Container orchestration decides where containers run, keeps them alive, scales them and connects them. Kubernetes dominates, but there is a wide range of ways to get it, plus several non-Kubernetes alternatives. Start with 3. Container Images and Build Tools and 1. Docker, then see how Kubernetes itself works in 2. Kubernetes Architecture – from Container Images to Running Pods.

Options at a glance

flowchart TD
    A[Need to run containers] --> B{Do you need Kubernetes itself?}
    B -->|No - just run my containers| C[Serverless containers: Cloud Run, Container Apps, App Runner, Fargate]
    B -->|Simple multi-host, few ops| D[ECS, Nomad, Docker Swarm]
    B -->|Yes| E{Who runs the control plane?}
    E -->|Cloud provider| F[EKS, GKE, AKS and other managed services]
    E -->|Vendor platform with support| G[OpenShift, Rancher, Tanzu]
    E -->|Me| H{Where?}
    H -->|Data centre / bare metal| I[kubeadm, Kubespray, RKE2, Talos]
    H -->|Edge or small hardware| J[k3s, MicroK8s, k0s]
    H -->|Laptop| K[minikube, kind, k3d]

Kubernetes (the baseline)

Open source, vendor-neutral orchestrator from the CNCF. You get the control plane, nodes, networking (CNI), storage (CSI) and everything else covered in the chapters.

• Upstream / "vanilla": you install and upgrade it yourself
• Core objects: Pod, Deployment, Service, Ingress, ConfigMap, Secret, PersistentVolume, Namespace, and custom resources
• Extensible through CRDs and operators
• Every option below that says "Kubernetes" is conformant, so the same manifests and kubectl commands work, which is what makes migration and multi-cloud possible

Self-managed installers

Self-managing means you own control plane upgrades, etcd backups, certificates, HA and security patching.

Managed Kubernetes on the Big Three

The provider runs the control plane (API server, etcd, scheduler, controllers); you manage workloads and, depending on mode, nodes. Cluster-level background in 3. Installing Your First Kubernetes Cluster.

flowchart LR
    subgraph Provider[Managed by the cloud provider]
        CP[Control plane: kube-apiserver, etcd, scheduler, controller-manager]
    end
    subgraph You[Managed by you]
        NP[Node pools / VMs]
        W[Workloads: Deployments, Services, Ingress]
    end
    CP --> NP --> W

# EKS
$ eksctl create cluster --name demo --region af-south-1 --nodes 2
$ aws eks update-kubeconfig --name demo --region af-south-1

# GKE
$ gcloud container clusters create-auto demo --region=europe-west1
$ gcloud container clusters get-credentials demo --region=europe-west1

# AKS
$ az aks create -g rg-demo -n demo --node-count 2 --generate-ssh-keys
$ az aks get-credentials -g rg-demo -n demo

Related notes: 3. AWS Compute (EKS, ECS, Fargate), 2. Azure Compute (AKS, Container Apps), 5. AWS Networking, 3. Azure Networking.

Other managed offerings

• DigitalOcean DOKS, Linode LKE, Vultr VKE, Civo: simple, low cost, free or cheap control planes
• Oracle OKE, IBM Cloud Kubernetes Service, Alibaba ACK: provider-integrated
• Hybrid / multi-cloud control: Google Distributed Cloud (Anthos), Azure Arc-enabled Kubernetes, EKS Anywhere, Rancher

Red Hat OpenShift and Enterprise Platforms

OpenShift

Red Hat OpenShift is Kubernetes plus a full, opinionated, supported platform: security defaults, built-in registry, CI/CD (Pipelines on Tekton, GitOps with Argo CD), service mesh, monitoring, an operator catalogue (OperatorHub), a developer / admin web console and the oc CLI (a superset of kubectl).

Things that differ from plain Kubernetes: - Security Context Constraints (SCC): containers run as a random non-root UID by default, so images that assume root often fail. Build with OpenShift in mind (non-root, group-writable dirs, Red Hat UBI bases) - Routes (HAProxy-based) in addition to Ingress - Projects = namespaces with extras; ImageStreams; BuildConfigs with Source-to-Image (3. Container Images and Build Tools) - Operators manage the platform itself, including upgrades - Subscription licensing

$ oc login https://api.cluster.example.com:6443
$ oc new-project demo
$ oc new-app python~https://github.com/sharif/shop-api.git
$ oc expose service shop-api
$ oc get routes

Rancher and RKE2 (SUSE)

• Rancher is a management plane to create, import and govern many clusters (EKS, GKE, AKS, on-prem) from one UI with RBAC, monitoring and catalogues
• RKE2 is a security-hardened distribution; k3s is the lightweight sibling
• Rancher Desktop for local development

VMware Tanzu (now under Broadcom)

Kubernetes integrated with vSphere (vSphere with Tanzu), plus the Tanzu Application Platform. Common in VMware-centric data centres.

Others

• Canonical Charmed Kubernetes / MicroK8s: Ubuntu-centred
• Nutanix NKE, HPE Ezmeral, Mirantis (MKE): enterprise offerings

Lightweight and Edge Kubernetes

Not-Kubernetes Alternatives

Choose these when Kubernetes' power is more than you need.

AWS ECS (Elastic Container Service) and Fargate

AWS's own orchestrator: task definitions, services, clusters. Run on EC2 or Fargate (serverless, no hosts to manage). Simpler than Kubernetes, but AWS-specific. Also App Runner for "give me a URL for this image or repo". See 3. AWS Compute.

Google Cloud Run

Serverless containers: deploy an image, get an HTTPS URL, scale to zero, pay per request. Built on Knative.

Azure Container Apps and Container Instances

Container Apps: serverless containers on Kubernetes with KEDA, Dapr and revisions, no cluster to run. ACI: single container groups for short jobs. See 2. Azure Compute.

HashiCorp Nomad

Single binary scheduler for containers and VMs, binaries and batch jobs. Integrates with Consul and Vault. Simple to operate, smaller ecosystem.

Docker Swarm

Built into Docker: docker swarm init, docker stack deploy -c compose.yaml app. Easy for small clusters, but largely in maintenance mode; the community moved to Kubernetes. Compose files carry over (2. Docker Compose).

Apache Mesos / Marathon

Historically important (Twitter, Apple), now legacy and effectively retired; mentioned for context only.

Choosing

Factors to weigh

• Operational capacity: who patches, upgrades, backs up etcd and carries the pager?
• Portability: Kubernetes-conformant services are portable; ECS and Cloud Run are not
• Cost: control plane fee + nodes + load balancers + egress; serverless saves at low or spiky load
• Ecosystem fit: IAM / identity, registry, networking and monitoring in your cloud
• Compliance and support contracts
• Skills: Kubernetes is broad; a smaller tool may suit a small team better

Concept Map

Pipeline integration for all of these is in CI-CD Pipelines.