Skip to content

borghei/Claude-Skills / engineering/azure-cloud-architect

azure-cloud-architect

Verified

Design, review, and validate Azure cloud architectures. Use when choosing Azure compute, storage, networking, or identity services, or applying the Azure Well-Architected Framework to a workload.

references/azure-services-reference.md

View on GitHub
# Azure Services Reference

Per-service depth on the Azure services most commonly used in cloud architectures: pricing tiers, SLAs, limits, when to pick each, when to upgrade, and operational defaults.

This is a working reference, not exhaustive. Always cross-check current Azure docs for limits and pricing — Azure changes quickly.

---

## Compute services

### Azure Kubernetes Service (AKS)

**When to use:** Container workloads requiring full Kubernetes API; multi-service platform engineering; need for CRDs / Operators / custom schedulers.

**When NOT to use:** Single web app (App Service is cheaper and simpler). Batch jobs (Container Apps Jobs / Batch).

**Pricing model:**
- Control plane: free (Free tier) or $0.10/hr (Standard with uptime SLA) or $0.60/hr (Premium with extra support)
- Worker nodes: standard VM pricing (VMSS-managed)
- Add-ons: monitoring (Container Insights via Log Analytics), Azure Policy, Azure Monitor agents

**Operational defaults you should set:**
- **Enable Managed Identity** for kubelet (system-assigned) and for workloads (workload identity)
- **Enable Azure RBAC + Entra ID integration** (don't use local accounts)
- **Enable Network Policies** (Calico or Azure NPM)
- **Use Container Insights** for cluster + node + pod metrics
- **Enable Defender for Containers** (security posture management)
- **Use private cluster** unless public API endpoint is justified
- **Set up cluster autoscaler** with reasonable bounds
- **Set up Pod Disruption Budgets** on critical workloads

**Limits to know:**
- Max nodes per cluster: 1000 (most regions)
- Max pods per node: 250 (default), can go to 500 (advanced networking)
- Max node pools: 100
- AKS doesn't manage etcd backups; consider Velero for cluster state backup

### Azure App Service

**When to use:** Standard web apps (Node, Python, .NET, Java, PHP, Ruby) where you don't want to manage containers/orchestration.

**When NOT to use:** Need GPU; need cron-style scheduled jobs (use Logic Apps or Functions); need fine-grained control over OS / sidecar (use AKS).

**Tiers (Linux):**
- **Free / Shared** — dev/test only; no SLA; shared CPU
- **Basic (B1/B2/B3)** — small prod workloads; manual scale; 99.95% SLA per instance
- **Standard (S1/S2/S3)** — autoscale, deployment slots, 99.95% SLA
- **Premium v3 (P1v3/P2v3/P3v3)** — VNet integration, faster scale, dedicated cores, 99.95% SLA
- **Isolated v2** — App Service Environment v3; dedicated VNet, regulatory workloads

**Tip:** Premium v3 is now usually cheaper than Premium v2 for equivalent specs. Always check current pricing.

**Operational defaults:**
- **Use Premium v3 if you need VNet integration** (most production workloads do)
- **Enable deployment slots** (staging slot + swap) for zero-downtime deploys
- **Reference secrets from Key Vault** via App Settings: `@Microsoft.KeyVault(SecretUri=...)`
- **Enable Application Insights** for distributed tracing + metrics
- **Set custom domains + managed certificates**
- **Use Managed Identity** for outbound calls to Azure services

### Azure Container Apps

**When to use:** Microservices needing scale-to-zero, KEDA-based event scaling, container runtime without K8s overhead.

**When NOT to use:** Need full Kubernetes API (use AKS); single web app (App Service is simpler).

**Pricing model:**
- Per-vCPU-second + per-GiB-second of memory allocated, while requests are active
- Free monthly allowance: 180,000 vCPU-seconds + 360,000 GiB-seconds + 2M requests

**Operational defaults:**
- **Use revisions** for blue/green deployment
- **Use Managed Identity** for service-to-service auth
- **Use Dapr** if your services would benefit from pub/sub + state + secrets abstractions
- **Set CPU/memory limits** matching expected load (autoscale handles bursts)
- **Use private endpoints** for connections to PaaS

### Azure Functions

**When to use:** Event-driven workloads; HTTP APIs with sporadic traffic; lightweight processing; integration "glue."

**Plans:**
- **Consumption (Y1)** — pay per execution + memory-seconds; auto-scale; cold start latency; no VNet
- **Premium (EP1/EP2/EP3)** — always-warm; VNet integration; longer timeouts; per-instance pricing
- **Dedicated (App Service Plan)** — runs in your App Service Plan; predictable cost; reuse existing plan

**Operational defaults:**
- **Use Application Insights** (auto-enabled)
- **Use Premium plan for VNet** (Consumption can't reach private endpoints)
- **Use Functions v4 runtime** + supported language version
- **Set host timeout deliberately** (default 5 min Consumption; 30 min Premium)
- **Use durable functions** for orchestration patterns (don't write your own)

### Virtual Machine Scale Sets (VMSS)

**When to use:** Legacy stateful apps that need full VMs; specialized workloads (HPC, GPU); apps that can't be containerized.

**When NOT to use:** Anything stateless that fits a managed service.

**Operational defaults:**
- **Use availability zones** for HA (not single-zone)
- **Use Managed Identity**
- **Use Azure Update Manager** for OS patching
- **Tag all VMs** for cost allocation
- **Enable Defender for Servers** (Microsoft Defender Cloud)

---

## Data services

### Azure SQL Database

**When to use:** OLTP relational workload; standard SQL Server compatibility.

**Tiers / deployment models:**
- **Single Database** — most common; vCore or DTU model
- **Elastic Pool** — multiple DBs sharing resources; good for SaaS multi-tenant
- **Hyperscale** — DB > 4TB with fast restore; serverless variant available
- **Managed Instance** — SQL Server feature parity (cross-DB queries, agent, etc.)

**vCore model:**
- **General Purpose** — Standard SLA, remote storage; balanced
- **Business Critical** — local SSD, in-memory OLTP; high HA via AlwaysOn
- **Hyperscale** — separate compute / log / page servers; near-instant restore

**Operational defaults:**
- **Use Managed Identity** for app connection (no SQL auth)
- **Enable auditing → Log Analytics**
- **Enable Microsoft Defender for SQL**
- **Configure backups**: PITR retention 7-35 days; LTR up to 10 years
- **Configure geo-replication** for DR (Active geo-replication or auto-failover groups)
- **Enable Always Encrypted** for sensitive columns (column-level encryption)
- **Use Private Endpoint** (disable public access)

### Azure Cosmos DB

**When to use:** Global distribution; multi-region writes; flexible schema; need for multiple consistency models.

**APIs:**
- **NoSQL (Core)** — JSON documents; Azure-native; most features
- **MongoDB** — Mongo wire protocol compatibility
- **Cassandra** — CQL compatibility
- **Gremlin** — graph DB
- **Table** — Azure Table Storage compatibility
- **PostgreSQL (Citus)** — distributed PG; columnar; HTAP

**Consistency levels (NoSQL API):**
- **Strong** — linearizable, single-region only effectively
- **Bounded Staleness** — lag bounded by time/version
- **Session** — read-your-own-writes within a session (default; usually right)
- **Consistent Prefix** — reads see writes in order
- **Eventual** — cheapest; no ordering guarantees

**Pricing model:**
- **Manual throughput (RU/s)** — provision capacity; predictable cost
- **Autoscale** — sets a ceiling; pays for max(used, 10% of ceiling)
- **Serverless** — per-request; good for low/spiky traffic, max 1TB

**Operational defaults:**
- **Use Session consistency** unless you have specific reason for stronger
- **Use partition key thoughtfully** — hot partitions = throttling
- **Index policy**: opt out of indexing fields not queried (saves RU/cost)
- **Use private endpoint**
- **Enable continuous backup** (point-in-time restore)
- **Monitor RU consumption** per partition

### Azure Database for PostgreSQL Flexible Server

**When to use:** Need actual PostgreSQL (extensions, behavior). Most modern PG workloads.

**Tiers:**
- **Burstable** — small workloads; dev/test
- **General Purpose** — most workloads
- **Memory Optimized** — analytical / cache-heavy

**Operational defaults:**
- **Use private endpoint**
- **Enable HA** (zone-redundant if available in region)
- **Configure backups**: PITR retention 7-35 days
- **Use Managed Identity** for auth (instead of password)
- **Set up read replicas** for read scale-out
- **Enable extensions selectively** — `pg_cron`, `pg_stat_statements`, `pgvector`, etc.

### Azure Storage

**Account types:**
- **Standard general-purpose v2** — most workloads
- **Premium block blobs** — high-throughput blob (low-latency reads)
- **Premium file shares** — performance-tier file
- **Premium page blobs** — for VM disks

**Blob tiers:**
- **Hot** — frequent access; highest storage cost, lowest access cost
- **Cool** — < 30 days, infrequent access
- **Cold** — < 90 days, rare access
- **Archive** — long-term, retrieval in hours

**Operational defaults:**
- **Disable public blob access** at account level
- **Enable soft delete** for blobs + containers (7-365 day retention)
- **Enable versioning** (or use immutable storage / WORM for regulated)
- **Use Private Endpoint**
- **Use Managed Identity** for app access
- **Enable lifecycle management** to auto-tier old data
- **Enable Microsoft Defender for Storage**

### Azure Cache for Redis

**Tiers:**
- **Basic** — single node; no SLA
- **Standard** — 2 nodes (primary + replica); 99.9% SLA
- **Premium** — clustering, persistence, VNet, geo-replication; 99.95% SLA
- **Enterprise / Enterprise Flash** — Redis Enterprise features; multiple primary nodes; very large memory

**Operational defaults:**
- **Use Standard at minimum for production**
- **Premium if you need VNet integration**
- **Configure maxmemory-policy** appropriately for your access pattern
- **Set up alerts** on memory pressure, eviction count, replication lag
- **Use private endpoint**

---

## Networking services

### Virtual Network (VNet) + subnets

**Default address spaces** to avoid: 10.0.0.0/16 (very common); pick something distinct (e.g., 10.42.0.0/16) to enable peering with other environments without conflict.

**Subnets** to plan:
- App subnets (per workload)
- Database subnets (for delegation to managed PG/MySQL)
- AKS subnets (one per node pool if using kubenet; one per cluster if using Azure CNI)
- Private Endpoint subnet
- Gateway subnet (mandatory for VPN/ExpressRoute)
- Azure Firewall subnet (if used)
- Bastion subnet (if used)

### Application Gateway

**Tiers:**
- **Standard v2** — basic L7 LB
- **WAF v2** — adds OWASP Core Rule Set 3.x

**When to use:** Internal-facing or regional LB with WAF; URL-based routing; SSL termination; cookie affinity.

**When NOT to use:** Global (use Front Door); simple TCP load balancing (use Standard Load Balancer).

### Front Door

**Tiers:**
- **Standard** — basic
- **Premium** — adds private link to origin, managed rules, bot protection

**When to use:** Global L7 routing; static + dynamic content acceleration; WAF at edge; multi-region failover.

**Cache TTLs:** configure per route; default is what your origin says (Cache-Control); override if you need.

### Azure Firewall

**Tiers:**
- **Standard** — L3-L4 + FQDN allowlist
- **Premium** — TLS inspection, IDPS, URL filtering
- **Basic** — small workloads (< 250 Mbps)

**Force-tunneling pattern:** All spoke VNet egress UDR'd through the firewall in the hub. Inspect/log all outbound.

---

## Identity services

### Microsoft Entra ID (formerly Azure AD)

**Tiers:**
- **Free** — basic identity, sync from on-prem
- **P1** — Conditional Access, MFA, password write-back, Identity Protection (lite)
- **P2** — Identity Protection, PIM (Privileged Identity Management), access reviews

**P1 is the practical minimum for any organization with security requirements.**

### Key Vault

**Tiers:**
- **Standard** — software-protected keys; suitable for most secrets/certificates
- **Premium** — HSM-protected keys; required for some compliance (FIPS 140-2 Level 2)

**Operational defaults:**
- **Enable soft-delete** (mandatory)
- **Enable purge protection** (mandatory for prod)
- **Use Managed Identity** for app access; never embed keys
- **Use Private Endpoint** (disable public network access)
- **Use RBAC mode** (not access policies) for permission management
- **Rotate keys** periodically (Key Vault can auto-rotate certain keys)
- **Audit logs → Log Analytics**

---

## Monitoring & observability

### Azure Monitor + Log Analytics

**Workspaces:**
- One workspace per environment (prod / staging / dev), or per region for sovereignty
- **Pricing tiers**: Pay-as-you-go ($2.30/GB ingested typical) or Commitment tier ($/day)
- **Retention**: 30 days default (free); paid up to 730 days; archive tier beyond

**Data sources to send:**
- All Azure resource diagnostic logs (audit + resource logs)
- AKS Container Insights
- VM Insights (if using VMs)
- Application Insights (per app)

### Application Insights

**When to use:** Every web/API workload that you want distributed tracing + metrics for.

**Modes:**
- **Workspace-based** (recommended): connected to a Log Analytics workspace
- **Classic** (deprecated): standalone instance

**Sampling:** Adaptive (default) keeps cost in check; fixed-rate if you want predictable. Set sample rate to 100% in dev, lower in prod.

### Defender for Cloud

**When to use:** Always. Microsoft's CSPM + CWPP suite. Includes regulatory compliance dashboard mapping to ISO 27001 / SOC 2 / PCI-DSS / etc.

**Plans:**
- Free / Foundational CSPM
- Defender for Servers (P1/P2) — agent-based, vulnerability mgmt
- Defender for SQL — threat detection on DBs
- Defender for Containers — AKS security
- Defender for Storage — anomalous access detection
- Defender for App Service — runtime protection
- Defender for Key Vault — anomalous access

---

## Service summary table (typical defaults for production)

| Service | Tier (typical) | Why |
|---------|----------------|-----|
| AKS | Standard with uptime SLA | Need SLA for prod |
| App Service | Premium v3 | VNet integration + scale + slots |
| Container Apps | n/a (consumption-based) | No tier choice |
| Functions | Premium EP1 | For VNet + always-warm; Consumption only for sporadic |
| SQL Database | General Purpose, Gen5, vCore | Balanced; upgrade if you hit IOPS limits |
| Cosmos DB | Autoscale or serverless | Match traffic shape |
| PG Flexible Server | General Purpose with HA | HA is critical |
| Storage | Standard GPv2 | Premium only if you need it |
| Cache for Redis | Standard or Premium | Premium for VNet |
| Application Gateway | WAF v2 | WAF always on |
| Front Door | Premium | If you need global edge + WAF |
| Key Vault | Standard | Premium only for HSM mandate |
| Log Analytics | PAYG | Commitment tier above ~100GB/day |
| Entra ID | P1 | Minimum for Conditional Access |
| Defender for Cloud | Per service: enable | Cost grows with services protected |

---

## Common pricing gotchas

- **Inbound traffic is free; outbound (egress) is charged.** Plan accordingly (caching, CDN, region affinity).
- **Cross-region traffic is charged at outbound rate.** Replication, backup-to-other-region, cross-region API calls.
- **Cross-AZ traffic in same region is usually free.** (Within VNet.)
- **Read-replicas, geo-replication, etc.** double or more storage cost.
- **Premium SSDs cost much more than Standard SSDs.** Premium only if your IOPS need it.
- **Log Analytics ingestion** can dominate budget. Filter at source; archive cold logs.
- **Snapshots** are charged for what changes; sparse snapshots are cheap; full backups are not.
- **Disk size, not used capacity, is billed** for unmanaged disks. Right-size.
- **Idle resources still bill** — unused IPs, unused disks, stopped (but not deallocated) VMs.

---

## Region selection criteria

| Criteria | Lever |
|----------|-------|
| Latency to users | Pick a region closest to majority of users |
| Latency to other Azure services | Same region preferred; same paired region good DR |
| Compliance | EU workloads → EU regions; financial → specific regions sometimes mandated |
| Cost | Some regions consistently cheaper (e.g., West Europe < UK South often) |
| Service availability | Not all services in all regions; check before designing |
| Capacity | New regions sometimes constrained; check before large workloads |

**Paired regions** (Microsoft's term for Azure region pairs): used for some replication features. Most cross-region setups don't need pairs strictly, but it's a known-good DR choice.

---

## When to graduate

| You currently have | Consider upgrading to |
|--------------------|----------------------|
| App Service Basic | Premium v3 if you need VNet, scale, or slots |
| SQL DB DTU model | vCore model (more flexible) |
| Cosmos DB manual throughput | Autoscale if traffic varies |
| Single-region | Multi-region with Traffic Manager / Front Door |
| Public endpoints | Private Endpoint via Private Link |
| Shared keys for auth | Managed Identity |
| Spread tags / no governance | Azure Policy + Management Groups |
| Manual deployment | Bicep + GitHub Actions / Azure DevOps |
| Log Analytics PAYG with > 100GB/day | Commitment tier |
| AKS Free control plane | Standard (with uptime SLA) for prod |
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall