OpenYurt is a Kubernetes distribution designed to extend native Kubernetes capabilities to edge computing environments. Hosted by the Cloud Native Computing Foundation as an Incubating Level Project, OpenYurt addresses the specific challenges of managing applications across cloud-edge infrastructure while maintaining full Kubernetes API compatibility. The project was first released on May 29th, 2020 as version 0.1.0-beta.1, with the latest release being version 1.7.0 on May 6th, 2026.
The architecture follows a classic cloud-edge design pattern with a centralized Kubernetes control plane in the cloud managing multiple edge nodes distributed across edge sites. OpenYurt organizes edge nodes into logical groupings called Pools, allowing applications to span multiple physical regions. The core system consists of several key components. YurtHub runs as a static pod on worker nodes and acts as a node sidecar, intercepting and handling requests from local components like Kubelet and Kubeproxy to the kube-apiserver. Yurt-Manager contains all controllers and webhooks required for edge operations. Raven-Agent focuses on edge-to-edge and edge-to-cloud communication, providing layer 3 network connectivity among pods across different physical regions. YurtIoTDock deploys one instance per edge NodePool to bridge the EdgeX Foundry platform and manage edge devices using Kubernetes CRDs. The project also includes a Node resource manager for handling additional edge node resources such as LVM, QuotaPath, and Persistent Memory.
OpenYurt is built in Go and designed to solve specific cloud-edge orchestration challenges including unreliable or disconnected cloud-edge networking, edge autonomy requirements, edge device management, and region-aware deployment. The system preserves the Kubernetes user experience while addressing edge-specific operational needs. OpenYurt is currently certified to support Kubernetes versions up to 1.34, with compatibility expected for subsequent versions pending verification.
Installation of OpenYurt is divided into two parts: installing the control plane components and joining nodes to the cluster. The project maintains an active community with multiple communication channels including a mailing list, Slack channel for English speakers, and DingTalk for Chinese-speaking contributors.
