cluster-rules-default.yaml

apiVersion: events.kubesphere.io/v1alpha1
kind: Rule
metadata:
  name: cluster-rules-default
  namespace: system
  labels:
    kubesphere.io/rule-scope: cluster
spec:
  rules:
    - name: ContainerCreated
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Created" and involvedObject.fieldPath != ""
      enable: true
      type: notification
    - name: ContainerStarted
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Started" and involvedObject.fieldPath != ""
      enable: true
      type: notification
    - name: ContainerFailed
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="Failed" and involvedObject.fieldPath != ""
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Container failed
        summaryCn: 容器失败
      enable: true
      type: alert
    - name: ContainerKilling
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Killing" and involvedObject.fieldPath != ""
      enable: true
      type: notification
    - name: ContainerPreempting
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="Preempting"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Container is preemting
        summaryCn: 容器抢占中
      type: alert
    - name: ContainerBackoff
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="BackOff" and involvedObject.fieldPath != "" and count>3
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Container back-off
        summaryCn: 容器回退
      enable: true
      type: alert
    - name: ContainerUnhealthy
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="Unhealthy" and count>3
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Container is unhealthy
        summaryCn: 容器状态不良
      enable: true
      type: alert
    - name: ContainerProbeWarning
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="ProbeWarning" and count>3
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Warning to perform a probe to the container
        summaryCn: 容器探测警告
      enable: true
      type: alert
    - name: PodKillingExceededGracePeriod
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="ExceededGracePeriod"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Pod killing exceeded specified grace period
        summaryCn: pod终止超时
      enable: true
      type: alert
    - name: PodKillFailed
      condition: |
        type="Warning" and reason="FailedKillPod"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to kill pod
        summaryCn: pod终止失败
      enable: true
      type: alert
    - name: PodContainerCreateFailed
      condition: |
        type="Warning" and reason="FailedCreatePodContainer"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to create pod container
        summaryCn: pod容器创建失败
      enable: true
      type: alert
    - name: PodFailed
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="Failed" and involvedObject.fieldPath=""
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Pod failed
        summaryCn: pod失败
      enable: true
      type: alert
    - name: PodNetworkNotReady
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="NetworkNotReady"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Pod network is not ready
        summaryCn: Pod网络异常
      enable: true
      type: alert
    - name: ImagePulling
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Pulling"
      enable: true
      type: notification
    - name: ImagePulled
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Pulled"
      enable: true
      type: notification
    - name: ImagePullPolicyError
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="ErrImageNeverPull"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Wrong image pull policy
        summaryCn: 镜像拉取策略错误
      enable: true
      type: alert
    - name: ImageInspectFailed
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="InspectFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to inspect image
        summaryCn: 镜像检查失败
      enable: true
      type: alert
    - name: NodeReady
      condition: |
        type="Normal" and involvedObject.kind="Node" and reason="NodeReady"
      enable: true
      type: notification
    - name: NodeNotReady
      condition: |
        type="Normal" and involvedObject.kind="Node" and reason="NodeNotReady"
      enable: true
      type: notification
    - name: NodeSchedulable
      condition: |
        type="Normal" and involvedObject.kind="Node" and reason="NodeSchedulable"
      enable: true
      type: notification
    - name: NodeNotSchedulable
      condition: |
        type="Normal" and involvedObject.kind="Node" and reason="NodeNotSchedulable"
      enable: true
      type: notification
    - name: KubeletStarting
      condition: |
        type="Normal" and involvedObject.kind="Node" and reason="Starting"
      enable: true
      type: notification
    - name: KubeletSetupFailed
      condition: |
        type="Warning" and involvedObject.kind="Node" and reason="KubeletSetupFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to setup kubelet
        summaryCn: kubelet安装失败
      enable: true
      type: alert
    - name: VolumeAttachFailed
      condition: |
        type="Warning" and reason="FailedAttachVolume"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to attach volume
        summaryCn: 存储卷装载失败
      enable: true
      type: alert
    - name: VolumeMountFailed
      condition: |
        type="Warning" and reason="FailedMount"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to mount volume
        summaryCn: 存储卷挂载失败
      enable: true
      type: alert
    - name: VolumeResizeFailed
      condition: |
        type="Warning" and reason="VolumeResizeFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to expand/reduce volume
        summaryCn: 存储卷扩缩容失败
      enable: true
      type: alert
    - name: VolumeResizeSuccess
      condition: |
        type="Normal" and reason="VolumeResizeSuccessful"
      enable: true
      type: notification
    - name: FileSystemResizeFailed
      condition: |
        type="Warning" and reason="FileSystemResizeFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to expand/reduce file system
        summaryCn: 文件系统扩缩容失败
      enable: true
      type: alert
    - name: FileSystemResized
      condition: |
        type="Normal" and reason="FileSystemResizeSuccessful"
      enable: true
      type: notification
    - name: VolumeMapFailed
      condition: |
        type="Warning" and reason="FailedMapVolume"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to map volume
        summaryCn: 存储卷映射失败
      enable: true
      type: alert
    - name: VolumeAlreadyMounted
      condition: |
        type="Warning" and reason="AlreadyMountedVolume"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Volume is already mounted
        summaryCn: 存储卷已被挂载
      enable: true
      type: alert
    - name: VolumeAttached
      condition: |
        type="Normal" and reason="SuccessfulAttachVolume"
      enable: true
      type: notification
    - name: VolumeMounted
      condition: |
        type="Normal" and reason="SuccessfulMountVolume"
      enable: true
      type: notification
    - name: NodeRebooted
      condition: |
        type="Warning" and involvedObject.kind="Node" and reason="Rebooted"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Node Rebooted
        summaryCn: 节点重启
      enable: true
      type: alert
    - name: ContainerGCFailed
      condition: |
        type="Warning" and reason="ContainerGCFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Container GC failed
        summaryCn: 容器GC失败
      enable: true
      type: alert
    - name: ImageGCFailed
      condition: |
        type="Warning" and reason="ImageGCFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Image GC failed
        summaryCn: 镜像GC失败
      enable: true
      type: alert
    - name: NodeAllocatableEnforcementFailed
      condition: |
        type="Warning" and reason="FailedNodeAllocatableEnforcement"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Node allocatable enforcement failed
        summaryCn: 节点可分配资源更新失败
      enable: true
      type: alert
    - name: NodeAllocatableEnforcedSuccess
      condition: |
        type="Normal" and involvedObject.kind="Node" and reason="NodeAllocatableEnforced"
      enable: true
      type: notification
    - name: SandboxChanged
      condition: |
        type="Normal" and reason="SandboxChanged"
      enable: true
      type: notification
    - name: SandboxCreateFailed
      condition: |
        type="Warning" and reason="FailedCreatePodSandBox"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to create sandbox
        summaryCn: Sandbox创建失败
      enable: true
      type: alert
    - name: SandboxStatusFailed
      condition: |
        type="Warning" and reason="FailedPodSandBoxStatus"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to get sandbox status
        summaryCn: 获取Sandbox状态错误
      enable: true
      type: alert
    - name: DiskCapacityInvalid
      condition: |
        type="Warning" and reason="InvalidDiskCapacity"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Invalid disk capacity
        summaryCn: 磁盘容量配置不合法
      enable: true
      type: alert
    - name: DiskSpaceFreeFailed
      condition: |
        type="Warning" and reason="FreeDiskSpaceFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to free disk space
        summaryCn: 磁盘空间释放失败
      enable: true
      type: alert
    - name: PodStatusSyncFailed
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="FailedSync"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed To Sync Pod Status
        summaryCn: Pod状态同步失败
      enable: true
      type: alert
    - name: ConfigurationValidationFailed
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="FailedValidation"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Configuration Validation Failed
        summaryCn: 配置验证失败
      enable: true
      type: alert
    - name: LifecycleHookPostStartFailed
      condition: |
        type="Warning" and reason="FailedPostStartHook"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to postStart LifecycleHook
        summaryCn: 容器启动后的生命周期钩子运行失败
      enable: true
      type: alert
    - name: LifecycleHookPreStopFailed
      condition: |
        type="Warning" and reason="FailedPreStopHook"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to preStop LifecycleHook
        summaryCn: 容器停止前的生命周期钩子运行失败
      enable: true
      type: alert
    - name: HPASelectorError
      condition: |
        type="Warning" and involvedObject.kind="HorizontalPodAutoscaler" and reason in ("SelectorRequired","InvalidSelector")
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: HPA selector error
        summaryCn: HPA选择器错误
      enable: true
      type: alert
    - name: HPAMetricError
      condition: |
        type="Warning" and involvedObject.kind="HorizontalPodAutoscaler" and reason in ("FailedGetObjectMetric","InvalidMetricSourceType")
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: HPA metric error
        summaryCn: HPA对象指标错误
      enable: true
      type: alert
    - name: HPAConvertFailed
      condition: |
        type="Warning" and involvedObject.kind="HorizontalPodAutoscaler" and reason="FailedConvertHPA"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to convert HPA
        summaryCn: HPA转换失败
      enable: true
      type: alert
    - name: HPAGetScaleFailed
      condition: |
        type="Warning" and involvedObject.kind="HorizontalPodAutoscaler" and reason="FailedGetScale"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to get HPA scale
        summaryCn: HPA规模获取失败
      enable: true
      type: alert
    - name: HPAComputeReplicasFailed
      condition: |
        type="Warning" and involvedObject.kind="HorizontalPodAutoscaler" and reason="FailedComputeMetricsReplicas"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to compute HPA replicas
        summaryCn: HPA副本计算失败
      enable: true
      type: alert
    - name: HPARescaleFailed
      condition: |
        type="Warning" and involvedObject.kind="HorizontalPodAutoscaler" and reason="FailedRescale"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to rescale HPA size
        summaryCn: HPA规模调整失败
      enable: true
      type: alert
    - name: HPARescaleSuccess
      condition: |
        type="Normal" and involvedObject.kind="HorizontalPodAutoscaler" and reason="SuccessfulRescale"
      enable: true
      type: notification
    - name: NodeSystemOOM
      condition: |
        type="Warning" and involvedObject.kind="Node" and reason="SystemOOM"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Node system OOM encountered
        summaryCn: 节点内存溢出
      enable: true
      type: alert
    - name: VolumeBindingFailed
      condition: |
        type="Warning" and reason="FailedBinding"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Volume binding failed
        summaryCn: 存储卷绑定失败
      enable: true
      type: alert
    - name: VolumeMismatch
      condition: |
        type="Warning" and reason="VolumeMismatch"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Volume Mismatch
        summaryCn: 存储卷不匹配
      enable: true
      type: alert
    - name: VolumeRecycleFailed
      condition: |
        type="Warning" and reason="VolumeFailedRecycle"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to recycle volume
        summaryCn: 存储卷回收失败
      enable: true
      type: alert
    - name: VolumeRecycled
      condition: |
        type="Normal" and reason="VolumeRecycled"
      enable: true
      type: notification
    - name: VolumeRecyclerPodError
      condition: |
        type="Warning" and reason="RecyclerPod"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Volume Recycler pod error
        summaryCn: 存储卷回收器错误
      enable: true
      type: alert
    - name: VolumeDeleted
      condition: |
        type="Normal" and reason="VolumeDelete"
      enable: true
      type: notification
    - name: VolumeDeleteFailed
      condition: |
        type="Warning" and reason="VolumeFailedDelete"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to delete volume
        summaryCn: 存储卷删除失败
      enable: true
      type: alert
    - name: VolumeProvisionFailed
      condition: |
        type="Warning" and reason="ProvisioningFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to provision volume
        summaryCn: 存储申请失败
      enable: true
      type: alert
    - name: VolumeProvisioned
      condition: |
        type="Normal" and reason="ProvisioningSucceeded"
      enable: true
      type: notification
    - name: VolumeProvisionCleanupFailed
      condition: |
        type="Warning" and reason="ProvisioningCleanupFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to clean up provision volume
        summaryCn: 清理存储失败
      enable: true
      type: alert
    - name: VolumeExternalExpandingError
      condition: |
        type="Warning" and reason="ExternalExpanding"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Error for volume external expanding
        summaryCn: 存储外部扩展错误
      enable: true
      type: alert
    - name: PodScheduleFailed
      condition: |
        type="Warning" and involvedObject.kind="Pod" and reason="FailedScheduling"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to schedule pod
        summaryCn: pod调度失败
      enable: true
      type: alert
    - name: PodSchedulePreempted
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Preempted"
      enable: true
      type: notification
    - name: PodScheduled
      condition: |
        type="Normal" and involvedObject.kind="Pod" and reason="Scheduled"
      enable: true
      type: notification
    - name: PodCreateFailed
      condition: |
        type="Warning" and involvedObject.kind in ("Pod","ReplicaSet","DaemonSet","StatefulSet","Job") and reason="FailedCreate"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to create pod
        summaryCn: pod创建失败
      enable: true
      type: alert
    - name: PodCreated
      condition: |
        type="Normal" and involvedObject.kind in ("Pod","ReplicaSet","DaemonSet","StatefulSet","Job") and reason="SuccessfulCreate"
      enable: true
      type: notification
    - name: PodDeleteFailed
      condition: |
        type="Warning" and involvedObject.kind in ("Pod","ReplicaSet","DaemonSet","StatefulSet","Job") and reason="FailedDelete"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to delete pod
        summaryCn: pod删除失败
      enable: true
      type: alert
    - name: PodDeleted
      condition: |
        type="Normal" and involvedObject.kind in ("Pod","ReplicaSet","DaemonSet","StatefulSet","Job") and reason="SuccessfulDelete"
      enable: true
      type: notification
    - name: ReplicaSetCreateError
      condition: |
        type="Warning" and reason="ReplicaSetCreateError"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Error to create replica set for deployment
        summaryCn: 副本集创建错误
      enable: true
      type: alert
    - name: DeploymentRollbackFailed
      condition: |
        type="Warning" and reason in("DeploymentRollbackRevisionNotFound","DeploymentRollbackTemplateUnchanged")
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to rollback deployment
        summaryCn: 部署回滚失败
      enable: true
      type: alert
    - name: DeploySelectorAll
      condition: |
        type="Warning" and involvedObject.kind="Deployment" and reason="SelectingAll"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: The deploy is selecting all pods
        summaryCn: deploy选择了所有pod
      enable: true
      type: alert
    - name: DaemonSelectorAll
      condition: |
        type="Warning" and involvedObject.kind="DaemonSet" and reason="SelectingAll"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: The daemon set is selecting all pods
        summaryCn: daemonset选择了所有pod
      enable: true
      type: alert
    - name: DaemonPodFailed
      condition: |
        type="Warning" and involvedObject.kind="DaemonSet" and reason in ("FailedDaemonPod","FailedPlacement")
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed daemon pod
        summaryCn: daemonset的pod失败
      enable: true
      type: alert
    - name: LoadBalancerSyncFailed
      condition: |
        type="Warning" and reason="SyncLoadBalancerFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Error syncing load balancer
        summaryCn: 负载据衡器不可用
      enable: true
      type: alert
    - name: LoadBalancerDeleting
      condition: |
        type="Normal" and reason="DeletingLoadBalancer"
      enable: true
      type: notification
    - name: LoadBalancerEnsuring
      condition: |
        type="Normal" and reason="EnsuringLoadBalancer"
      enable: true
      type: notification
    - name: LoadBalancerEnsured
      condition: |
        type="Normal" and reason="EnsuredLoadBalancer"
      enable: true
      type: notification
    - name: LoadBalancerUnAvailable
      condition: |
        type="Warning" and reason="UnAvailableLoadBalancer"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Load balancer is not available
        summaryCn: 负载据衡器不可用
      enable: true
      type: alert
    - name: LoadBalancerUpdated
      condition: |
        type="Normal" and reason="UpdatedLoadBalancer"
      enable: true
      type: notification
    - name: LoadBalancerUpdateFailed
      condition: |
        type="Warning" and reason="UpdateLoadBalancerFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to update load balancer
        summaryCn: 更新负载据衡器失败
      enable: true
      type: alert
    - name: LoadBalancerDeleting
      condition: |
        type="Normal" and reason="DeletingLoadBalancer"
      enable: true
      type: notification
    - name: LoadBalancerDeleted
      condition: |
        type="Normal" and reason="DeletedLoadBalancer"
      enable: true
      type: notification
    - name: LoadBalancerDeleteFailed
      condition: |
        type="Warning" and reason="DeleteLoadBalancerFailed"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to delete load balancer
        summaryCn: 负载据衡器删除失败
      enable: true
      type: alert
    - name: JobGetFailed
      condition: |
        type="Warning" and involvedObject.kind="CronJob" and reason="FailedGet"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to get job
        summaryCn: 任务获取失败
      enable: true
      type: alert
    - name: JobCreated
      condition: |
        type="Normal" and involvedObject.kind="CronJob" and reason="SuccessfulCreate"
      enable: true
      type: notification
    - name: JobCreateFailed
      condition: |
        type="Warning" and involvedObject.kind="CronJob" and reason="FailedCreate"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to create job
        summaryCn: 任务创建失败
      enable: true
      type: alert
    - name: JobDeleted
      condition: |
        type="Normal" and involvedObject.kind="CronJob" and reason="SuccessfulDelete"
      enable: true
      type: notification
    - name: JobDeleteFailed
      condition: |
        type="Warning" and involvedObject.kind="CronJob" and reason="FailedDelete"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: Failed to delete job
        summaryCn: 任务删除失败
      enable: true
      type: alert
    - name: JobCompleted
      condition: |
        type="Normal" and involvedObject.kind="CronJob" and reason="SawCompletedJob"
      enable: true
      type: notification
    - name: JobUnexpected
      condition: |
        type="Warning" and involvedObject.kind="CronJob" and reason="UnexpectedJob"
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: CronJob saw unexpected job
        summaryCn: 任务非预期
      enable: true
      type: alert
    - name: JobMissing
      condition: |
        type="Normal" and involvedObject.kind="CronJob" and reason="MissingJob"
      enable: true
      type: notification
    - name: JobScheduleFailed
      condition: |
        type="Warning" and involvedObject.kind="CronJob" and reason in ("MissSchedule","FailedNeedsStart")
      labels:
        severity: warning
      annotations:
        message: '%message'
        summary: CronJob failed to schedule job
        summaryCn: 任务调度失败
      enable: true
      type: alert