使用Prometheus Operator监控kubetnetes集群

一、简介

Prometheus

Prometheus 是一个开源监控系统，它本身已经成为了云原生中指标监控的事实标准，几乎所有 Kubernetes 的核心组件以及其它云原生系统都以 Prometheus 的指标格式输出自己的运行时监控信息。

主要特性：

• 使用指标名称及键值对标识的多维度数据模型。
• 采用弹性查询语言PromQL。
• 不依赖分布式存储，为自治的单点服务。
• 使用http完成对监控数据的拉取。
• 通过网关支持时序数据的推送。
• 支持多种图形和Dashboard的展示。

另外在Prometheus的整个生态系统中有各种可选组件，用于功能的扩充。

Prometheus-Operator

CoreOS提供了一种名为Operator的管理工具，他是管理特定应用程序的控制器。通过扩展Kubernetes API以软件的方式帮助用户创建、配置和管理复杂的或又状态的应用程序实例（如etcd、Redis、MySQL、Prometheus等）。

它通过Kubernetes的CRD（Custom Resource Definition，自定义资源定义）对Prometheus和Prometheus需要监控的服务进行部署和配置。

Prometheus-Operator使用下面两种资源来配置Prometheus及其要监控的服务。

• Prometheus：为Prometheus的deployment。
• ServiceMonitor：用于描述Prometheus监控的服务。

首先我们先来了解下 Prometheus-Operator 的架构图：

上图是 Prometheus-Operator 官方提供的架构图，其中 Operator 是最核心的部分，作为一个控制器，他会去创建 Prometheus 、 ServiceMonitor 、 AlertManager 以及 PrometheusRule 4个 CRD 资源对象，然后会一直监控并维持这4个资源对象的状态。

其中创建的 prometheus 这种资源对象就是作为 Prometheus Server 存在，而 ServiceMonitor 就是 exporter 的各种抽象， exporter是用来提供专门提供 metrics 数据接口的工具， Prometheus 就是通过 ServiceMonitor 提供的 metrics 数据接口去 pull 数据的。

当然 alertmanager 这种资源对象就是对应的 AlertManager 的抽象，而 PrometheusRule 是用来被 Prometheus 实例使用的报警规则文件。

二、说明

• 本文档基于Prometheus Operator对Prometheus监控系统而进行，完整的配置文件请参考https://github.com/coreos/prometheus-operator

• 本文使用的k8s集群是基于kubeadm搭建的，具体搭建的详细请参考https://blog.51cto.com/billy98/2350660

三、部署

1. 环境准备

配置Prometheus-Operator之前需要先准备以下几个环境：

• Helm环境：请参考https://blog.51cto.com/billy98/2338415
• gluster-heketi环境：因为prometheus和alertmanager都是有状态的statefulsets集群，所以需要使用到公共存储。请参考https://blog.51cto.com/billy98/2337874
• Ingress环境：请参考https://blog.51cto.com/billy98/2337874

2. 配置域名解析

将alert.cnlinux.club、grafana.cnlinux.club、prom.cnlinux.club三个域名的A记录解析到负责均衡的IP10.31.90.200。

3. 修改scheduler、controller-manager监听地址

修改/etc/kubernetes/manifests/ 目录下kube-controller-manager.yaml和kube-scheduler.yaml
将监听地址改成--address=0.0.0.0，重启kubelet服务

systemctl restart kubelet.service 

4. 创建Namespace

kubectl create ns monitoring

5. 创建Secret

因为etcd是使用https访问的，所以prometheus的容器中也必须要etcd的证书去监控etcd集群，创建Secret就是将证书挂载到prometheus容器中,后续还需要在Prometheus-Operator的配置文件中使用此Secret。

kubectl  -n monitoring create secret generic etcd-certs --from-file=/etc/kubernetes/pki/etcd/healthcheck-clien t.crt --from-file=/etc/kubernetes/pki/etcd/healthcheck-client.key --from-file=/etc/kubernetes/pki/etcd/ca.crt 

6. 配置Prometheus-Operator

1）首先使用helm下载Prometheus-Operator文件包。

helm fetch stable/prometheus-operator

2）解压下载的压缩包

tar zxf prometheus-operator-1.8.0.tgz

并修改prometheus-operator目录下的values.yaml。

具体修改的配置如下(配置过多，其他未修改的就不再展示了)：

nameOverride: "p"
alertmanager:
  ingress:
    enabled: true
    annotations: 
      kubernetes.io/ingress.class: nginx
    labels: {}
    hosts: 
      - alert.cnlinux.club
    tls: []
  alertmanagerSpec:
    storage: 
      volumeClaimTemplate:
        spec:
          storageClassName: gluster-heketi
          accessModes: ["ReadWriteOnce"]
          resources:
            requests:
              storage: 20Gi
        selector: {}
grafana:
  enabled: true
  adminPassword: admin
    #grafana登录密码
  ingress:
    enabled: true
    annotations: 
      kubernetes.io/ingress.class: nginx
    labels: {}
    hosts: 
      - grafana.cnlinux.club
kubeApiServer:
  enabled: true
  tlsConfig:
    serverName: kubernetes
    insecureSkipVerify: true
  serviceMonitor:
    jobLabel: component
    selector:
      matchLabels:
        component: apiserver
        provider: kubernetes
kubelet:
  enabled: true
  namespace: kube-system
  serviceMonitor:
    https: true
kubeControllerManager:
  enabled: true
  endpoints: []
  service:
    port: 10252
    targetPort: 10252
    selector:
      component: kube-controller-manager
coreDns:
  enabled: true
  service:
    port: 9153
    targetPort: 9153
    selector:
      k8s-app: kube-dns
kubeEtcd:
  enabled: true
  endpoints: []
  service:
    port: 2379
    targetPort: 2379
    selector:
      component: etcd
  serviceMonitor:
    scheme: https
    insecureSkipVerify: false
    serverName: ""
    caFile: /etc/prometheus/secrets/etcd-certs/ca.crt
    certFile: /etc/prometheus/secrets/etcd-certs/healthcheck-client.crt
    keyFile: /etc/prometheus/secrets/etcd-certs/healthcheck-client.key
    #secret etcd-certs挂载在prometheus的路径是/etc/prometheus/secrets/etcd-certs，证书文件名和secret一样
kubeScheduler:
  enabled: true
  endpoints: []
  service:
    port: 10251

    targetPort: 10251
    selector:
      component: kube-scheduler
prometheus:
  ingress:
    enabled: true
    annotations: 
      kubernetes.io/ingress.class: nginx
    labels: {}
    hosts: 
      - prom.cnlinux.club
  prometheusSpec:
    secrets: [etcd-certs]
    #上面步骤创建etcd证书的secret
    storageSpec: 
      volumeClaimTemplate:
        spec:
          storageClassName: gluster-heketi
          accessModes: ["ReadWriteOnce"]
          resources:
            requests:
              storage: 20Gi
        selector: {}

3) 安装prometheus-operator

[root@node-01 ~]# helm install --name p --namespace monitoring ./prometheus-operator
NAME:   p
LAST DEPLOYED: Tue Feb 26 14:30:52 2019
NAMESPACE: monitoring
STATUS: DEPLOYED

RESOURCES:
==> v1beta1/DaemonSet
NAME                        DESIRED  CURRENT  READY  UP-TO-DATE  AVAILABLE  NODE SELECTOR  AGE
p-prometheus-node-exporter  6        6        1      6           1          <none>         5s

==> v1beta2/Deployment
NAME       DESIRED  CURRENT  UP-TO-DATE  AVAILABLE  AGE
p-grafana  1        1        1           0          5s

==> v1/PrometheusRule
NAME                                    AGE
p-alertmanager.rules                    4s
p-etcd                                  4s
p-general.rules                         4s
p-k8s.rules                             4s
p-kube-apiserver.rules                  4s
p-kube-prometheus-node-alerting.rules   4s
p-kube-prometheus-node-recording.rules  4s
p-kube-scheduler.rules                  4s
p-kubernetes-absent                     4s
p-kubernetes-apps                       4s
p-kubernetes-resources                  4s
p-kubernetes-storage                    4s
p-kubernetes-system                     4s
p-node.rules                            4s
p-prometheus-operator                   4s
p-prometheus.rules                      4s

==> v1/Pod(related)
NAME                                  READY  STATUS             RESTARTS  AGE
p-prometheus-node-exporter-48lw9      0/1    Running            0         5s
p-prometheus-node-exporter-7lpvx      0/1    Running            0         5s
p-prometheus-node-exporter-8q577      1/1    Running            0         5s
p-prometheus-node-exporter-ls8cx      0/1    Running            0         5s
p-prometheus-node-exporter-nbl2g      0/1    Running            0         5s
p-prometheus-node-exporter-v7tb5      0/1    Running            0         5s
p-grafana-fcf4dc6bb-9c6pg             0/3    ContainerCreating  0         5s
p-kube-state-metrics-57d788d69-vmh42  0/1    Running            0         5s
p-operator-666b958c4f-wvd4h           1/1    Running            0         5s

==> v1beta1/ClusterRole
NAME                            AGE
p-kube-state-metrics            6s
psp-p-prometheus-node-exporter  6s

==> v1/Service
NAME                        TYPE       CLUSTER-IP      EXTERNAL-IP  PORT(S)    AGE
p-grafana                   ClusterIP  10.245.103.159  <none>       80/TCP     5s
p-kube-state-metrics        ClusterIP  10.245.150.181  <none>       8080/TCP   5s
p-prometheus-node-exporter  ClusterIP  10.245.98.70    <none>       9100/TCP   5s
p-alertmanager              ClusterIP  10.245.10.5     <none>       9093/TCP   5s
p-coredns                   ClusterIP  None            <none>       9153/TCP   5s
p-kube-controller-manager   ClusterIP  None            <none>       10252/TCP  5s
p-kube-etcd                 ClusterIP  None            <none>       2379/TCP   5s
p-kube-scheduler            ClusterIP  None            <none>       10251/TCP  5s
p-operator                  ClusterIP  10.245.31.238   <none>       8080/TCP   5s
p-prometheus                ClusterIP  10.245.109.85   <none>       9090/TCP   5s

==> v1/ClusterRoleBinding
NAME                          AGE
p-grafana-clusterrolebinding  6s
p-alertmanager                6s
p-operator                    6s
p-operator-psp                6s
p-prometheus                  6s
p-prometheus-psp              6s

==> v1beta1/ClusterRoleBinding
NAME                            AGE
p-kube-state-metrics            6s
psp-p-prometheus-node-exporter  6s

==> v1beta1/Role
NAME       AGE
p-grafana  6s

==> v1/RoleBinding
NAME                 AGE
p-prometheus-config  5s
p-prometheus         4s
p-prometheus         4s

==> v1/Deployment
NAME        DESIRED  CURRENT  UP-TO-DATE  AVAILABLE  AGE
p-operator  1        1        1           1          5s

==> v1/Alertmanager
NAME            AGE
p-alertmanager  5s

==> v1/Secret
NAME                         TYPE    DATA  AGE
p-grafana                    Opaque  3     6s
alertmanager-p-alertmanager  Opaque  1     6s

==> v1/ServiceAccount
NAME                        SECRETS  AGE
p-grafana                   1        6s
p-kube-state-metrics        1        6s
p-prometheus-node-exporter  1        6s
p-alertmanager              1        6s
p-operator                  1        6s
p-prometheus                1        6s

==> v1beta1/Deployment
NAME                  DESIRED  CURRENT  UP-TO-DATE  AVAILABLE  AGE
p-kube-state-metrics  1        1        1           0          5s

==> v1beta1/Ingress
NAME            HOSTS                 ADDRESS  PORTS  AGE
p-grafana       grafana.cnlinux.club  80       5s
p-alertmanager  alert.cnlinux.club    80       5s
p-prometheus    prom.cnlinux.club     80       5s

==> v1beta1/PodSecurityPolicy
NAME                        PRIV   CAPS      SELINUX   RUNASUSER  FSGROUP    SUPGROUP  READONLYROOTFS  VOLUMES
p-grafana                   false  RunAsAny  RunAsAny  RunAsAny   RunAsAny   false     configMap,emptyDir,projected,secret,downwardAPI,persistentVolumeClaim
p-prometheus-node-exporter  false  RunAsAny  RunAsAny  MustRunAs  MustRunAs  false     configMap,emptyDir,projected,secret,downwardAPI,persistentVolumeClaim,hostPath
p-alertmanager              false  RunAsAny  RunAsAny  MustRunAs  MustRunAs  false     configMap,emptyDir,projected,secret,downwardAPI,persistentVolumeClaim
p-operator                  false  RunAsAny  RunAsAny  MustRunAs  MustRunAs  false     configMap,emptyDir,projected,secret,downwardAPI,persistentVolumeClaim
p-prometheus                false  RunAsAny  RunAsAny  MustRunAs  MustRunAs  false     configMap,emptyDir,projected,secret,downwardAPI,persistentVolumeClaim

==> v1/ConfigMap
NAME                         DATA  AGE
p-grafana-config-dashboards  1     6s
p-grafana                    1     6s
p-grafana-datasource         1     6s
p-etcd                       1     6s
p-grafana-coredns-k8s        1     6s
p-k8s-cluster-rsrc-use       1     6s
p-k8s-node-rsrc-use          1     6s
p-k8s-resources-cluster      1     6s
p-k8s-resources-namespace    1     6s
p-k8s-resources-pod          1     6s
p-nodes                      1     6s
p-persistentvolumesusage     1     6s
p-pods                       1     6s
p-statefulset                1     6s

==> v1beta1/RoleBinding
NAME       AGE
p-grafana  5s

==> v1/Prometheus
NAME          AGE
p-prometheus  4s

==> v1/ServiceMonitor
NAME                       AGE
p-alertmanager             4s
p-coredns                  4s
p-apiserver                4s
p-kube-controller-manager  4s
p-kube-etcd                4s
p-kube-scheduler           4s
p-kube-state-metrics       4s
p-kubelet                  4s
p-node-exporter            4s
p-operator                 4s
p-prometheus               4s

==> v1/ClusterRole
NAME                   AGE
p-grafana-clusterrole  6s
p-alertmanager         6s
p-operator             6s
p-operator-psp         6s
p-prometheus           6s
p-prometheus-psp       6s

==> v1/Role
NAME                 AGE
p-prometheus-config  6s
p-prometheus         4s
p-prometheus         4s

NOTES:
The Prometheus Operator has been installed. Check its status by running:
  kubectl --namespace monitoring get pods -l "release=p"

Visit https://github.com/coreos/prometheus-operator for instructions on how
to create & configure Alertmanager and Prometheus instances using the Operator.

6. 部署中遇到的问题

在部署中有几个坑，我在此列举一下，大家配置的时候需要注意一下

1）名称过长导致pvc创建失败

由于alertmanager和prometheus都是有状态的statefulsets，所以我们使用了gluster的存储，并通过 prometheus-operator自动创建pvc，如果charts的release 名称过长会导致pvc创建失败。
所以在上面的安装中指定了release的名字为phelm install --name p --namespace monitoring ./prometheus-operator,并且在配置文件中也修改了namenameOverride: "p"。

Warning    ProvisioningFailed  3s (x2 over 40s)  persistentvolume-controller  Failed to provision volume with StorageClass "gluster-heketi": failed to create volume: failed to create endpoint/service
default/glusterfs-dynamic-72488422-3428-11e9-a74b-005056824bdc: failed to create endpoint: Endpoints "glusterfs-dynamic-72488422-3428-11e9-a74b-005056824bdc" is invalid:
metadata.labels: Invalid value: "alertmanager-prom-alertmanager-db-alertmanager-prom-alertmanager-0": must be no more than 63 characters

2）配置文件中labels

首先要查看pod的标签，然后修改修改prometheus-operator目录下的values.yaml对应的标签。

[root@node-01 ~]# kubectl -n kube-system get pod --show-labels
NAME                                        READY   STATUS    RESTARTS   AGE     LABELS
coredns-7f65654f74-6gxps                    1/1     Running   8          5d22h   k8s-app=kube-dns,pod-template-hash=7f65654f74
etcd-node-01                                1/1     Running   1          32d     component=etcd,tier=control-plane
kube-controller-manager-node-01             1/1     Running   0          39h     component=kube-controller-manager,tier=control-plane
kube-scheduler-node-01                      1/1     Running   0          23h     component=kube-scheduler,tier=control-plane
...

需要注意的是一定要修改prometheus-operator目录下values.yaml对应的标签，不能在安装的时候指定外部的配置文件来覆盖labels值，这可能是个bug，指定外部配置时无法覆盖labels而是追加，会导致prometheus无法抓取到数据。

7. 更新和删除

如果在修改了配置文件values.yaml，可以使用以下命令更新prometheus-operator

helm upgrade RELEASE_NAME ./prometheus-operator

如果需要删除，可以使用以下命令

helm del --purge RELEASE_NAME

kubectl -n monitoring delete crd prometheuses.monitoring.coreos.com
kubectl -n monitoring delete crd prometheusrules.monitoring.coreos.com
kubectl -n monitoring delete crd servicemonitors.monitoring.coreos.com
kubectl -n monitoring delete crd alertmanagers.monitoring.coreos.com

8. 验证

部署完后可以在浏览器访问prometheushttp://prom.cnlinux.club/targets,可以看到如下图，所有的项都有数据，并且是UP状态的。

浏览器访问grafana http://grafana.cnlinux.club/，可以看到各种资源的监控图。
用户名为admin，密码为values.yaml配置文件中指定的密码

浏览器访问Alertmanager
http://alert.cnlinux.club/，可以看到报警项。

至此所有的安装完成，下一篇将详细说明使用prometheus监控自定义服务，以及报警设置。如有问题欢迎在下面留言交流。希望大家多多关注和点赞，谢谢！

转载于:https://blog.51cto.com/billy98/2355587