Kubernetes
Kubernetes is the open-source container orchestrator Google donated to the Cloud Native Computing Foundation (graduated project, Apache 2.0 license). Current release is v1.36 "Haru" (April 22, 2026), patched to v1.36.2 on June 9, 2026; v1.37.0 ships August 26, 2026 under CNCF's N-2 support window (~14 months per minor). Its control plane runs etcd, kube-apiserver, kube-scheduler, and kube-controller-manager, plus a kubelet on every node β five components a team must run before deploying a workload. Per CNCF's 2025 Annual Survey (published Jan 2026), 82% of container users run it in production, it holds 92% of orchestration-tool market share, 79% of users run it managed (EKS/GKE/AKS) rather than self-hosted, and 77% of Fortune 100 firms run it in production. Current version/status: v1.36.2 "Haru" (June 9, 2026); v1.37.0 due Aug 26, 2026. License: Apache 2.0. Pricing: Kubernetes itself is free; managed control planes cost extra β EKS standard $0.10/hr (~$73/mo, jumps to $0.60/hr after 14-month extended support), GKE $0.10/hr offset by a $74.40/mo credit covering one cluster, AKS free on the Free tier or $0.10-$0.60/hr on Standard/Premium tiers; ECS has no control-plane fee at all. Maintained by CNCF (graduated project); originally created and open-sourced by Google in 2014.
Pick Kubernetes when you have 15+ services, multiple teams, and need one API that works identically on AWS, GCP, and Azure β that portability is the entire reason it exists. Skip it for a 3-person shop running five services: ECS has zero control-plane fee versus EKS's ~$73/mo, and Nomad replaces etcd+apiserver+scheduler+controller-manager+kubelet with a single binary. Want Kubernetes-shaped without the pain? K3s swaps etcd for SQLite and is CNCF-certified for edge and sub-100-node clusters. Even Kubernetes' own ecosystem admits the weakness: teams routinely burn 40+ hours a month on control-plane care before a single app ships. Known weakness: Kubernetes practitioners themselves acknowledge the operational tax: standing up a production-ready cluster requires running etcd, kube-apiserver, kube-scheduler, and kube-controller-manager (plus kubelet per node), and small teams report 40+ hours a month of control-plane maintenance before deploying a single application.
See how it ranks β