1. 企业级容器编排需求背景当企业应用从单体架构转向微服务架构时服务数量往往呈指数级增长。传统手工部署方式在50个以上容器实例的场景中会面临三大核心挑战首先是服务调度效率低下运维人员需要手动登录每台主机部署容器其次是资源利用率不均衡某些节点负载过高而其他节点闲置最重要的是缺乏自愈能力容器崩溃后无法自动恢复。这些痛点直接催生了对容器编排系统的刚性需求。Kubernetes作为CNCF毕业项目其设计哲学完美匹配企业级诉求。控制平面Control Plane通过API Server集中管理集群状态Scheduler智能分配计算资源Controller Manager确保系统始终朝向期望状态收敛。这种声明式API控制器循环的架构范式使得Kubernetes在阿里巴巴双11、京东618等超大规模场景中经受住了实战检验。2. Kubernetes集群架构解析2.1 控制平面核心组件etcd作为分布式键值存储采用Raft一致性算法保证集群状态数据的强一致性。生产环境中建议部署3/5/7个节点组成高可用集群写入性能约1000-2000 QPS取决于value大小。以下是通过etcdctl检查集群健康状态的典型命令ETCDCTL_API3 etcdctl \ --endpointshttps://192.168.1.100:2379 \ --cacert/etc/kubernetes/pki/etcd/ca.crt \ --cert/etc/kubernetes/pki/etcd/server.crt \ --key/etc/kubernetes/pki/etcd/server.key \ endpoint healthkube-apiserver作为唯一入口其水平扩展能力直接影响集群规模。建议部署至少3个实例做负载均衡启用--max-requests-inflight800和--max-mutating-requests-inflight400参数使用专用etcd集群分离业务数据与系统数据2.2 工作节点优化配置kubelet的--max-pods参数需要根据节点规格精细调整。例如AWS c5.2xlarge实例8vCPU/16GB内存建议配置apiVersion: kubelet.config.k8s.io/v1beta1 kind: KubeletConfiguration maxPods: 110 cpuManagerPolicy: static reservedSystemCPUs: 0-1 kubeReserved: cpu: 500m memory: 1Gi容器运行时推荐containerd而非Docker因其更轻量且CNCF认证。关键配置在/etc/containerd/config.toml[plugins.io.containerd.grpc.v1.cri] sandbox_image registry.k8s.io/pause:3.6 [plugins.io.containerd.grpc.v1.cri.containerd] snapshotter overlayfs disable_snapshot_annotations true3. 企业级部署实践3.1 高可用拓扑设计生产环境推荐采用多可用区部署每个区部署独立的控制平面。网络方案选择需考虑Calico适合需要网络策略的场景Cilium提供eBPF加速和Service Mesh集成Flannel配置简单但功能有限使用kubeadm初始化集群时指定--control-plane-endpoint指向负载均衡器kubeadm init \ --control-plane-endpoint k8s-api.example.com:6443 \ --upload-certs \ --pod-network-cidr192.168.0.0/163.2 应用编排示例以下是一个电商平台的典型部署组合前端Deployment HPA Ingress订单服务StatefulSet Headless ServiceRedisOperator管理的集群MySQL通过Custom Resource Definition部署示例Deployment配置包含就绪探针和资源限制apiVersion: apps/v1 kind: Deployment metadata: name: payment-service spec: replicas: 3 strategy: rollingUpdate: maxSurge: 1 maxUnavailable: 0 template: spec: containers: - name: payment image: registry.example.com/payment:v1.2.3 readinessProbe: httpGet: path: /health port: 8080 initialDelaySeconds: 5 periodSeconds: 10 resources: requests: cpu: 500m memory: 512Mi limits: cpu: 2 memory: 2Gi4. 运维监控体系4.1 可观测性方案Prometheus-Operator自动抓取指标建议配置apiVersion: monitoring.coreos.com/v1 kind: Prometheus metadata: name: k8s spec: serviceMonitorSelector: matchLabels: team: backend resources: requests: memory: 8Gi retention: 15d日志收集采用FluentdElasticsearch组合时需注意为Fluentd配置buffer_queue_limit和buffer_chunk_limitES集群至少3个master节点2个data节点使用Curator定期清理旧索引4.2 故障排查流程当Pod处于CrashLoopBackOff状态时按以下步骤诊断查看事件日志kubectl describe pod/[pod-name]检查容器日志kubectl logs -p [pod-name]若需进入容器kubectl exec -it [pod-name] -- sh网络连通性测试kubectl run debug --imagenicolaka/netshoot常见问题处理ImagePullBackOff检查镜像地址和pull secretPending状态describe查看资源不足或亲和性规则冲突NodeNotReady检查kubelet日志journalctl -u kubelet5. 安全加固措施5.1 认证与授权启用RBAC并遵循最小权限原则。例如限制开发人员只能访问特定namespaceapiVersion: rbac.authorization.k8s.io/v1 kind: Role metadata: namespace: shop-dev name: developer rules: - apiGroups: [] resources: [pods, services] verbs: [get, list, create]5.2 网络策略禁止default命名空间的所有入站流量apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: deny-all-ingress namespace: default spec: podSelector: {} policyTypes: - Ingress5.3 运行时安全使用PodSecurityPolicy或新版PodSecurity标准限制特权容器apiVersion: policy/v1beta1 kind: PodSecurityPolicy metadata: name: restricted spec: privileged: false runAsUser: rule: MustRunAsNonRoot seLinux: rule: RunAsAny volumes: - configMap - emptyDir6. 性能调优实战6.1 API Server优化通过kube-apiserver的指标监控请求延迟kubectl get --raw /metrics | grep apiserver_request_duration_seconds关键参数调整--watch-cache-sizes# 根据资源类型设置--target-ram-mb 设置为物理内存的70%启用--enable-priority-and-fairness6.2 etcd性能提升定期执行碎片整理ETCDCTL_API3 etcdctl defrag \ --endpointshttps://127.0.0.1:2379 \ --cacert/etc/kubernetes/pki/etcd/ca.crt监控指标重点关注etcd_disk_wal_fsync_duration_secondsetcd_server_leader_changes_seen_total6.3 调度器配置设置合适的Pod优先级apiVersion: scheduling.k8s.io/v1 kind: PriorityClass metadata: name: high-priority value: 1000000 globalDefault: false description: 用于关键业务Pod7. 持续交付流水线7.1 GitOps实践ArgoCD同步策略配置示例apiVersion: argoproj.io/v1alpha1 kind: Application metadata: name: product-service spec: destination: namespace: production server: https://kubernetes.default.svc source: path: k8s/overlays/prod repoURL: gitgithub.com:company/gitops-repo.git targetRevision: HEAD syncPolicy: automated: prune: true selfHeal: true7.2 金丝雀发布策略通过Istio实现流量渐进式切换apiVersion: networking.istio.io/v1alpha3 kind: VirtualService metadata: name: reviews spec: hosts: - reviews http: - route: - destination: host: reviews subset: v1 weight: 90 - destination: host: reviews subset: v2 weight: 108. 存储方案选型8.1 动态供给配置AWS EBS存储类示例apiVersion: storage.k8s.io/v1 kind: StorageClass metadata: name: gp3-encrypted provisioner: ebs.csi.aws.com parameters: type: gp3 encrypted: true volumeBindingMode: WaitForFirstConsumer allowVolumeExpansion: true8.2 本地存储优化使用Local PersistentVolume时需注意设置合适的nodeAffinity回收策略为Retain监控磁盘健康状态示例定义apiVersion: v1 kind: PersistentVolume metadata: name: local-pv spec: capacity: storage: 500Gi accessModes: - ReadWriteOnce persistentVolumeReclaimPolicy: Retain storageClassName: local-storage local: path: /mnt/ssd nodeAffinity: required: nodeSelectorTerms: - matchExpressions: - key: kubernetes.io/hostname operator: In values: - node-19. 混合云部署策略9.1 集群联邦配置使用Kubefederation v2管理多集群kubefedctl join cluster1 \ --host-cluster-contextcluster1 \ --v2关键注意事项需要统一的CA证书体系跨集群服务发现需依赖DNS资源同步存在秒级延迟9.2 应用跨区部署通过Topology Spread Constraints实现apiVersion: apps/v1 kind: Deployment metadata: name: global-service spec: template: spec: topologySpreadConstraints: - maxSkew: 1 topologyKey: topology.kubernetes.io/zone whenUnsatisfiable: DoNotSchedule labelSelector: matchLabels: app: global-service10. 版本升级方案10.1 滚动升级步骤使用kubeadm升级控制平面# 1. 升级kubeadm yum install -y kubeadm-1.25.0 --disableexcludeskubernetes # 2. 检查升级计划 kubeadm upgrade plan # 3. 应用升级 kubeadm upgrade apply v1.25.0 # 4. 逐节点升级kubelet kubectl drain node --ignore-daemonsets yum install -y kubelet-1.25.0 kubectl-1.25.0 systemctl restart kubelet kubectl uncordon node10.2 兼容性检查关键验证点检查API弃用通知kubectl get --raw /apis | grep removalRelease验证CRD是否使用已废弃的API版本测试关键业务工作负载11. 成本优化技巧11.1 资源回收策略设置命名空间资源配额apiVersion: v1 kind: ResourceQuota metadata: name: team-quota spec: hard: requests.cpu: 20 requests.memory: 100Gi pods: 10011.2 弹性伸缩配置Cluster Autoscaler参数建议--scale-down-unneeded-time10m --scale-down-delay-after-add10m --max-node-provision-time15m结合HPA实现双层扩缩容apiVersion: autoscaling/v2 kind: HorizontalPodAutoscaler metadata: name: php-apache spec: scaleTargetRef: apiVersion: apps/v1 kind: Deployment name: php-apache minReplicas: 3 maxReplicas: 10 metrics: - type: Resource resource: name: cpu target: type: Utilization averageUtilization: 5012. 灾备恢复方案12.1 定期备份策略使用Velero执行全集群备份velero install \ --provider aws \ --bucket backup-bucket \ --secret-file ./credentials \ --use-volume-snapshotsfalse \ --plugins velero/velero-plugin-for-aws:v1.3.0 # 创建每日备份 velero schedule create daily \ --scheduleevery 24h \ --include-namespacesproduction12.2 故障转移演练模拟节点故障测试# 随机选择工作节点 NODE$(kubectl get nodes -l node-role.kubernetes.io/worker | awk NR1 {print $1} | shuf -n 1) # 隔离节点 kubectl cordon $NODE kubectl drain $NODE --ignore-daemonsets --delete-emptydir-data # 观察Pod重新调度情况 watch kubectl get pods -o wide13. 服务网格集成13.1 Istio sidecar注入自动注入配置示例apiVersion: v1 kind: Namespace metadata: name: payment labels: istio-injection: enabled13.2 流量管理策略基于版本的流量切分apiVersion: networking.istio.io/v1alpha3 kind: DestinationRule metadata: name: reviews spec: host: reviews subsets: - name: v1 labels: version: v1 - name: v2 labels: version: v214. 机器学习负载支持14.1 GPU资源调度配置nvidia-device-pluginapiVersion: v1 kind: Pod metadata: name: gpu-pod spec: containers: - name: cuda-container image: nvidia/cuda:11.0-base resources: limits: nvidia.com/gpu: 214.2 分布式训练框架使用Kubeflow TFJob运行分布式训练apiVersion: kubeflow.org/v1 kind: TFJob metadata: name: mnist spec: tfReplicaSpecs: Worker: replicas: 3 template: spec: containers: - name: tensorflow image: tensorflow/tensorflow:2.6.0-gpu command: [python, /mnist.py] resources: limits: nvidia.com/gpu: 115. 边缘计算场景15.1 K3s轻量部署单节点快速安装curl -sfL https://get.k3s.io | INSTALL_K3S_VERSIONv1.24.2k3s2 sh -15.2 边缘节点管理使用KubeEdge建立边缘连接keadm init --kubeedge-version1.11.0 \ --advertise-address192.168.1.100 \ --k8s-server-ip10.10.10.10016. 安全合规审计16.1 CIS基准检测使用kube-bench运行检查docker run --rm --pidhost -v /etc:/etc:ro -v /var:/var:ro aquasec/kube-bench:latest run --targetsmaster16.2 网络策略验证使用audit2rbac生成策略建议kubectl get pods --all-namespaces -o json | audit2rbac -f -17. 自定义扩展开发17.1 Operator开发使用Operator SDK脚手架operator-sdk init \ --domainexample.com \ --repogithub.com/example/mysql-operator17.2 Webhook配置验证准入控制链kubectl get validatingwebhookconfigurations,mutatingwebhookconfigurations18. 性能基准测试18.1 集群压力测试使用kubemark模拟大规模集群go run hack/kubemark/start-kubemark.go \ --nodes1000 \ --kubeconfig$HOME/.kube/config18.2 网络性能测量使用iperf3测试Pod间带宽kubectl run iperf-server --imagenetworkstatic/iperf3 -it -- --server kubectl run iperf-client --imagenetworkstatic/iperf3 -it -- --client iperf-server --time 6019. 多租户管理19.1 虚拟集群方案使用vcluster创建隔离环境vcluster create team-a \ -n vcluster-team-a \ --connectfalse19.2 资源隔离配置通过Pod优先级实现资源保障apiVersion: scheduling.k8s.io/v1 kind: PriorityClass metadata: name: guaranteed value: 1000000 globalDefault: false description: 关键业务Pod优先级20. 新兴技术集成20.1 Serverless集成Knative服务部署示例apiVersion: serving.knative.dev/v1 kind: Service metadata: name: hello spec: template: spec: containers: - image: gcr.io/knative-samples/helloworld-go env: - name: TARGET value: Kubernetes20.2 WebAssembly支持使用Krustlet运行WASI负载kubectl apply -f https://raw.githubusercontent.com/deislabs/krustlet/master/docs/howto/hello-world-rust.yaml Kubernetes企业级容器编排实战指南
1. 企业级容器编排需求背景当企业应用从单体架构转向微服务架构时服务数量往往呈指数级增长。传统手工部署方式在50个以上容器实例的场景中会面临三大核心挑战首先是服务调度效率低下运维人员需要手动登录每台主机部署容器其次是资源利用率不均衡某些节点负载过高而其他节点闲置最重要的是缺乏自愈能力容器崩溃后无法自动恢复。这些痛点直接催生了对容器编排系统的刚性需求。Kubernetes作为CNCF毕业项目其设计哲学完美匹配企业级诉求。控制平面Control Plane通过API Server集中管理集群状态Scheduler智能分配计算资源Controller Manager确保系统始终朝向期望状态收敛。这种声明式API控制器循环的架构范式使得Kubernetes在阿里巴巴双11、京东618等超大规模场景中经受住了实战检验。2. Kubernetes集群架构解析2.1 控制平面核心组件etcd作为分布式键值存储采用Raft一致性算法保证集群状态数据的强一致性。生产环境中建议部署3/5/7个节点组成高可用集群写入性能约1000-2000 QPS取决于value大小。以下是通过etcdctl检查集群健康状态的典型命令ETCDCTL_API3 etcdctl \ --endpointshttps://192.168.1.100:2379 \ --cacert/etc/kubernetes/pki/etcd/ca.crt \ --cert/etc/kubernetes/pki/etcd/server.crt \ --key/etc/kubernetes/pki/etcd/server.key \ endpoint healthkube-apiserver作为唯一入口其水平扩展能力直接影响集群规模。建议部署至少3个实例做负载均衡启用--max-requests-inflight800和--max-mutating-requests-inflight400参数使用专用etcd集群分离业务数据与系统数据2.2 工作节点优化配置kubelet的--max-pods参数需要根据节点规格精细调整。例如AWS c5.2xlarge实例8vCPU/16GB内存建议配置apiVersion: kubelet.config.k8s.io/v1beta1 kind: KubeletConfiguration maxPods: 110 cpuManagerPolicy: static reservedSystemCPUs: 0-1 kubeReserved: cpu: 500m memory: 1Gi容器运行时推荐containerd而非Docker因其更轻量且CNCF认证。关键配置在/etc/containerd/config.toml[plugins.io.containerd.grpc.v1.cri] sandbox_image registry.k8s.io/pause:3.6 [plugins.io.containerd.grpc.v1.cri.containerd] snapshotter overlayfs disable_snapshot_annotations true3. 企业级部署实践3.1 高可用拓扑设计生产环境推荐采用多可用区部署每个区部署独立的控制平面。网络方案选择需考虑Calico适合需要网络策略的场景Cilium提供eBPF加速和Service Mesh集成Flannel配置简单但功能有限使用kubeadm初始化集群时指定--control-plane-endpoint指向负载均衡器kubeadm init \ --control-plane-endpoint k8s-api.example.com:6443 \ --upload-certs \ --pod-network-cidr192.168.0.0/163.2 应用编排示例以下是一个电商平台的典型部署组合前端Deployment HPA Ingress订单服务StatefulSet Headless ServiceRedisOperator管理的集群MySQL通过Custom Resource Definition部署示例Deployment配置包含就绪探针和资源限制apiVersion: apps/v1 kind: Deployment metadata: name: payment-service spec: replicas: 3 strategy: rollingUpdate: maxSurge: 1 maxUnavailable: 0 template: spec: containers: - name: payment image: registry.example.com/payment:v1.2.3 readinessProbe: httpGet: path: /health port: 8080 initialDelaySeconds: 5 periodSeconds: 10 resources: requests: cpu: 500m memory: 512Mi limits: cpu: 2 memory: 2Gi4. 运维监控体系4.1 可观测性方案Prometheus-Operator自动抓取指标建议配置apiVersion: monitoring.coreos.com/v1 kind: Prometheus metadata: name: k8s spec: serviceMonitorSelector: matchLabels: team: backend resources: requests: memory: 8Gi retention: 15d日志收集采用FluentdElasticsearch组合时需注意为Fluentd配置buffer_queue_limit和buffer_chunk_limitES集群至少3个master节点2个data节点使用Curator定期清理旧索引4.2 故障排查流程当Pod处于CrashLoopBackOff状态时按以下步骤诊断查看事件日志kubectl describe pod/[pod-name]检查容器日志kubectl logs -p [pod-name]若需进入容器kubectl exec -it [pod-name] -- sh网络连通性测试kubectl run debug --imagenicolaka/netshoot常见问题处理ImagePullBackOff检查镜像地址和pull secretPending状态describe查看资源不足或亲和性规则冲突NodeNotReady检查kubelet日志journalctl -u kubelet5. 安全加固措施5.1 认证与授权启用RBAC并遵循最小权限原则。例如限制开发人员只能访问特定namespaceapiVersion: rbac.authorization.k8s.io/v1 kind: Role metadata: namespace: shop-dev name: developer rules: - apiGroups: [] resources: [pods, services] verbs: [get, list, create]5.2 网络策略禁止default命名空间的所有入站流量apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: deny-all-ingress namespace: default spec: podSelector: {} policyTypes: - Ingress5.3 运行时安全使用PodSecurityPolicy或新版PodSecurity标准限制特权容器apiVersion: policy/v1beta1 kind: PodSecurityPolicy metadata: name: restricted spec: privileged: false runAsUser: rule: MustRunAsNonRoot seLinux: rule: RunAsAny volumes: - configMap - emptyDir6. 性能调优实战6.1 API Server优化通过kube-apiserver的指标监控请求延迟kubectl get --raw /metrics | grep apiserver_request_duration_seconds关键参数调整--watch-cache-sizes# 根据资源类型设置--target-ram-mb 设置为物理内存的70%启用--enable-priority-and-fairness6.2 etcd性能提升定期执行碎片整理ETCDCTL_API3 etcdctl defrag \ --endpointshttps://127.0.0.1:2379 \ --cacert/etc/kubernetes/pki/etcd/ca.crt监控指标重点关注etcd_disk_wal_fsync_duration_secondsetcd_server_leader_changes_seen_total6.3 调度器配置设置合适的Pod优先级apiVersion: scheduling.k8s.io/v1 kind: PriorityClass metadata: name: high-priority value: 1000000 globalDefault: false description: 用于关键业务Pod7. 持续交付流水线7.1 GitOps实践ArgoCD同步策略配置示例apiVersion: argoproj.io/v1alpha1 kind: Application metadata: name: product-service spec: destination: namespace: production server: https://kubernetes.default.svc source: path: k8s/overlays/prod repoURL: gitgithub.com:company/gitops-repo.git targetRevision: HEAD syncPolicy: automated: prune: true selfHeal: true7.2 金丝雀发布策略通过Istio实现流量渐进式切换apiVersion: networking.istio.io/v1alpha3 kind: VirtualService metadata: name: reviews spec: hosts: - reviews http: - route: - destination: host: reviews subset: v1 weight: 90 - destination: host: reviews subset: v2 weight: 108. 存储方案选型8.1 动态供给配置AWS EBS存储类示例apiVersion: storage.k8s.io/v1 kind: StorageClass metadata: name: gp3-encrypted provisioner: ebs.csi.aws.com parameters: type: gp3 encrypted: true volumeBindingMode: WaitForFirstConsumer allowVolumeExpansion: true8.2 本地存储优化使用Local PersistentVolume时需注意设置合适的nodeAffinity回收策略为Retain监控磁盘健康状态示例定义apiVersion: v1 kind: PersistentVolume metadata: name: local-pv spec: capacity: storage: 500Gi accessModes: - ReadWriteOnce persistentVolumeReclaimPolicy: Retain storageClassName: local-storage local: path: /mnt/ssd nodeAffinity: required: nodeSelectorTerms: - matchExpressions: - key: kubernetes.io/hostname operator: In values: - node-19. 混合云部署策略9.1 集群联邦配置使用Kubefederation v2管理多集群kubefedctl join cluster1 \ --host-cluster-contextcluster1 \ --v2关键注意事项需要统一的CA证书体系跨集群服务发现需依赖DNS资源同步存在秒级延迟9.2 应用跨区部署通过Topology Spread Constraints实现apiVersion: apps/v1 kind: Deployment metadata: name: global-service spec: template: spec: topologySpreadConstraints: - maxSkew: 1 topologyKey: topology.kubernetes.io/zone whenUnsatisfiable: DoNotSchedule labelSelector: matchLabels: app: global-service10. 版本升级方案10.1 滚动升级步骤使用kubeadm升级控制平面# 1. 升级kubeadm yum install -y kubeadm-1.25.0 --disableexcludeskubernetes # 2. 检查升级计划 kubeadm upgrade plan # 3. 应用升级 kubeadm upgrade apply v1.25.0 # 4. 逐节点升级kubelet kubectl drain node --ignore-daemonsets yum install -y kubelet-1.25.0 kubectl-1.25.0 systemctl restart kubelet kubectl uncordon node10.2 兼容性检查关键验证点检查API弃用通知kubectl get --raw /apis | grep removalRelease验证CRD是否使用已废弃的API版本测试关键业务工作负载11. 成本优化技巧11.1 资源回收策略设置命名空间资源配额apiVersion: v1 kind: ResourceQuota metadata: name: team-quota spec: hard: requests.cpu: 20 requests.memory: 100Gi pods: 10011.2 弹性伸缩配置Cluster Autoscaler参数建议--scale-down-unneeded-time10m --scale-down-delay-after-add10m --max-node-provision-time15m结合HPA实现双层扩缩容apiVersion: autoscaling/v2 kind: HorizontalPodAutoscaler metadata: name: php-apache spec: scaleTargetRef: apiVersion: apps/v1 kind: Deployment name: php-apache minReplicas: 3 maxReplicas: 10 metrics: - type: Resource resource: name: cpu target: type: Utilization averageUtilization: 5012. 灾备恢复方案12.1 定期备份策略使用Velero执行全集群备份velero install \ --provider aws \ --bucket backup-bucket \ --secret-file ./credentials \ --use-volume-snapshotsfalse \ --plugins velero/velero-plugin-for-aws:v1.3.0 # 创建每日备份 velero schedule create daily \ --scheduleevery 24h \ --include-namespacesproduction12.2 故障转移演练模拟节点故障测试# 随机选择工作节点 NODE$(kubectl get nodes -l node-role.kubernetes.io/worker | awk NR1 {print $1} | shuf -n 1) # 隔离节点 kubectl cordon $NODE kubectl drain $NODE --ignore-daemonsets --delete-emptydir-data # 观察Pod重新调度情况 watch kubectl get pods -o wide13. 服务网格集成13.1 Istio sidecar注入自动注入配置示例apiVersion: v1 kind: Namespace metadata: name: payment labels: istio-injection: enabled13.2 流量管理策略基于版本的流量切分apiVersion: networking.istio.io/v1alpha3 kind: DestinationRule metadata: name: reviews spec: host: reviews subsets: - name: v1 labels: version: v1 - name: v2 labels: version: v214. 机器学习负载支持14.1 GPU资源调度配置nvidia-device-pluginapiVersion: v1 kind: Pod metadata: name: gpu-pod spec: containers: - name: cuda-container image: nvidia/cuda:11.0-base resources: limits: nvidia.com/gpu: 214.2 分布式训练框架使用Kubeflow TFJob运行分布式训练apiVersion: kubeflow.org/v1 kind: TFJob metadata: name: mnist spec: tfReplicaSpecs: Worker: replicas: 3 template: spec: containers: - name: tensorflow image: tensorflow/tensorflow:2.6.0-gpu command: [python, /mnist.py] resources: limits: nvidia.com/gpu: 115. 边缘计算场景15.1 K3s轻量部署单节点快速安装curl -sfL https://get.k3s.io | INSTALL_K3S_VERSIONv1.24.2k3s2 sh -15.2 边缘节点管理使用KubeEdge建立边缘连接keadm init --kubeedge-version1.11.0 \ --advertise-address192.168.1.100 \ --k8s-server-ip10.10.10.10016. 安全合规审计16.1 CIS基准检测使用kube-bench运行检查docker run --rm --pidhost -v /etc:/etc:ro -v /var:/var:ro aquasec/kube-bench:latest run --targetsmaster16.2 网络策略验证使用audit2rbac生成策略建议kubectl get pods --all-namespaces -o json | audit2rbac -f -17. 自定义扩展开发17.1 Operator开发使用Operator SDK脚手架operator-sdk init \ --domainexample.com \ --repogithub.com/example/mysql-operator17.2 Webhook配置验证准入控制链kubectl get validatingwebhookconfigurations,mutatingwebhookconfigurations18. 性能基准测试18.1 集群压力测试使用kubemark模拟大规模集群go run hack/kubemark/start-kubemark.go \ --nodes1000 \ --kubeconfig$HOME/.kube/config18.2 网络性能测量使用iperf3测试Pod间带宽kubectl run iperf-server --imagenetworkstatic/iperf3 -it -- --server kubectl run iperf-client --imagenetworkstatic/iperf3 -it -- --client iperf-server --time 6019. 多租户管理19.1 虚拟集群方案使用vcluster创建隔离环境vcluster create team-a \ -n vcluster-team-a \ --connectfalse19.2 资源隔离配置通过Pod优先级实现资源保障apiVersion: scheduling.k8s.io/v1 kind: PriorityClass metadata: name: guaranteed value: 1000000 globalDefault: false description: 关键业务Pod优先级20. 新兴技术集成20.1 Serverless集成Knative服务部署示例apiVersion: serving.knative.dev/v1 kind: Service metadata: name: hello spec: template: spec: containers: - image: gcr.io/knative-samples/helloworld-go env: - name: TARGET value: Kubernetes20.2 WebAssembly支持使用Krustlet运行WASI负载kubectl apply -f https://raw.githubusercontent.com/deislabs/krustlet/master/docs/howto/hello-world-rust.yaml