使用“KubeEdge”管理边缘节点

1 .Manage Edge Nodes with KubeEdge and Case Study Yulin Sun, yulin.sun@huawei.com; Li Xing, Li.xing1@huawei.com; Seattle Cloud Lab, Huawei R&D USA, Bellevue WA 

2 .Agenda • Edge scenarios/characters • KubeEdge Architecture • KubeBus • Edge metadata service • Kubernetes extension • Case study • Water utility management • Campus surveillance system 

3 .Edge Scenario/Characters (vs Data Center) • Similar requirement • Edge/Cloud nodes management • Application management • Inter-service communication • … • Edge special characters • Edge node running in private network • connect to Cloud behind NAT • Edge node’s connect to Cloud through Internet • Low bandwidth/High latency • Unstable connectivity • Heterogeneous hardware • Memory: 64MB ~ 100+ GB • CPU: AMD64/ARM/MIPS… 

4 .Manage Edge Nodes with KubeEdge K8S Master ETCD • Goal • Manage Edge Nodes K8S Node1 K8S Node2 K8S Node n Cloud together with Nodes in cloud as one Cluster EdgeController • Address the Edge special ... KubeBus characters Virtual Router • Edge nodes and cloud nodes in one VPN K8S Cluster • Edge nodes offline execution KubeEdge VPN autonomously • Lightweight edge agent • Components KubeBus KubeBus • KubeBus: Edge Network EdgeMetadataService EdgeMetadataService • EdgeMetadataService • Kubernetes extension App Engine App Engine ... Edge • Edge Controller • App Engine Edge Node Edge Node 

5 . KubeBus Cloud Cluster VM VM VM Container Network, e.g. Flannel 10.244.0.0/16 ... KubeBus ... Virtual Router Edge Network IP: 192.168.6.4 EIP: 35.160.180.205 192.168.23.129 192.168.23.130 VM IP: 192.168.23.131 • Scenario • Edge Nodes have no external IP and running behind NAT • One VM in Cloud Cluster has EIP • No direct connection between Edge nodes KubeBus KubeBus Edge Network: 192.168.6.2 Edge Network: 192.168.6.3 • Goal • Cloud VM and Edge nodes run as in one cluster Container Network 10.243.0.0/16 Container Network 10.242.0.0/16 ... • KubeBus Virtual Router • Public Endpoint for Edge connection • Router to connect EdgeNode EdgeNode • Cloud VM subnet • Cloud Container subnet • Edge Node virtual subnet Container • Edge Container subnet 

6 .Edge Metadata Service Cloud Cluster • Goal Cloud Service1 Cloud Service2 ... • Offline Edge Metadata Store CRUD CRUD • Cloud/Edge metadata bi-direction sync ETCD@Cloud • Component • Edge Metadata Store: Etcd Cloud • Sync Service: Configuration Status • Mirror data between Etcd@Cloud and Etcd@Edge • Base on Etcd watch/get Edge Edge MetadataService • Sync Algorithm SyncService • Eventually consistent • Atomic ETCD@Edge CRUD CRUD Edge Edge ... Service1 Service2 Edge Node 

7 . Edge Controller/App Engine • Goal Current Kubernetes Kubernetes edge extension • Offline autonomous – network connectivity KubeMaster KubeMaster • Only sync change data – network Configuration (Pod Spec) Status performance Edge Controller • Lightweight Kubelet – resource constrain edge node Etcd@Cloud • Components Sync Service • Edge Controller: representative of Edge nodes to talk to Kubelet Etcd@Edge KubeMaster • Get configuration (e.g. pod spec) App Engine • Reports Edge nodes status Compute Node Edge Node • AppEngine: Control containers 

8 .Case Study#1 --Water utilities management • Device • Liquid level sensor • Water pressure sensor • Valve actuator • Goal • Adjust valve based on water level of water tank in the area and water pressure nearby • Requirement • Management plane in Cloud • Edge to edge communication • Edge autonomous when offline … • Lightweight edge node 

9 .Case Study#2 --Campus surveillance system • Goal • Cloud/Edge collaborated campus surveillance system • Requirement • Workload balance between Edge/Cloud • Edge autonomous • Edge Cluster 

10 .Reference 1. Extend cloud execution environment to edge with KubeEdge, Yulin,Sun; Ying Xiong; Li Xing, Ying Huang, https://schd.ws/hosted_files/kccncchina2018chinese/cb/KubeEdge.pdf 

11 .