如何利用Prometheus Operator实现自定义指标监控,很多新手对此不是很清楚,为了帮助大家解决这个难题,下面小编将为大家详细讲解,有这方面需求的人可以来学习下,希望你能有所收获。
创新互联-专业网站定制、快速模板网站建设、高性价比全州网站开发、企业建站全套包干低至880元,成熟完善的模板库,直接使用。一站式全州网站制作公司更省心,省钱,快速模板网站建设找我们,业务覆盖全州地区。费用合理售后完善,10余年实体公司更值得信赖。
在过去的文章中,我们花了相当大的篇幅来聊关于监控的话题。这是因为当你正在管理Kubernetes集群时,一切都会以极快的速度发生变化。因此有一个工具来监控集群的健康状态和资源指标极为重要。
在Rancher 2.5中,我们引入了基于Prometheus Operator的新版监控,它可以提供Prometheus以及相关监控组件的原生Kubernetes部署和管理。Prometheus Operator可以让你监控集群节点、Kubernetes组件和应用程序工作负载的状态和进程。同时,它还能够通过Prometheus收集的指标来定义告警并且创建自定义仪表盘,通过Grafana可以轻松地可视化收集到的指标。
新版本的监控也采用prometheus-adapter,开发人员可以利用其基于自定义指标和HPA扩展他们的工作负载。
我们将探索如何利用Prometheus Operator来抓取自定义指标并利用这些指标进行高级工作负载管理。
安装Prometheus
在Rancher 2.5中安装Prometheus极为简单。仅需访问Cluster Explorer -> Apps并安装rancher-monitoring即可。
你需要了解以下默认设置:
prometheus-adapter
将会作为chart安装的一部分启用ServiceMonitorNamespaceSelector
留为空,允许 Prometheus 在所有命名空间中收集 ServiceMonitors
安装完成后,我们可以从Cluster Explorer访问监控组件。
部署工作负载
现在让我们部署一个从应用层暴露自定义指标的示例工作负载。该工作负载暴露了一个简单的应用程序,该应用程序已经使用Prometheus client_golang库进行了检测,并在/metric
端点上提供了一些自定义指标。
它有两个指标:
http_requests_total
http_request_duration_seconds
以下manifest部署了工作负载、相关服务以及访问该工作负载的ingress:
apiVersion: apps/v1 kind: Deployment metadata: labels: app.kubernetes.io/name: prometheus-example-app name: prometheus-example-app spec: replicas: 1 selector: matchLabels: app.kubernetes.io/name: prometheus-example-app template: metadata: labels: app.kubernetes.io/name: prometheus-example-app spec: containers: - name: prometheus-example-app image: gmehta3/demo-app:metrics ports: - name: web containerPort: 8080 --- apiVersion: v1 kind: Service metadata: name: prometheus-example-app labels: app.kubernetes.io/name: prometheus-example-app spec: selector: app.kubernetes.io/name: prometheus-example-app ports: - protocol: TCP port: 8080 targetPort: 8080 name: web --- apiVersion: networking.k8s.io/v1beta1 kind: Ingress metadata: name: prometheus-example-app spec: rules: - host: hpa.demo http: paths: - path: / backend: serviceName: prometheus-example-app servicePort: 8080
部署ServiceMonitor
ServiceMonitor是一个自定义资源定义(CRD),可以让我们声明性地定义如何监控一组动态服务。
你可以访问以下链接查看完整的ServiceMonitor规范:
https://github.com/prometheus-operator/prometheus-operator/blob/master/Documentation/api.md#servicemonitor
现在,我们来部署ServiceMonitor,Prometheus用它来收集组成prometheus-example-app Kubernetes服务的pod。
kind: ServiceMonitor metadata: name: prometheus-example-app spec: selector: matchLabels: app.kubernetes.io/name: prometheus-example-app endpoints: - port: web
如你所见,现在用户可以在Rancher监控中浏览ServiceMonitor。
不久之后,新的service monitor和服务相关联的pod应该会反映在Prometheus服务发现中。
我们也能够在Prometheus中看到指标。
部署Grafana仪表盘
在Rancher 2.5中,监控可以让用户将Grafana仪表盘存储为cattle-dashboards
命名空间中的ConfigMaps。
用户或集群管理员现在可以在这一命名空间中添加更多的仪表盘以扩展Grafana的自定义仪表盘。
Dashboard ConfigMap Example
apiVersion: v1 kind: ConfigMap metadata: name: prometheus-example-app-dashboard namespace: cattle-dashboards labels: grafana_dashboard: "1" data: prometheus-example-app.json: | { "annotations": { "list": [ { "builtIn": 1, "datasource": "-- Grafana --", "enable": true, "hide": true, "iconColor": "rgba(0, 211, 255, 1)", "name": "Annotations & Alerts", "type": "dashboard" } ] }, "editable": true, "gnetId": null, "graphTooltip": 0, "links": [], "panels": [ { "aliasColors": {}, "bars": false, "dashLength": 10, "dashes": false, "datasource": null, "fieldConfig": { "defaults": { "custom": {} }, "overrides": [] }, "fill": 1, "fillGradient": 0, "gridPos": { "h": 9, "w": 12, "x": 0, "y": 0 }, "hiddenSeries": false, "id": 2, "legend": { "avg": false, "current": false, "max": false, "min": false, "show": true, "total": false, "values": false }, "lines": true, "linewidth": 1, "nullPointMode": "null", "percentage": false, "pluginVersion": "7.1.5", "pointradius": 2, "points": false, "renderer": "flot", "seriesOverrides": [], "spaceLength": 10, "stack": false, "steppedLine": false, "targets": [ { "expr": "rate(http_requests_total{code=\"200\",service=\"prometheus-example-app\"}[5m])", "instant": false, "interval": "", "legendFormat": "", "refId": "A" } ], "thresholds": [], "timeFrom": null, "timeRegions": [], "timeShift": null, "title": "http_requests_total_200", "tooltip": { "shared": true, "sort": 0, "value_type": "individual" }, "type": "graph", "xaxis": { "buckets": null, "mode": "time", "name": null, "show": true, "values": [] }, "yaxes": [ { "format": "short", "label": null, "logBase": 1, "max": null, "min": null, "show": true }, { "format": "short", "label": null, "logBase": 1, "max": null, "min": null, "show": true } ], "yaxis": { "align": false, "alignLevel": null } }, { "aliasColors": {}, "bars": false, "dashLength": 10, "dashes": false, "datasource": null, "description": "", "fieldConfig": { "defaults": { "custom": {} }, "overrides": [] }, "fill": 1, "fillGradient": 0, "gridPos": { "h": 8, "w": 12, "x": 0, "y": 9 }, "hiddenSeries": false, "id": 4, "legend": { "avg": false, "current": false, "max": false, "min": false, "show": true, "total": false, "values": false }, "lines": true, "linewidth": 1, "nullPointMode": "null", "percentage": false, "pluginVersion": "7.1.5", "pointradius": 2, "points": false, "renderer": "flot", "seriesOverrides": [], "spaceLength": 10, "stack": false, "steppedLine": false, "targets": [ { "expr": "rate(http_requests_total{code!=\"200\",service=\"prometheus-example-app\"}[5m])", "interval": "", "legendFormat": "", "refId": "A" } ], "thresholds": [], "timeFrom": null, "timeRegions": [], "timeShift": null, "title": "http_requests_total_not_200", "tooltip": { "shared": true, "sort": 0, "value_type": "individual" }, "type": "graph", "xaxis": { "buckets": null, "mode": "time", "name": null, "show": true, "values": [] }, "yaxes": [ { "format": "short", "label": null, "logBase": 1, "max": null, "min": null, "show": true }, { "format": "short", "label": null, "logBase": 1, "max": null, "min": null, "show": true } ], "yaxis": { "align": false, "alignLevel": null } } ], "schemaVersion": 26, "style": "dark", "tags": [], "templating": { "list": [] }, "time": { "from": "now-15m", "to": "now" }, "timepicker": { "refresh_intervals": [ "5s", "10s", "30s", "1m", "5m", "15m", "30m", "1h", "2h", "1d" ] }, "timezone": "", "title": "prometheus example app", "version": 1 }
现在,用户应该能够在Grafana中访问prometheus example app的仪表盘。
自定义指标的HPA
这一部分假设你已经将prometheus-adapter
作为监控的一部分安装完毕了。实际上,在默认情况下,监控安装程序会安装prometheus-adapter。
用户现在可以创建一个HPA spec,如下所示:
apiVersion: autoscaling/v2beta2 kind: HorizontalPodAutoscaler metadata: name: prometheus-example-app-hpa spec: scaleTargetRef: apiVersion: apps/v1 kind: Deployment name: prometheus-example-app minReplicas: 1 maxReplicas: 5 metrics: - type: Object object: describedObject: kind: Service name: prometheus-example-app metric: name: http_requests target: averageValue: "5" type: AverageValue
你可以查看以下链接获取关于HPA的更多信息:
https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/
我们将使用自定义的http_requests_total指标来执行pod自动伸缩。
现在我们可以生成一个样本负载来查看HPA的运行情况。我可以使用hey
进行同样的操作。
hey -c 10 -n 5000 http://hpa.demo
开发人员和集群管理员可以利用该堆栈来监控它们的工作负载,部署可视化,并利用Kubernetes内可用的高级工作负载管理功能。
看完上述内容是否对您有帮助呢?如果还想对相关知识有进一步的了解或阅读更多相关文章,请关注创新互联行业资讯频道,感谢您对创新互联的支持。
分享名称:如何利用PrometheusOperator实现自定义指标监控
转载来于:http://scyingshan.cn/article/psogdj.html