warehouses, large databases | | GPU | NCv3/NDv4 | Standard_NC6s_v3 | 6 | 112 GB | ML training, rendering |
Always use v5 generation or newer — better price-performance than older series.
Database Services
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | Relational, SQL Server compatible | Azure SQL Database | | Relational, PostgreSQL | Azure Database for PostgreSQL Flexible Server | | Relational, MySQL | Azure Database for MySQL Flexible Server | | Document / multi-model, global distribution | Cosmos DB | | Key-value cache, sessions | Azure Cache for Redis | | Time-series, IoT data | Azure Data Explorer (Kusto) | | Full-text search | Azure AI Search (formerly Cognitive Search) | | Graph database | Cosmos DB (Gremlin API) |
Cosmos DB vs Azure SQL vs PostgreSQL
| Feature | Cosmos DB | Azure SQL | PostgreSQL Flexible | |---------|-----------|-----------|-------------------| | Data model | Document, key-value, graph, table, column | Relational | Relational + JSON | | Global distribution | Native multi-region writes | Geo-replication (async) | Read replicas | | Consistency | 5 levels (strong to eventual) | Strong | Strong | | Scaling | RU/s (auto or manual) | DTU or vCore | vCore | | Serverless tier | Yes | Yes | No | | Best for | Global apps, variable schema, low-latency reads | OLTP, complex queries, transactions | PostgreSQL ecosystem, extensions | | Pricing model | Per RU/s + storage | Per DTU or per vCore | Per vCore | | Managed backups | Continuous + point-in-time | Automatic + long-term retention | Automatic |
Opinionated recommendation:
- Default to Azure SQL Serverless for most relational workloads — auto-pause saves money in dev/staging.
- Use PostgreSQL Flexible when you need PostGIS, full-text search, or specific PostgreSQL extensions.
- Use Cosmos DB only when you need global distribution, sub-10ms latency, or flexible schema.
- Never use Cosmos DB for workloads that need complex joins or transactions across partitions.
Azure SQL Tier Selection
| Tier | Use Case | Compute | Cost Range | |------|----------|---------|------------| | Basic / S0 | Dev/test, tiny workloads | 5 DTUs | $5/month | | General Purpose (Serverless) | Variable workloads, dev/staging | 0.5-40 vCores (auto-pause) | $40-800/month | | General Purpose (Provisioned) | Steady production workloads | 2-80 vCores | $150-3000/month | | Business Critical | High IOPS, low latency, readable secondary | 2-128 vCores | $400-8000/month | | Hyperscale | Large databases (>4 TB), instant scaling | 2-128 vCores | $200-5000/month |
Storage Services
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | Unstructured data (files, images, backups) | Blob Storage | | File shares (SMB/NFS) | Azure Files | | High-performance file shares | Azure NetApp Files | | Data Lake (analytics, big data) | Data Lake Storage Gen2 | | Disk storage for VMs | Managed Disks | | Queue-based messaging (simple) | Queue Storage | | Table data (simple key-value) | Table Storage (or Cosmos DB Table API) |
Blob Storage Tiers
| Tier | Access Pattern | Cost (per GB/month) | Access Cost | Use Case | |------|---------------|---------------------|-------------|----------| | Hot | Frequent access | $0.018 | Low | Active data, web content | | Cool | Infrequent (30+ days) | $0.01 | Medium | Backups, older data | | Cold | Rarely accessed (90+ days) | $0.0036 | Higher | Compliance archives | | Archive | Almost never (180+ days) | $0.00099 | High (rehydrate required) | Long-term retention |
Always set lifecycle management policies. Rule of thumb: Hot for 30 days, Cool for 90 days, Cold or Archive after that.
Messaging and Events
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | Pub/sub, event routing, reactive | Event Grid | | Reliable message queues, transactions | Service Bus | | High-throughput event streaming | Event Hubs | | Simple task queues | Queue Storage | | IoT device telemetry | IoT Hub |
Event Grid vs Service Bus vs Event Hubs
| Feature | Event Grid | Service Bus | Event Hubs | |---------|-----------|-------------|------------| | Pattern | Pub/Sub events | Message queue / topic | Event streaming | | Delivery | At-least-once | At-least-once (peek-lock) | At-least-once (partitioned) | | Ordering | No guarantee | FIFO (sessions) | Per partition | | Max message size | 1 MB | 256 KB (Standard), 100 MB (Premium) | 1 MB (Standard), 20 MB (Premium) | | Retention | 24 hours | 14 days (Standard) | 1-90 days | | Throughput | Millions/sec | Thousands/sec | Millions/sec | | Best for | Reactive events, webhooks | Business workflows, commands | Telemetry, logs, analytics | | Dead letter | Yes | Yes | Via capture to storage |
Opinionated recommendation:
- Event Grid for reactive, fan-out scenarios (blob uploaded, resource created, custom events).
- Service Bus for reliable business messaging (orders, payments, workflows). Use topics for pub/sub, queues for point-to-point.
- Event Hubs for high-volume telemetry, log aggregation, and streaming analytics.
Networking
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | Global HTTP load balancing + CDN + WAF | Azure Front Door | | Regional Layer 7 load balancing + WAF | Application Gateway | | Regional Layer 4 load balancing | Azure Load Balancer | | DNS management | Azure DNS | | DNS-based global traffic routing | Traffic Manager | | Private connectivity to PaaS | Private Endpoints | | Site-to-site VPN | VPN Gateway | | Dedicated private connection | ExpressRoute | | Outbound internet from VNet | NAT Gateway | | DDoS protection | Azure DDoS Protection |
Front Door vs Application Gateway vs Load Balancer
| Feature | Front Door | Application Gateway | Load Balancer | |---------|-----------|-------------------|--------------| | Layer | 7 (HTTP/HTTPS) | 7 (HTTP/HTTPS) | 4 (TCP/UDP) | | Scope | Global | Regional | Regional | | WAF | Yes (Premium) | Yes (v2) | No | | SSL termination | Yes | Yes | No | | CDN | Built-in | No | No | | Health probes | Yes | Yes | Yes | | Best for | Global web apps, multi-region | Single-region web apps | TCP/UDP workloads, internal LB |
Security and Identity
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | User authentication | Entra ID (Azure AD) | | B2C customer identity | Entra External ID (Azure AD B2C) | | Secrets, keys, certificates | Key Vault | | Service-to-service auth | Managed Identity | | Network access control | NSGs + Private Endpoints | | Web application firewall | Front Door WAF or App Gateway WAF | | Threat detection | Microsoft Defender for Cloud | | Policy enforcement | Azure Policy | | Privileged access management | Entra ID PIM |
Managed Identity Usage
| Scenario | Configuration | |----------|---------------| | App Service accessing SQL | System-assigned MI + Azure SQL Entra auth | | Functions accessing Key Vault | System-assigned MI + Key Vault RBAC | | AKS pods accessing Cosmos DB | Workload Identity + Cosmos DB RBAC | | VM accessing Storage | System-assigned MI + Storage RBAC | | DevOps pipeline deploying | Workload Identity Federation (no secrets) |
Rule: Every Azure service that supports Managed Identity should use it. No connection strings with passwords, no service principal secrets in config.
Monitoring and Observability
Decision Matrix
| Requirement | Recommended Service | |-------------|---------------------| | Application performance monitoring | Application Insights | | Log aggregation and queries | Log Analytics (KQL) | | Metrics and alerts | Azure Monitor | | Dashboards | Azure Dashboard or Grafana (managed) | | Distributed tracing | Application Insights (OpenTelemetry) | | Cost monitoring | Cost Management + Budgets | | Security monitoring | Microsoft Defender for Cloud | | Compliance monitoring | Azure Policy + Regulatory Compliance |
Every resource should have diagnostic settings sending logs and metrics to a Log Analytics workspace. Non-negotiable for production.