Google Kubernetes Engine

Google Kubernetes Engine (GKE) is Google Cloud's managed Kubernetes service, with an Autopilot mode in which Google runs the nodes, sold pay as you go.

GKE runs containerized applications on clusters whose control plane Google installs, patches and upgrades, so a team deploys workloads without operating Kubernetes itself. It is a Google product, sold and documented as part of Google Cloud. It is a managed implementation of the open source Kubernetes platform, and its certification under the CNCF Kubernetes conformance program was announced in 2017. It is one offering with two modes of operation rather than separate editions. In Autopilot, Google manages the nodes and their scaling; in Standard, teams run their own node pools on Compute Engine virtual machines. Release channels move the control plane and nodes to new Kubernetes versions automatically, and the Extended channel keeps a cluster on one version for longer. The Horizontal Pod Autoscaler changes replica counts on resource use, load balancer traffic or custom metrics. The cluster autoscaler resizes Standard node pools as demand rises and falls. A Kubernetes Ingress object sets up Google Cloud's external Application Load Balancer, with the Gateway API as a second route in. Network policy enforcement decides which Pods and Services may talk to each other. Persistent volumes are provisioned on demand through the Compute Engine persistent disk CSI driver, and the Secret Manager add-on mounts secrets into Pods as volumes. Fleets gather clusters from Google Cloud, other clouds and on-premises into one console, while Backup for GKE protects namespaces and Multi Cluster Ingress routes traffic across clusters. GKE suits teams already building on Google Cloud that want Kubernetes without running its control plane. Autopilot fits application teams with little Kubernetes operations experience, because node sizing and scaling stay with Google. Standard fits platform engineers who want to configure and run their own node pools. AI and machine learning teams can train and serve models on GPU and TPU accelerators, beside Balanced, Scale-Out and Arm compute classes. Organizations that also run Kubernetes on AWS or Azure can bring those clusters into one fleet through GKE Multicloud. A delivery pipeline can build container images with Cloud Build, keep them and their Helm charts in Artifact Registry, and roll them out to GKE. Rolling updates replace Pods gradually, and kubectl rollout undo returns a Deployment to its previous revision when a release misbehaves. Platform teams commit namespaces, roles and policies to the repository Config Sync watches, so every cluster in a fleet carries the same configuration. Packaged software from Cloud Marketplace installs onto a cluster as a Helm chart or a kubectl configuration. That lets a Google Cloud team run its web services and its AI workloads on shared clusters under one set of Kubernetes policies.