Managed Kubernetes platforms — Amazon EKS, Azure AKS, and Google GKE — remove the operational burden of running the Kubernetes control plane, letting teams focus on workloads instead of infrastructure. Each cloud provider wraps vanilla Kubernetes with distinct networking models, autoscaling strategies, identity integrations, and fully-managed tiers such as GKE Autopilot, AKS Automatic, and EKS Auto Mode. Kubernetes 1.33 "Octarine" brings in-place pod resizing to GA, while Dynamic Resource Allocation (DRA) — stable in 1.35 — unlocks structured GPU scheduling beyond the legacy nvidia.com/gpu extended-resource model. The Gateway API has graduated to GA and now supersedes Ingress for new clusters, following Ingress-NGINX's retirement in March 2026.
What This Cheat Sheet Covers
This topic spans 16 focused tables and 187 indexed concepts. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.
Table 1: Platform Comparison (EKS, AKS, GKE)
Understanding the pricing model, default CNI, and managed-tier options of each platform before designing a cluster prevents costly surprises. AKS offers a three-tier control-plane pricing model; GKE gives a monthly zonal credit; EKS charges separately for Auto Mode node management.
| Platform | Example | Description |
|---|---|---|
eksctl create cluster --name prod --region us-east-1 | • Control plane: 0.10/hr** (standard, first 14 months per version)<br>• Extended support: **0.60/hr (months 15–26 after release)• Workers: separate EC2/Fargate cost; CNI: AWS VPC CNI | |
az aks create -g rg -n prod --tier standard | • Free tier: no SLA, ≤10 nodes recommended • Standard: 0.10/hr, 99.95% SLA<br>• **Premium**: 0.60/hr, 24-month LTS, 99.95% SLA; CNI default: Azure CNI Overlay | |
gcloud container clusters create prod --region us-central1 | • Standard: 0.10/hr per cluster (one **free zonal cluster / 74.40/month credit**)• Autopilot: pay-per-pod; Dataplane V2 eBPF default for Autopilot | |
gcloud container clusters create-auto autopilot-prod | • Fully-managed node pools • GKE handles all node operations • billed per vCPU/memory/storage of running pods | |
kubectl apply -f fargate-profile.yaml | • Serverless pods — no EC2 nodes • billed per vCPU and memory per second • use for burst or compliance workloads |