1. 为什么需要 kube-status-metrics
Kubernetes 的监控主要关注两类指标:
- 基础性能指标
CPU、内存、磁盘、网络等指标，可以通过 DaemonSet 部署 node-exporter，由 Prometheus 抓取相关指标。
- 资源对象指标
Deployment 的副本数量、Pod 的运行状态等。这些指标需要 kube-status-metrics 轮询 Kubernetes 的 API 查询，并暴露给 Prometheus 才能够看到。
2. kube-status-metrics 默认提供了哪些指标
指标类别包括:
CertificateSigningRequest Metrics
ConfigMap Metrics
CronJob Metrics
DaemonSet Metrics
Deployment Metrics
Endpoint Metrics
Horizontal Pod Autoscaler Metrics
Ingress Metrics
Job Metrics
Lease Metrics
LimitRange Metrics
MutatingWebhookConfiguration Metrics
Namespace Metrics
NetworkPolicy Metrics
Node Metrics
PersistentVolume Metrics
PersistentVolumeClaim Metrics
Pod Disruption Budget Metrics
Pod Metrics
ReplicaSet Metrics
ReplicationController Metrics
ResourceQuota Metrics
Secret Metrics
Service Metrics
StatefulSet Metrics
StorageClass Metrics
ValidatingWebhookConfiguration Metrics
VerticalPodAutoscaler Metrics
VolumeAttachment Metrics
以 Pod 为例：
kube_pod_annotations
kube_pod_info
kube_pod_ips
kube_pod_start_time
kube_pod_completion_time
kube_pod_owner
kube_pod_labels
kube_pod_nodeselectors
kube_pod_status_phase
kube_pod_status_ready
kube_pod_status_scheduled
kube_pod_containeHnfo
kube_pod_container_status_waiting
kube_pod_container_status_waiting_reason
kube_pod_container_status_running
kube_pod_container_state_started
kube_pod_container_status_terminated
kube_pod_container_status_terminated_reason
kube_pod_container_status_last_terminated_reason
kube_pod_container_status_ready
kube_pod_container_status_restarts_total
kube_pod_container_resource_requests
kube_pod_container_resource_limits
kube_pod_overhead_cpu_cores
kube_pod_overhead_memory_bytes
kube_pod_runtimeclass_name_info
kube_pod_created
kube_pod_deletion_timestamp
kube_pod_restart_policy
kube_pod_init_container_info
kube_pod_init_container_status_waiting
kube_pod_init_container_status_waiting_reason
kube_pod_init_container_status_running
kube_pod_init_container_status_terminated
kube_pod_init_container_status_terminated_reason
kube_pod_init_container_status_last_terminated_reason
kube_pod_init_container_status_ready
kube_pod_init_container_status_restarts_total
kube_pod_init_containerLresource_limits
kube_pod_init_container^resource_requests
kube_pod_spec_volumes_persistentvolumeclaims_info
kube_pod_spec_volumes_persistentvolumeclaims_readonly
kube_pod_status_reason
kube_pod_status_scheduled_time
kube_pod_status_unschedulable
相关的指标非常丰富，基本能够观测 Kubernetes 的运行状态。
3. 如何抓取 label、annotations
默认情况下，kube_pod_labels 和 kube_pod_annotations 指标仅包含名称和命名空间标签。
如果需要监控更多 labels 和 annotations，就需要用到 kube-status-metrics 的两个启动参数 --metric-labels-allowlist 和 --metric-annotations-allowlist。
需要注意的是，低版本的 kube-status-metrics 并不完全支持这两个参数，下面的配置中使用的是 2.4.2 版本。
| |
- 准备一个 Pod 作为观测目标
| |
- 观测 kube_pod_labels
开启 kube-status-metrics 开关之前
kube_pod_labels{app_kubernetes_io_component=“metrics”, app_kubernetes_io_instance=“prometheus”, app_kubernetes_io_managed_by=“Helm”, app_kubernetes_io_name=“kube-state-metrics”, app_kubernetes_io_part_of=“kube-state-metrics”, app_kubernetes_io_version=“2.3.0”, exported_namespace=“tekton-pipelines”, helm_sh_chart=“kube-state-metrics-4.4.3”, instance=“10.233.96.11:8080”, job=“kubernetes-service-endpoints”, namespace=“monitor”, node=“node2”, pod=“tekton-pipelines-controller-6f449d874b-mc7nl”, service=“prometheus-kube-state-metrics”, uid=“412f8383-1c5c-4f61-8198-453bdb204911”}
开启 kube-status-metrics 开关之后
kube_pod_labels{app_kubernetes_io_component=“metrics”, app_kubernetes_io_instance=“prometheus”, app_kubernetes_io_managed_by=“Helm”, app_kubernetes_io_name=“kube-state-metrics”, app_kubernetes_io_part_of=“kube-state-metrics”, app_kubernetes_io_version=“2.3.0”, exported_namespace=“tekton-pipelines”, helm_sh_chart=“kube-state-metrics-4.4.3”, instance=“10.233.105.11:8080”, job=“kubernetes-service-endpoints”, label_app=“tekton-pipelines-controller”, label_app_kubernetes_io_component=“controller”, label_app_kubernetes_io_instance=“default”, label_app_kubernetes_io_name=“controller”, label_app_kubernetes_io_part_of=“tekton-pipelines”, label_app_kubernetes_io_version=“v0.24.1”, label_pipeline_tekton_dev_release=“v0.24.1”, label_pod_template_hash=“6f449d874b”, label_version=“v0.24.1”, namespace=“monitor”, node=“node4”, pod=“tekton-pipelines-controller-6f449d874b-mc7nl”, service=“prometheus-kube-state-metrics”, uid=“412f8383-1c5c-4f61-8198-453bdb204911”}
会增加很多 label_ 开头的标签。
- kube_pod_annotations
开启 kube-status-metrics 开关之前
kube_pod_annotations{app_kubernetes_io_component=“metrics”, app_kubernetes_io_instance=“prometheus”, app_kubernetes_io_managed_by=“Helm”, app_kubernetes_io_name=“kube-state-metrics”, app_kubernetes_io_part_of=“kube-state-metrics”, app_kubernetes_io_version=“2.3.0”, exported_namespace=“tekton-pipelines”, helm_sh_chart=“kube-state-metrics-4.4.3”, instance=“10.233.96.11:8080”, job=“kubernetes-service-endpoints”, namespace=“monitor”, node=“node2”, pod=“tekton-pipelines-controller-6f449d874b-mc7nl”, service=“prometheus-kube-state-metrics”, uid=“412f8383-1c5c-4f61-8198-453bdb204911”}
开启 kube-status-metrics 开关之后
kube_pod_annotations{annotation_cluster_autoscaler_kubernetes_io_safe_to_evict=“false”, annotation_cni_projectcalico_org_container_id=“8a505a530b501ad80ce471e86b553257e4ec3541313bc4245233f60a04dd3619”, annotation_cni_projectcalico_org_pod_ip=“10.233.105.3/32”, annotation_cni_projectcalico_org_pod_ips=“10.233.105.3/32”, app_kubernetes_io_component=“metrics”, app_kubernetes_io_instance=“prometheus”, app_kubernetes_io_managed_by=“Helm”, app_kubernetes_io_name=“kube-state-metrics”, app_kubernetes_io_part_of=“kube-state-metrics”, app_kubernetes_io_version=“2.3.0”, exported_namespace=“tekton-pipelines”, helm_sh_chart=“kube-state-metrics-4.4.3”, instance=“10.233.105.11:8080”, job=“kubernetes-service-endpoints”, namespace=“monitor”, node=“node4”, pod=“tekton-pipelines-controller-6f449d874b-mc7nl”, service=“prometheus-kube-state-metrics”, uid=“412f8383-1c5c-4f61-8198-453bdb204911”}
会增加很多 annotation_ 开头的标签。
开启这两个开关之后，对 Prometheus 的内存、CPU、存储都会增加压力。
在我测试的环境下，集群中有 2000 个，其中仅 40 个处于 Running 状态，全部采集时 Prometheus 的内存消耗瞬间就增加了大约 400 MB，如下图:
Pod 的状态不影响 kube-status-metrics 对其指标的采集。