边缘计算云原生实践
边缘计算将计算能力从云端延伸到网络边缘，实现数据就近处理和低延迟响应。本文深入探讨K3s、KubeEdge等边缘计算平台的架构、部署和最佳实践。
🌐 边缘计算架构
边缘-云协同架构
yaml
edge_cloud_architecture:
cloud_datacenter:
role: "集中管理和大数据处理"
responsibilities:
- "统一控制平面"
- "应用编排和调度"
- "模型训练和优化"
- "长期数据存储和分析"
- "全局策略管理"
typical_workloads:
- "大数据批处理"
- "机器学习训练"
- "跨区域数据聚合"
- "全局业务逻辑"
technologies:
- "Kubernetes集群"
- "大数据平台"
- "AI/ML平台"
- "数据仓库"
edge_nodes:
role: "就近处理和快速响应"
responsibilities:
- "实时数据处理"
- "本地推理执行"
- "离线自治运行"
- "数据预处理和过滤"
- "设备管理和控制"
typical_workloads:
- "实时视频分析"
- "IoT数据处理"
- "边缘AI推理"
- "本地缓存服务"
technologies:
- "K3s/KubeEdge"
- "边缘AI框架"
- "时序数据库"
- "本地存储"
end_devices:
role: "数据采集和简单处理"
responsibilities:
- "传感器数据采集"
- "设备状态监控"
- "简单数据过滤"
- "协议转换"
typical_devices:
- "IoT传感器"
- "工业PLC"
- "智能摄像头"
- "边缘网关"
technologies:
- "MQTT协议"
- "Modbus协议"
- "边缘代理"
- "设备SDK"yaml
communication_patterns:
cloud_to_edge:
description: "云到边缘的下行通信"
use_cases:
application_deployment:
- "应用更新推送"
- "配置下发"
- "模型分发"
device_management:
- "设备配置"
- "固件升级"
- "策略更新"
protocols:
- "MQTT over TLS"
- "gRPC"
- "WebSocket"
- "HTTP/2"
edge_to_cloud:
description: "边缘到云的上行通信"
use_cases:
data_reporting:
- "指标上报"
- "日志收集"
- "事件通知"
data_synchronization:
- "离线数据同步"
- "增量更新"
- "状态上报"
optimization:
- "数据压缩"
- "批量上传"
- "智能过滤"
- "断点续传"
edge_to_edge:
description: "边缘节点间通信"
use_cases:
- "本地服务发现"
- "数据共享"
- "协同计算"
- "故障转移"
technologies:
- "mDNS"
- "边缘服务网格"
- "本地消息队列"🚀 K3s轻量级Kubernetes
K3s架构和部署
yaml
k3s_features:
lightweight:
binary_size: "< 100MB"
memory_footprint: "~512MB"
benefits:
- "快速部署"
- "低资源占用"
- "边缘设备友好"
- "IoT场景适用"
simplified_components:
differences_from_k8s:
removed:
- "Cloud Provider"
- "Storage Plugins(使用local-path)"
- "Legacy特性"
replaced:
etcd: "SQLite(默认)或Embedded etcd"
docker: "containerd内置"
flannel: "默认CNI"
added:
- "Helm Controller"
- "Traefik Ingress"
- "ServiceLB"
- "Local Storage Provider"
production_ready:
certifications:
- "CNCF认证Kubernetes"
- "100% Kubernetes API兼容"
- "通过Conformance测试"
enterprise_features:
- "高可用部署"
- "自动备份"
- "SELinux支持"
- "Secret加密"yaml
k3s_deployment:
single_server:
installation: |
# 单节点安装
curl -sfL https://get.k3s.io | sh -
# 验证安装
k3s kubectl get nodes
# 获取kubeconfig
sudo cat /etc/rancher/k3s/k3s.yaml
configuration: |
# K3s配置文件 /etc/rancher/k3s/config.yaml
write-kubeconfig-mode: "0644"
tls-san:
- "k3s.example.com"
- "192.168.1.100"
# 禁用默认组件
disable:
- traefik
- servicelb
# 数据目录
data-dir: "/var/lib/rancher/k3s"
# Node标签
node-label:
- "node-type=edge"
- "location=factory-a"
high_availability:
embedded_etcd: |
# HA集群部署（Embedded etcd）
# 第一个server节点
curl -sfL https://get.k3s.io | sh -s - server \
--cluster-init \
--tls-san=k3s-cluster.example.com
# 获取token
sudo cat /var/lib/rancher/k3s/server/node-token
# 添加其他server节点
curl -sfL https://get.k3s.io | sh -s - server \
--server https://k3s-server1:6443 \
--token ${NODE_TOKEN}
external_datastore: |
# 使用外部数据库（PostgreSQL/MySQL）
curl -sfL https://get.k3s.io | sh -s - server \
--datastore-endpoint="postgres://username:password@hostname:5432/k3s"
load_balancer: |
# 负载均衡器配置（HAProxy示例）
global
maxconn 50000
defaults
mode tcp
timeout connect 10s
timeout client 30s
timeout server 30s
frontend k3s-api
bind *:6443
default_backend k3s-servers
backend k3s-servers
balance roundrobin
server k3s-1 192.168.1.11:6443 check
server k3s-2 192.168.1.12:6443 check
server k3s-3 192.168.1.13:6443 check
agent_nodes:
join_cluster: |
# Worker节点加入集群
curl -sfL https://get.k3s.io | K3S_URL=https://k3s-server:6443 \
K3S_TOKEN=${NODE_TOKEN} sh -
node_labels: |
# 添加节点标签和污点
curl -sfL https://get.k3s.io | K3S_URL=https://k3s-server:6443 \
K3S_TOKEN=${NODE_TOKEN} \
sh -s - agent \
--node-label location=edge-site-a \
--node-label hardware=raspberry-pi \
--node-taint edge=true:NoScheduleK3s边缘应用部署
yaml
edge_workloads:
video_analytics:
deployment: |
apiVersion: apps/v1
kind: Deployment
metadata:
name: video-analytics
namespace: edge-apps
spec:
replicas: 1
selector:
matchLabels:
app: video-analytics
template:
metadata:
labels:
app: video-analytics
spec:
nodeSelector:
location: edge-site-a
hardware: gpu-enabled
containers:
- name: analytics
image: video-analytics:v1.0
resources:
requests:
memory: "512Mi"
cpu: "500m"
limits:
memory: "2Gi"
cpu: "2"
nvidia.com/gpu: 1
env:
- name: RTSP_URL
value: "rtsp://camera.local/stream"
- name: MODEL_PATH
value: "/models/yolov5s.pt"
volumeMounts:
- name: models
mountPath: /models
- name: output
mountPath: /output
volumes:
- name: models
persistentVolumeClaim:
claimName: model-storage
- name: output
hostPath:
path: /var/lib/analytics
type: DirectoryOrCreate
iot_gateway:
deployment: |
apiVersion: v1
kind: ConfigMap
metadata:
name: mqtt-config
data:
mosquitto.conf: |
listener 1883
allow_anonymous false
password_file /mosquitto/config/passwd
listener 8883
cafile /mosquitto/config/ca.crt
certfile /mosquitto/config/server.crt
keyfile /mosquitto/config/server.key
persistence true
persistence_location /mosquitto/data/
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: mqtt-broker
spec:
replicas: 1
template:
spec:
nodeSelector:
node-type: edge
containers:
- name: mosquitto
image: eclipse-mosquitto:2
ports:
- containerPort: 1883
name: mqtt
- containerPort: 8883
name: mqtts
volumeMounts:
- name: config
mountPath: /mosquitto/config
- name: data
mountPath: /mosquitto/data
volumes:
- name: config
configMap:
name: mqtt-config
- name: data
persistentVolumeClaim:
claimName: mqtt-data
edge_caching:
deployment: |
# 边缘缓存服务
apiVersion: apps/v1
kind: DaemonSet
metadata:
name: edge-cache
spec:
selector:
matchLabels:
app: edge-cache
template:
metadata:
labels:
app: edge-cache
spec:
nodeSelector:
node-type: edge
hostNetwork: true
containers:
- name: nginx
image: nginx:alpine
ports:
- containerPort: 80
hostPort: 80
volumeMounts:
- name: cache
mountPath: /var/cache/nginx
- name: config
mountPath: /etc/nginx/nginx.conf
subPath: nginx.conf
volumes:
- name: cache
hostPath:
path: /var/cache/edge-nginx
type: DirectoryOrCreate
- name: config
configMap:
name: nginx-cache-configyaml
edge_autonomy:
offline_capability:
local_registry: |
# 本地镜像仓库
apiVersion: apps/v1
kind: Deployment
metadata:
name: local-registry
spec:
replicas: 1
template:
spec:
containers:
- name: registry
image: registry:2
ports:
- containerPort: 5000
env:
- name: REGISTRY_STORAGE_FILESYSTEM_ROOTDIRECTORY
value: /var/lib/registry
volumeMounts:
- name: registry-data
mountPath: /var/lib/registry
volumes:
- name: registry-data
persistentVolumeClaim:
claimName: registry-storage
---
# 镜像同步Job
apiVersion: batch/v1
kind: CronJob
metadata:
name: image-sync
spec:
schedule: "0 2 * * *"
jobTemplate:
spec:
template:
spec:
containers:
- name: sync
image: skopeo:latest
command:
- /bin/bash
- -c
- |
# 同步关键镜像到本地仓库
for image in $(cat /config/images.txt); do
skopeo copy \
docker://$image \
docker://local-registry:5000/$image
done
volumeMounts:
- name: config
mountPath: /config
volumes:
- name: config
configMap:
name: sync-images-list
restartPolicy: OnFailure
local_dns: |
# CoreDNS本地DNS配置
apiVersion: v1
kind: ConfigMap
metadata:
name: coredns-custom
namespace: kube-system
data:
edge.server: |
edge.local:53 {
errors
cache 30
forward . 192.168.1.1
reload
}
# 本地服务解析
local.cluster:53 {
file /etc/coredns/local.hosts
reload
}
local.hosts: |
local-registry.local.cluster. IN A 192.168.1.100
mqtt-broker.local.cluster. IN A 192.168.1.101
data_persistence: |
# 本地数据持久化
apiVersion: apps/v1
kind: StatefulSet
metadata:
name: timescaledb
spec:
serviceName: timescaledb
replicas: 1
template:
spec:
containers:
- name: timescaledb
image: timescale/timescaledb:latest-pg14
env:
- name: POSTGRES_PASSWORD
valueFrom:
secretKeyRef:
name: postgres-secret
key: password
volumeMounts:
- name: data
mountPath: /var/lib/postgresql/data
volumeClaimTemplates:
- metadata:
name: data
spec:
accessModes: ["ReadWriteOnce"]
storageClassName: local-path
resources:
requests:
storage: 50Gi🔗 KubeEdge边缘计算平台
KubeEdge架构
yaml
kubeedge_architecture:
cloud_components:
cloudcore:
description: "云端核心组件"
modules:
cloudhub:
- "WebSocket服务器"
- "边缘节点连接管理"
- "消息路由"
edgecontroller:
- "边缘节点管理"
- "应用下发"
- "状态同步"
devicecontroller:
- "设备模型管理"
- "设备数据同步"
- "设备twin维护"
edge_components:
edgecore:
description: "边缘端核心组件"
modules:
edgehub:
- "云端通信客户端"
- "消息路由"
- "离线缓存"
edged:
- "轻量级Kubelet"
- "Pod生命周期管理"
- "本地资源管理"
eventbus:
- "MQTT事件总线"
- "设备数据路由"
- "模块间通信"
servicebus:
- "HTTP服务总线"
- "REST API代理"
- "服务发现"
devicetwin:
- "设备状态管理"
- "设备shadow维护"
- "设备数据同步"
metamanager:
- "元数据管理"
- "本地存储(SQLite)"
- "持久化保证"yaml
kubeedge_deployment:
cloudcore_installation: |
# 云端安装
# 1. 准备Kubernetes集群
# 2. 安装CloudCore
# 使用Helm安装
helm repo add kubeedge https://raw.githubusercontent.com/kubeedge/kubeedge/master/build/helm
helm install cloudcore kubeedge/cloudcore \
--set cloudCore.modules.cloudHub.advertiseAddress="{cloud-server-ip}" \
--set cloudCore.modules.cloudHub.nodeLimit=1000 \
--namespace kubeedge \
--create-namespace
# 生成边缘节点Token
kubectl get secret -n kubeedge tokensecret -o jsonpath='{.data.tokendata}' | base64 -d
edgecore_installation: |
# 边缘端安装
# 下载EdgeCore
wget https://github.com/kubeedge/kubeedge/releases/download/v1.15.0/keadm-v1.15.0-linux-amd64.tar.gz
tar -zxvf keadm-v1.15.0-linux-amd64.tar.gz
# 安装EdgeCore
sudo ./keadm join \
--cloudcore-ipport=${CLOUDCORE_IP}:10000 \
--token=${EDGE_NODE_TOKEN} \
--edgenode-name=edge-node-1 \
--cgroupdriver=systemd \
--remote-runtime-endpoint=unix:///var/run/containerd/containerd.sock
edge_configuration: |
# 边缘节点配置 /etc/kubeedge/config/edgecore.yaml
apiVersion: edgecore.config.kubeedge.io/v1alpha2
kind: EdgeCore
modules:
edgehub:
enable: true
heartbeat: 15
tlsCaFile: /etc/kubeedge/ca/rootCA.crt
tlsCertFile: /etc/kubeedge/certs/edge.crt
tlsPrivateKeyFile: /etc/kubeedge/certs/edge.key
httpServer: https://${CLOUDCORE_IP}:10002
websocket:
enable: true
server: ${CLOUDCORE_IP}:10000
edged:
enable: true
cgroupDriver: systemd
clusterDNS: "169.254.96.16"
clusterDomain: cluster.local
devicePluginEnabled: true
gpuPluginEnabled: false
imageGCHighThreshold: 80
imageGCLowThreshold: 40
maximumDeadContainersPerPod: 1
eventBus:
enable: true
mqttMode: 2 # 2: internal mqtt broker
mqttServerInternal: tcp://127.0.0.1:1883
mqttServerExternal: tcp://0.0.0.0:1883
metaManager:
enable: true
metaServer:
enable: true
server: 127.0.0.1:10550设备管理
yaml
device_management:
device_model: |
# 设备模型定义
apiVersion: devices.kubeedge.io/v1alpha2
kind: DeviceModel
metadata:
name: temperature-sensor
namespace: default
spec:
properties:
- name: temperature
description: Temperature in celsius
type:
int:
accessMode: ReadOnly
maximum: 100
minimum: -50
unit: celsius
- name: humidity
description: Humidity percentage
type:
int:
accessMode: ReadOnly
maximum: 100
minimum: 0
unit: percentage
- name: status
description: Device status
type:
string:
accessMode: ReadWrite
defaultValue: active
device_instance: |
# 设备实例
apiVersion: devices.kubeedge.io/v1alpha2
kind: Device
metadata:
name: sensor-01
namespace: default
labels:
location: factory-a
type: temperature-sensor
spec:
deviceModelRef:
name: temperature-sensor
nodeSelector:
nodeSelectorTerms:
- matchExpressions:
- key: node-role.kubernetes.io/edge
operator: Exists
protocol:
modbus:
slaveID: 1
common:
collectCycle: 10000 # ms
reportCycle: 30000 # ms
propertyVisitors:
- propertyName: temperature
modbus:
register: holding
offset: 0
limit: 1
scale: 0.1
- propertyName: humidity
modbus:
register: holding
offset: 1
limit: 1
scale: 1
mapper_deployment: |
# Modbus Mapper部署
apiVersion: apps/v1
kind: Deployment
metadata:
name: modbus-mapper
spec:
replicas: 1
template:
spec:
nodeSelector:
node-role.kubernetes.io/edge: ""
hostNetwork: true
containers:
- name: mapper
image: kubeedge/modbus-mapper:v1.0
env:
- name: MQTT_SERVER
value: "tcp://127.0.0.1:1883"
securityContext:
privileged: trueyaml
edge_autonomy_kubeedge:
edge_rule: |
# 边缘规则引擎
apiVersion: rules.kubeedge.io/v1
kind: Rule
metadata:
name: temperature-alert
namespace: default
spec:
source: "temperature-sensor"
sourceResource:
path: "/temperature"
target: "alert-service"
targetResource:
path: "/alerts"
condition: "temperature > 80"
edge_stream: |
# Sedna边缘AI推理
apiVersion: sedna.io/v1alpha1
kind: IncrementalLearningJob
metadata:
name: helmet-detection
spec:
dataset:
name: "helmet-dataset"
trainProb: 0.8
trainSpec:
template:
spec:
nodeName: cloud-node
containers:
- image: helmet-detection-train:v1
name: train
evalSpec:
template:
spec:
nodeName: cloud-node
containers:
- image: helmet-detection-eval:v1
name: eval
deploySpec:
template:
spec:
nodeName: edge-node-1
containers:
- image: helmet-detection-infer:v1
name: infer📋 边缘计算面试重点
架构设计类
边缘计算与云计算的本质区别？
- 计算位置和延迟
- 数据处理模式
- 网络带宽要求
- 运维管理方式
如何设计边缘-云协同架构？
- 工作负载划分原则
- 数据同步策略
- 离线自治能力
- 故障容错机制
边缘计算的主要应用场景？
- 工业IoT
- 智慧城市
- 车联网
- AR/VR
K3s实践类
K3s与K8s的主要差异？
- 组件简化
- 资源占用
- 部署方式
- 适用场景
如何部署K3s高可用集群？
- Embedded etcd vs外部数据库
- 负载均衡配置
- 节点故障恢复
- 备份策略
K3s边缘节点管理最佳实践？
- 节点标签和污点
- 网络断连处理
- 本地镜像管理
- 资源限制策略
KubeEdge应用类
KubeEdge的架构组件？
- CloudCore功能
- EdgeCore模块
- 通信机制
- 设备管理
如何实现设备到云端的数据流转？
- 设备模型定义
- Mapper开发
- 数据路由规则
- 云端数据处理
边缘自治能力如何实现？
- 本地数据缓存
- 离线决策
- 规则引擎
- 状态同步
技术选型类
K3s vs KubeEdge的选择？
- 功能对比
- 设备管理能力
- 部署复杂度
- 适用场景
边缘AI推理的部署策略？
- 模型压缩优化
- 硬件加速方案
- 模型更新机制
- 性能监控
边缘安全的关键考虑？
- 设备认证
- 通信加密
- 数据隐私保护
- 安全更新机制
🔗 相关内容
- 新兴技术概述 - 云原生新兴技术全景
- Serverless平台 - Serverless架构实践
- 容器技术 - 容器基础和运行时
- Kubernetes核心 - Kubernetes深入实践
边缘计算是云原生技术向边缘延伸的重要趋势，通过K3s、KubeEdge等平台可以实现云边协同的应用架构。掌握边缘计算的核心概念和实践技能，对于构建现代IoT和实时处理系统至关重要。