prometheus☞02外置prometheus监控k8s

阅读量: zyh 2021-03-11 14:37:28
Categories: > Tags:

基本

prometheus 获取 k8s 指标,需要从三个方面获取:

而上述三个方面,均需要 prometheus 去访问 k8s apiserver。这里需要提前创建授权token.

不过kube-state-metrics和metrics-server之前还有很大不同的,二者主要区别如下

1.kube-state-metrics主要关注的是业务相关的一些元数据,比如Deployment、Pod、副本状态等

2.metrics-service主要关注的是资源度量API的实现,比如CPU、文件描述符、内存、请求延时等指标

授权认证

kube-apiserver-client访问证书

💥如果prometheus配置了kubernetes_sd_configs.tls_config.insecure_skip_verify: true,即禁用服务器证书认证,则无需这个步骤。

openssl genrsa -out prometheus.key 2048
openssl req -new -key prometheus.key -out prometheus.csr -subj "/CN=prometheus/O=it"
RequestStr=`cat prometheus.csr | base64 | tr -d "\n"`
# 提交申请到k8s
cat <<EOF | kubectl apply -f -
apiVersion: certificates.k8s.io/v1beta1
kind: CertificateSigningRequest
metadata:
  name: prometheus
spec:
  groups:
  - system:authenticated
  request: ${RequestStr}
  signerName: kubernetes.io/kube-apiserver-client
  usages:
  - client auth
EOF
# 如果不出错,可以审批
kubectl certificate approve prometheus
# 导出证书
kubectl get csr prometheus -o jsonpath='{.status.certificate}' | base64 --decode > prometheus.crt
# 检查证书有效期
openssl x509 -in prometheus.crt -noout -dates

保存好证书文件 prometheus.crt,prometheus配置所需

创建服务账户/集群角色/集群角色绑定对象

用于查询k8s资源。

kubectl create ns monitor

prometheus-rabc.yaml

apiVersion: v1
kind: ServiceAccount
metadata:
  name: prometheus
  namespace: monitor
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  name: prometheus
rules:
- apiGroups:
  - ""
  resources:
  - nodes
  - services
  - endpoints
  - pods
  - nodes/proxy
  verbs:
  - get
  - list
  - watch
- apiGroups:
  - "extensions"
  resources:
    - ingresses
  verbs:
  - get
  - list
  - watch
- apiGroups:
  - ""
  resources:
  - configmaps
  - nodes/metrics
  verbs:
  - get
- nonResourceURLs:
  - /metrics
  verbs:
  - get
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: prometheus
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: prometheus
subjects:
- kind: ServiceAccount
  name: prometheus
  namespace: monitor

因prometheus访问k8s需要拿到bearer_token,因此还需要获取服务账户prometheus的Tokens。

获取服务账户prometheus存储Token的secret对象名

➜   kubectl get sa prometheus -n monitor -o jsonpath={'.secrets[0].name'} | more
prometheus-token-2gljj

将Token进行解密

➜   kubectl get secret prometheus-token-2gljj -n monitor -o jsonpath={'.data.token'} | base64 -d > prometheus.bearer_token

保存好解密后的 prometheus.bearer_token,prometheus配置所需

安装Metrics-server

添加 metrics-server https://github.com/kubernetes-sigs/metrics-server#configuration

wget https://github.com/kubernetes-sigs/metrics-server/releases/latest/download/components.yaml
# 修改metrics-server容器参数部分,添加额外的启动参数(arg)
args:
  - --kubelet-preferred-address-types=InternalIP
  - --kubelet-insecure-tls
kubectl apply -f components.yaml
kubectl get --raw "/apis/metrics.k8s.io/v1beta1/nodes"

kubectl top 指令需要指标才能输出

安装kube-state-metrics

部署方式:点我

指标文档:https://github.com/kubernetes/kube-state-metrics/tree/master/docs

➜  standard git:(master) ✗ ls -l
total 20
-rw-r--r-- 1 zyh zyh  381 Mar 12 12:07 cluster-role-binding.yaml
-rw-r--r-- 1 zyh zyh 1744 Mar 12 12:07 cluster-role.yaml
-rw-r--r-- 1 zyh zyh 1134 Mar 12 17:29 deployment.yaml
-rw-r--r-- 1 zyh zyh  197 Mar 12 12:07 service-account.yaml
-rw-r--r-- 1 zyh zyh  410 Mar 12 12:07 service.yaml

需要注意的是,deployment.yaml中kube-state-metrics的容器地址是k8s.gcr.io/kube-state-metrics/kube-state-metrics,这个地址国内没法访问。因此需要替换为其它地址。

可以将项目clone下来,然后根据项目源代码里的Dockerfile来,自行build一个镜像。

ARG GOVERSION=1.17
ARG GOARCH=amd64
FROM golang:${GOVERSION} as builder
ARG GOARCH
ENV GOARCH=${GOARCH}
WORKDIR /go/src/k8s.io/kube-state-metrics/
COPY . /go/src/k8s.io/kube-state-metrics/

RUN make build-local

FROM gcr.io/distroless/static:latest-${GOARCH}
COPY --from=builder /go/src/k8s.io/kube-state-metrics/kube-state-metrics /

USER nobody

ENTRYPOINT ["/kube-state-metrics", "--port=8080", "--telemetry-port=8081"]

EXPOSE 8080 8081

GOARCH指定环境,其余无需修改。

根据情况修改 service 对象

集群外的prometheus无法直接访问集群内的kube-state-metrics服务,因此需要暴漏kube-state-metrics。

默认配置里,service对象是ClusterIP,仅针对了集群内部。

通过内网负载均衡器提供。

apiVersion: v1
kind: Service
metadata:
  labels:
    app.kubernetes.io/component: exporter
    app.kubernetes.io/name: kube-state-metrics
    app.kubernetes.io/version: 2.3.0
  name: kube-state-metrics
  namespace: kube-system
  annotations:
    service.beta.kubernetes.io/alibaba-cloud-loadbalancer-spec: slb.s1.small
    service.beta.kubernetes.io/alicloud-loadbalancer-address-type: intranet
    service.beta.kubernetes.io/alibaba-cloud-loadbalancer-additional-resource-tags: "project=cms-saas"
spec:
  type: LoadBalancer
  ports:
  - name: http-metrics
    port: 8080
    targetPort: http-metrics
  - name: telemetry
    port: 8081
    targetPort: telemetry
  selector:
    app.kubernetes.io/name: kube-state-metrics

安装

➜	kubectl apply -f .

检查服务是否正常

curl ip:8080/metrics

遇到的问题

性能不够(250m/512Mi,针对3000+ Job对象),导致接口数据返回过慢,引发prometheus数据采集失败。

配置Prometheus

https://prometheus.io/docs/prometheus/latest/configuration/configuration/#kubernetes_sd_config

通过kubernetes_sd_configs.role: node发现 kubernetes 的节点上的kubelet地址,并将其作为刮取数据源

通过kubernetes_sd_configs.role: endpoints发现 kubernetes service 对象的 endpoint 地址,从而找到 kube-state-metrics 的外部端点地址。

✨配置里需要将api_server替换成k8s的api_server服务地址.

  - job_name: 'k8s-cadvisor' # 抓容器, cadvisor被整合在kubelet中
    scheme: https
    metrics_path: /metrics
    tls_config:
      #ca_file: /etc/prometheus/conf/prometheus.crt
      insecure_skip_verify: true
    bearer_token_file: /etc/prometheus/conf/prometheus.bearer_token
    kubernetes_sd_configs:
    - role: node
      api_server: "https://<api_server_ip>:6443"
      tls_config:
        #ca_file: /etc/prometheus/conf/prometheus.crt
        insecure_skip_verify: true
      bearer_token_file: /etc/prometheus/conf/prometheus.bearer_token
    relabel_configs:
    - action: labelmap # 标签映射:抓取regex正则匹配的标签名,复制为replacement,期间标签值不变(replacement为$1时可以不写)
      regex: __meta_kubernetes_node_label_(.+)
      replacement: $1
    metric_relabel_configs:
    - source_labels: [instance]
      separator: ;
      regex: (.+)
      target_label: node
      replacement: $1
      action: replace
    - source_labels: [pod_name] # 兼容老集群,老集群曾用标签 pod_name
      separator: ;
      regex: (.+)
      target_label: pod
      replacement: $1
      action: replace
    - source_labels: [container_name] # 兼容老集群,老集群曾用标签 container_name
      separator: ;
      regex: (.+)
      target_label: container
      replacement: $1
      action: replace
      
  - job_name: 'kubernetes-apiservers'
    scheme: https
    tls_config:
      #ca_file: /etc/prometheus/conf/prometheus.crt
      insecure_skip_verify: true
    bearer_token_file: /etc/prometheus/conf/prometheus.bearer_token
    kubernetes_sd_configs:
    - role: endpoints
      api_server: "https://<api_server_ip>:6443"
      tls_config:
        #ca_file: /etc/prometheus/conf/prometheus.crt
        insecure_skip_verify: true
      bearer_token_file: /etc/prometheus/conf/prometheus.bearer_token
    relabel_configs:
    - source_labels: [__meta_kubernetes_namespace, __meta_kubernetes_service_name, __meta_kubernetes_endpoint_port_name]
      action: keep
      regex: default;kubernetes;https

  - job_name: "kubernetes-state-metrics" # 抓工作负载资源
    scheme: http
    tls_config:
      #ca_file: /etc/prometheus/conf/prometheus.crt
      insecure_skip_verify: true
    bearer_token_file: /etc/prometheus/conf/prometheus.bearer_token
    kubernetes_sd_configs:
    - role: endpoints
      api_server: "https://<api_server_ip>:6443"
      tls_config:
        #ca_file: /etc/prometheus/conf/prometheus.crt
        insecure_skip_verify: true
      bearer_token_file: /etc/prometheus/conf/prometheus.bearer_token
    relabel_configs:
    - source_labels: [__meta_kubernetes_service_name] # 将自动发现的endpoint只保留kube-state-metrics
      regex: kube-state-metrics
      replacement: $1
      action: keep
    metric_relabel_configs: # 排除部分kube-state-metrics采集项目
    - source_labels: [__name__]
      regex: "^go_.*|^kube_job_.*"  # 如果删除 job 会导致 job 对象监控失败.
      action: drop
    - source_labels: [pod]
      regex: "^无需监控的Pod名称前缀-.*"
      action: drop

添加grafana模板

https://grafana.com/grafana/dashboards/13105

image-20210313180815651

上图是我在阿里云ASK中测试的最终效果图

调整模板

如果你发现模板中,【网络带宽】之类的没有数据,则需要看下此板块SQL里的标签过滤是否有误。

在我的环境里,模板中过滤的 name=‘^k8s_.*’ 无法匹配到Pod。因此删除SQL里的这个过滤即可。

添加预警

需要明确,我们需要监控什么

  1. k8s本身各组件状态
  2. 调度了多少个replicas?现在可用的有几个?
  3. Pod是否启动成功
  4. Pod重启速率?
groups:
- name: kubernetes
  rules:
  - alert: kube endpoint down
    expr: (up{job="kube-state-metrics"} or up{job="kubernetes-apiservers"}) != 1  #0不正常,1正常
    for: 5m  #持续时间 ,表示持续5分钟获取不到信息,则触发报警
    labels:
      severity: error
      cluster: k8s
    annotations:
      summary: "Instance:{{ $labels.instance }}, Job {{ $labels.job }} stop "
      sourcedata: '{{ $labels.job }}'
  - alert: JobFailed
    expr: kube_job_status_failed == 1
    for: 5m
    labels:
      severity: error
      cluster: k8s
    annotations:
      summary: 'Namespace: {{ $labels.namespace }}, Job: {{ $labels.job }} run failed.'
      sourcedata: '{{ $labels.job }}'
  - alert: PodReady
    expr: kube_pod_container_status_ready != 1
    for: 5m   #Ready持续5分钟,说明启动有问题
    labels:
      severity: warning
      cluster: k8s
    annotations:
      summary: 'Namespace: {{ $labels.namespace }}, Pod: {{ $labels.pod }} not ready for 5m'
      sourcedata: '{{ $labels.job }}'
  - alert: PodRestart
    expr: kube_pod_container_status_last_terminated_reason == 1 and on(container) rate(kube_pod_container_status_restarts_total[5m]) * 300 > 1
    for: 5s
    labels:
      severity: error
      cluster: k8s
    annotations:
      summary: 'namespace: {{ $labels.namespace }}, pod: {{ $labels.pod }} restart rate > {{ $value }}'
      sourcedata: '{{ $labels.job }}'
  - alert: Deployment_replicas_available_num_low
    expr: kube_deployment_status_replicas - kube_deployment_status_replicas_available > 0
    for: 1m
    labels:
      severity: warning
      cluster: k8s
    annotations:
      summary: 'namespace: {{ $labels.namespace }}, deployment: {{ $labels.deployment }} available low'
      sourcedata: '{{ $labels.job }}'
  - alert: Deployment_replicas_unavailable
    expr: kube_deployment_status_replicas > 0 and kube_deployment_status_replicas_available == 0
    for: 1m
    labels:
      severity: error
      cluster: k8s
    annotations:
      summary: 'namespace: {{ $labels.namespace }}, deployment: {{ $labels.deployment }} unavailable'
      sourcedata: '{{ $labels.job }}'