Azure for AWS engineers · Lesson 6 of 9 · 5 min read
AKS in depth
Cluster basics
- Control plane managed by Azure. Pricing tiers: Free (no SLA, for dev), Standard (financially backed uptime SLA, recommended for production), Premium (adds long-term support for versions). Unlike EKS's per-cluster fee, the Free tier control plane costs nothing — a frequently mentioned difference.
- Node pools are VMSS-backed: a system node pool (runs CoreDNS, metrics-server, etc.; at least one required) and user node pools for workloads — with taints, labels, Spot node pools, Windows node pools, GPU pools.
- Availability zones — spread node pools across zones (zone-redundant node pools).
az login
az account set --subscription "Shop-Prod"
az group create --name rg-shop-prod --location centralindia
az aks create -g rg-shop-prod -n aks-shop-prod \
--tier standard --kubernetes-version 1.34 \
--node-count 3 --zones 1 2 3 \
--network-plugin azure --network-plugin-mode overlay --network-dataplane cilium \
--enable-oidc-issuer --enable-workload-identity \
--enable-managed-identity --attach-acr acrshopprod \
--enable-aad --enable-azure-rbac
az aks get-credentials -g rg-shop-prod -n aks-shop-prod
az aks nodepool add -g rg-shop-prod --cluster-name aks-shop-prod -n spot --priority Spot --node-count 0 --enable-cluster-autoscaler --min-count 0 --max-count 20
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.
Networking models
| Model | Pod IPs | Notes |
|---|---|---|
| kubenet (legacy) | Overlay range, NAT to node IP | Being retired |
| Azure CNI (flat) | Real VNet IPs (≈ AWS VPC CNI) | IP exhaustion risk; dynamic allocation option |
| Azure CNI Overlay | Private overlay CIDR; nodes in VNet | Scales without consuming VNet IPs — common default now |
| Azure CNI powered by Cilium | eBPF dataplane, network policy | Modern choice |
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.
Identity: workload identity
AKS Microsoft Entra Workload ID works like EKS IRSA:
- Enable the cluster OIDC issuer and workload identity.
- Create a user-assigned managed identity and grant it roles (e.g. "Key Vault Secrets User" on a vault).
- Create a federated identity credential on that identity trusting the cluster's issuer and
system:serviceaccount:<ns>:<sa>. - Annotate the ServiceAccount with the identity's client ID (
azure.workload.identity/client-id) and label podsazure.workload.identity/use: "true". - The Azure SDK exchanges the projected token for an Entra token — no secrets.
Cluster access for humans: Entra integration (users log in with Entra; kubelogin) with Azure RBAC for Kubernetes or Kubernetes RBAC bound to Entra groups (≈ EKS access entries).
Enlarge diagram
Swipe sideways to see all of it, or tap to open it full screen.
Secrets, registry, ingress, scaling, monitoring
- Key Vault + Secrets Store CSI driver (AKS add-on) mounts secrets as files/syncs to Kubernetes Secrets, authenticated with workload identity (≈ AWS Secrets Store CSI provider / External Secrets).
- ACR integration:
--attach-acrgrants the kubelet identity AcrPull — no imagePullSecrets needed (≈ EKS node role with ECR read). - Ingress: the managed application routing add-on (NGINX-based; check its roadmap, since the upstream ingress-nginx project was retired in March 2026), Application Gateway Ingress Controller (AGIC) / Application Gateway for Containers (Gateway API), or self-managed NGINX/Traefik.
- Autoscaling: cluster autoscaler per node pool, Node Auto Provisioning (Karpenter-based), HPA, KEDA add-on.
- Upgrades: control plane then node pools, with surge settings; auto-upgrade channels and planned maintenance windows; node image upgrades separately.
- Monitoring: Container Insights (Azure Monitor + Log Analytics), Azure Monitor managed Prometheus + Azure Managed Grafana (≈ AMP/AMG).
- Policy: the Azure Policy add-on for AKS runs Gatekeeper with built-in policy definitions.
- GitOps: the Flux extension for AKS (
az k8s-configuration flux create) — Flux knowledge from EKS transfers directly.
Resilience
Zone-redundant node pools and zone-redundant storage/databases; multiple regions behind Front Door (or Traffic Manager) for regional failover; geo-replicated ACR; geo-redundant data (SQL auto-failover groups, Cosmos DB multi-region writes).