DevOps & CI/CD Deep Dive · 9 of 18

EKS · AKS · GKE — Kubernetes the Cloud Runs for You

Same Kubernetes API, different operational opinions. The cloud handles the control plane (etcd, API server, scheduler) and patches it; you bring the worker nodes (or use a serverless mode) and the workloads. Pick by where the rest of your data lives.

EKSAKSGKEFargateAutopilotk3s
← Back to DevOps & CI/CD
Side-By-Side

The Three Big Ones

ServiceStrengthsWatch out for
GKE (Google)Most polished K8s experience. Autopilot mode hides nodes entirely. Aggressive feature releases since K8s came from Google.GCP regional footprint smaller than AWS.
EKS (AWS)Tightest integration with the broader AWS surface (IAM, ALB, EBS, VPC). Fargate for serverless pods.Most assembly required. IAM-for-pods (IRSA / Pod Identity), CNI, ALB controller — all bring-your-own.
AKS (Azure)Free control plane on the standard tier. Strong AAD/Entra integration. Decent Windows-node support.Some addons lag GKE/EKS; networking models (Azure CNI vs kubenet vs Cilium) are confusing.
Serverless Modes

Pods Without Nodes

  • GKE Autopilot — Google manages nodes; you pay per pod resource request. No node-level tuning, no DaemonSets, hardened defaults.
  • EKS Fargate — schedule pods onto invisible micro-VMs. Slower cold starts, no DaemonSets, but no node patching.
  • AKS virtual-node via ACI — bursty workloads land on Azure Container Instances.

All three are great for "just run my pods" workloads; less great when you need GPUs, host-level access, or persistent local disks.

Lighter Distributions

K8s Without the Cloud

  • k3s / k0s — single-binary K8s for edge, IoT, and small VMs.
  • kind / minikube — local clusters for dev and CI.
  • OpenShift (Red Hat) — opinionated K8s distribution with built-in CI/CD, registry, RBAC, and a developer-friendly UI.
  • Rancher — multi-cluster management UI, often paired with RKE2.
Tradeoffs

What to Watch Out For

  • Control-plane cost — EKS and GKE charge ~$72/mo per cluster; small environments add up.
  • Cluster upgrades are still your job. K8s ships a minor every ~4 months; upstream support is one year.
  • Networking egress charges bite when chatty pods cross AZs. Watch your topology hints.
  • "Managed" doesn't mean autopilot — addons (CNI, ingress, autoscaler, secrets) are still your responsibility on EKS and AKS.
Continue

Other DevOps & CI/CD Tools

← Kubernetes ECS / Swarm / Nomad → Helm → Service Mesh → ↑ Back to DevOps & CI/CD