Edge computing creates orchestration challenges that standard Kubernetes wasn't designed for — intermittent connectivity, resource-constrained nodes, and autonomous operation requirements. KubeEdge, a graduated CNCF project, extends Kubernetes to edge environments by adding offline-capable edge nodes, a lightweight EdgeCore binary (~70MB), local state caching via SQLite, and a device management framework that represents physical IoT devices as Kubernetes custom resources. The architecture uses CloudCore as a single gateway aggregating communication from up to 100,000 edge nodes, preventing API server overload. Unlike vanilla Kubernetes, disconnected edge nodes don't trigger pod evictions — they continue running autonomously and sync on reconnect. Real-world deployments include China's 100,000-node highway toll system and a satellite using onboard AI to reduce data transmission by 90%. Giant Swarm chose KubeEdge over K3s because it integrates with any existing Kubernetes control plane without separate lifecycle management. Known production gotchas include CNI compatibility issues (specifically Cilium with eBPF on certain edge hardware due to missing kernel flags).
Table of contents
The edge isn't the cloud, and that's the problemWhy Kubernetes makes sense (and where it breaks down)KubeEdge: extending Kubernetes to the edgeHow the architecture worksManaging physical devices as Kubernetes resourcesWhy we chose KubeEdge over K3sWhat to watch for in productionWhere to go from hereSort: