k3s 分布式存储 Longhorn1.7 版本升级记录

抽空水一下，有兴趣的需要先阅读傻瓜式教学:部署云原生存储 Longhorn

新特性

• 新增 cli 工具代替之前的脚本(新用户用的比较多)，其他用途暂时没了解, 官方说支持故障排查， 命令行文档
• RWX(NFS 协议)相关特性增强
  • RWX 卷快速故障转移，支持快速检测和响应 ShareManage 故障
  • 存储网络支持多次读写
• 当卷意外分离时自动删除工作负载 pod
• 支持定期和按需全量备份
• 增强了副本自动均衡功能
• 建议使用 6.7+ 内核版本

新 Bug

无法附加在 v1.5.2 和 v1.4.4 之前创建的卷

如果 Longhorn 集群包含具有以下特征的资源，请避免升级到 v1.7.0：

• 资源名称：格式为 。<volume name>-e-<8-char random id>
• 创建时间：集群上安装了 v1.5.2 和 v1.4.4 之前的 Longhorn 版本

运行以下命令以检查是否可以安全地将 Longhorn 集群升级到 v1.7.0：

[ $(kubectl -n longhorn-system get engines.longhorn.io -o name | grep -E '\-e\-[a-z0-9]{8}$' | wc -l) -gt 0 ] && echo "Please hold off on upgrading to v1.7.0 until v1.7.1 is available." || echo "Safe to upgrade to v1.7.0."

升级后体验感觉

自动重新平衡副本

之前手动干预比较多。调整如下参数阈值，会自动在同一节点内的另一个磁盘上重建相关副本

kubectl get settings.longhorn.io/replica-auto-balance-disk-pressure-percentage  -n longhorn-system

升级安装

相关镜像我已经同步到国内。

添加 Longhorn Helm 仓库

helm repo add longhorn https://charts.longhorn.io

更新

helm repo update

准备 values.yaml

我的环境限制，仅在存储节点运行。

global:
  nodeSelector:
    node-role.kubernetes.io/storage: "true"
  tolerations:
    - operator: Exists
      effect: NoSchedule

persistence:
  defaultClass: false
  defaultDataLocality: best-effort
  migratable: true

defaultSettings:
  createDefaultDiskLabeledNodes: true
  defaultDataPath: /data/k8s/longhorn
  defaultDataLocality: best-effort
  replicaAutoBalance: best-effort
  storageMinimalAvailablePercentage: 10
  systemManagedComponentsNodeSelector: "node-role.kubernetes.io/storage: true"
  taintToleration: ":NoSchedule"

longhornUI:
  replicas: 1

ingress:
  enabled: true
  ingressClassName: nginx
  host: lhsc.ysicing.local
  # annotations:
  #   nginx.ingress.kubernetes.io/auth-type: basic
  #   nginx.ingress.kubernetes.io/auth-secret: basic-auth

longhornManager:
  nodeSelector:
    node-role.kubernetes.io/storage: "true"
  tolerations:
    - operator: Exists
      effect: NoSchedule

longhornDriver:
  nodeSelector:
    node-role.kubernetes.io/storage: "true"
  tolerations:
    - operator: Exists
      effect: NoSchedule

image:
  longhorn:
    engine:
      repository: ccr.ccs.tencentyun.com/k7scn/longhorn-engine
    manager:
      repository: ccr.ccs.tencentyun.com/k7scn/longhorn-manager
    ui:
      repository: ccr.ccs.tencentyun.com/k7scn/longhorn-ui
    instanceManager:
      repository: ccr.ccs.tencentyun.com/k7scn/longhorn-instance-manager
    shareManager:
      repository: ccr.ccs.tencentyun.com/k7scn/longhorn-share-manager
    backingImageManager:
      repository: ccr.ccs.tencentyun.com/k7scn/backing-image-manager
    supportBundleKit:
      repository: ccr.ccs.tencentyun.com/k7scn/support-bundle-kit
  csi:
    attacher:
      repository: ccr.ccs.tencentyun.com/k7scn/csi-attacher
    provisioner:
      repository: ccr.ccs.tencentyun.com/k7scn/csi-provisioner
    nodeDriverRegistrar:
      repository: ccr.ccs.tencentyun.com/k7scn/csi-node-driver-registrar
    resizer:
      repository: ccr.ccs.tencentyun.com/k7scn/csi-resizer
    snapshotter:
      repository: ccr.ccs.tencentyun.com/k7scn/csi-snapshotter
    livenessProbe:
      repository: ccr.ccs.tencentyun.com/k7scn/livenessprobe

升级

helm upgrade -i longhorn longhorn/longhorn -n longhorn-system -f values.yaml