Kubernetes (often shortened to k8s) is the dominant container orchestration platform. It takes a fleet of servers and turns them into a single logical compute pool. You declare what you want to run (containers, replicas, networking, storage) as YAML manifests; Kubernetes makes it happen and keeps it running. If a container crashes, Kubernetes restarts it. If a server dies, Kubernetes reschedules the work elsewhere.
A Kubernetes cluster has a control plane (API server, scheduler, controller manager, etcd) and worker nodes (machines running kubelet + container runtime + kube-proxy). Users submit declarative manifests via kubectl apply. The control plane reconciles desired state vs actual state continuously, scheduling pods (groups of containers) onto nodes, configuring networking, attaching storage, and routing traffic via services and ingresses.
Common abstractions: Deployments (managed replicas), StatefulSets (stable identity for stateful workloads), Services (stable network endpoints), Ingresses (HTTP routing). Helm is the de-facto package manager. Operators encapsulate domain-specific operational logic (running databases, message queues) as custom controllers.
Kubernetes is the standard substrate for self-hosted application infrastructure in 2026. Cloud providers offer managed Kubernetes (EKS, GKE, AKS) that handle the control plane. For applications that need more than serverless can offer (long-running processes, GPUs, custom networking), Kubernetes is usually the answer.
The Docker cheat sheet covers the container layer that Kubernetes orchestrates.