What’s New in KubeSphere 3.2.0? GPU Scheduling, Multi‑Cluster Management & More

The most popular server‑side technology today is cloud native, and KubeSphere—an open‑source, Kubernetes‑based cloud‑native distributed operating system—continues the trend with its 3.2.0 release.

Announced on November 3, 2021, KubeSphere 3.2.0 builds on the previous 3.1.0 release that added edge‑node management and metering, and the 3.1.1 patch that integrated existing Prometheus instances.

GPU Scheduling and Quota Management

KubeSphere now offers visual creation of GPU workloads, tenant‑level GPU quota management, and support for Nvidia GPU or vGPU solutions, making GPU resource handling easier for AI and machine‑learning workloads.

Enhanced Observability

Observability is strengthened with custom monitoring panels, Grafana integration, default GPU monitoring templates, and a three‑pronged approach (Logging, Tracing, Metrics). Alerts can be sent via HTTPS to Elasticsearch, and new notification channels (email, DingTalk, WeChat, Webhook, Slack) are testable. ETCD leader nodes are automatically labeled with Leader.

Multi‑Cloud and Multi‑Cluster Management

KubeSphere supports federated deployments across multiple clusters, allowing users to specify total replica counts and weight distributions for flexible scaling.

Operations‑Friendly Storage Management

New storage volume management lets administrators configure cloning, snapshots, and expansion under StorageClass. The default Immediate binding mode is complemented by a WaitForFirstConsumer mode to improve scheduling for topology‑constrained storage. PV resources can now be viewed, edited, and deleted from the console, and VolumeSnapshotClass can be specified for snapshot creation.

Cluster‑Level Gateway

A global gateway shared by all projects replaces the previous project‑level gateways, supporting any v1/ingress Ingress Controller as a backend.

Authentication and Authorization

KubeSphere 3.2.0 adds built‑in OpenID Connect support, providing a standards‑based authentication service that can be used by other components.

Partner‑Open Application Store

Based on OpenPitrix, the application store now supports dynamic loading of Helm charts submitted by partners, removing version constraints. The default chart repository is https://github.com/kubesphere/helm-charts .

KubeSphere DevOps Becomes Independent

The DevOps component is evolving into the separate ks-devops project, with Jenkins integration, a new CRD PipelineRun for pipeline records, containerd‑based image builds, and future support for Tekton and ArgoCD.

More Flexible Cluster Deployment

KubeSphere offers two installers: KubeKey and ks‑installer. KubeKey supports the latest Kubernetes v1.22.1, automatic certificate renewal, internal load‑balancer HA, and updated components such as Istio, Jaeger, Prometheus Operator, Fluent Bit, KubeEdge, and Nginx Ingress.

User Experience Optimizations

Console UI text has been refactored for better localization, hard‑coded strings removed, and new features added such as Harbor image search, init‑container volume mounts, and preventing automatic pod restarts on volume expansion.

For detailed release notes and installation instructions, refer to the official KubeSphere 3.2.0 documentation.