# 实践：接收Prometheus的告警

本节我们将写一个程序，来接收Prometheus发送的告警，并将告警打印出来。

首先，我们安装好Prometheus、NodeExporter。Prometheus的配置prometheus.yml如下：

```
global:
  scrape_interval: 15s 
  evaluation_interval: 15s

rule_files:
  - "/usr/local/prometheus/rule_files/node.yml"

scrape_configs:
  - job_name: 'node-exporter'
    static_configs:
    - targets: ['192.168.2.101:9100']
      labels:
        IP: '192.168.2.101'

alerting:
  alertmanagers:
  - static_configs:
    - targets: [ '192.168.2.101:19093' ]
```

告警规则文件`node.yml`内容如下：

```
groups:
- name: Node
  rules:
  - alert: NodeCpuPressure
    expr: 100 * (1 - avg(irate(node_cpu_seconds_total{mode="idle"}[2m])) by(IP)) > 80
    for: 2m
    annotations:
      summary: "NodeCpuPressure, IP: {{$labels.IP}}, Value: {{$value}}%, Threshold: 80%"
```

在上面的prometheus.yml中，我们配置的alertmanager的地址为`192.168.2.101:19093`，这个地址就是接下来我们要启动的自已的程序，用来接收prometheus的告警信息并打印出来。

在主机上的某个目录下创建文件alertmanager.go，文件内容见本文附录，然后执行命令`go run alertmanager.go`将程序跑起来。

接着，我们启动Prometheus与NodeExporter。

然后，我们拉高主机CPU的使用率，不一会，我们便可以在Prometheus的Alert页面看到一个告警处理Pending状态

![](https://735702182-files.gitbook.io/~/files/v0/b/gitbook-legacy-files/o/assets%2F-M69DxRoC8HFhJcA2bUu%2Fsync%2F27b7482992cf82bc19e9d0ef79d60a7f61dd048e.png?generation=1592397053901885\&alt=media)

再等一会，等告警Firing之后，我们自已的程序会接收到该告警并打印出来：可以发现，`startsAt`就是上面截图中的`Active Since`的时间，也就是Alert变成Pending的时间。`endsAt`刚好是`startAt`加了三分钟（这个暂时不清楚为什么是三分钟）。

```
$ go run alertmanager.go
2020-02-17 20:48:58.411459407 +0800 CST m=+191.432028937
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:51:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]
```

让CPU一直保持在80%以上，我们会发现Prometheus会持续地向AlertManager发送Alert。我们发现，发送的时间间隔为75秒（为什么是75秒还不清楚，很有可能是`evaluation_interval`的五倍），`startsAt`一直没变，而`endsAt`一直在增加，增加的时间为Prometheus发送该告警的时间再加3m。

```
2020-02-17 20:48:58.411459407 +0800 CST m=+191.432028937
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:51:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:50:13.405880837 +0800 CST m=+266.426450346
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:53:13.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:51:28.405287147 +0800 CST m=+341.425856660
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:54:28.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:52:43.40958641 +0800 CST m=+416.430155922
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:55:43.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:53:58.405264448 +0800 CST m=+491.425833979
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:56:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:55:13.406017704 +0800 CST m=+22.465382727
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:58:13.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:56:28.405989243 +0800 CST m=+97.465354315
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 100%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:59:28.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]
```

接着，我们把CPU降到80%以下（时间大概在`20:56:28`到`20:56:58`之间）。我们会发现，`20:56:28`时发送的告警（上面的最后一条），其`endsAt`的时间为`20:59:28`，说明发送该条告警时，CPU还处于80%以上。而过了30秒后，Promethus又发送了一条告警（下面第一条），这一次没有等到75秒，是因为发送该条告警时CPU已经低于80%了。

这里我们就有两个疑问了：（1）我们如何判断发送下面第一条告警时CPU已经低于80%了呢？（2）下面第一条告警中，CPU的值显示的是81.76666666656578%呀？

其实原因是很简单：我们配置了`evaluation_interval`的值为15秒，Prometheus在上面最后一条告警（`20:56:28`）后，在`20:56:43`和`20:56:58`都会做一次Evaluation，在`20:56:43`这一次计算出来的CPU的值为81.76666666656578%，由于上一条告警还没有过75秒，所以不发送。在`20:56:58`时该计算出来的CPU已经小于80%，相当于告警已经解除，于是Prometheus会立即发送一条告警出去，表明告警已解除。也就是说，下面第一条告警其实是一条“解除告警”，为什么呢？因为`endsAt`的时间就是发送该条告警的时间，当AlertManager接收到以后，发现这个时间已经是一个过去的时间了，也就是说，这条告警已经结束了。既然已经通过`endsAt`标识该告警已经过时了，那么CPU的值也就不重要了，不过Prometheus用了“解除告警”的上一次计算出来的值还有让人有点容易混淆。

```
2020-02-17 20:56:58.405968564 +0800 CST m=+127.465333664
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 81.76666666656578%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:56:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:58:13.408337229 +0800 CST m=+202.467702513
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 81.76666666656578%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:56:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 20:59:28.406675061 +0800 CST m=+277.466040229
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 81.76666666656578%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:56:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 21:00:43.40653634 +0800 CST m=+352.465901514
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 81.76666666656578%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:56:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]

2020-02-17 21:01:58.405259946 +0800 CST m=+427.464625058
[{"labels":{"IP":"192.168.2.101","alertname":"NodeCpuPressure"},"annotations":{"summary":"NodeCpuPressure, IP: 192.168.2.101, Value: 81.76666666656578%, Threshold: 80%"},"startsAt":"2020-02-17T20:48:58.402147666+08:00","endsAt":"2020-02-17T20:56:58.402147666+08:00","generatorURL":"http://peng01:9090/graph?g0.expr=100+%2A+%281+-+avg+by%28IP%29+%28irate%28node_cpu_seconds_total%7Bmode%3D%22idle%22%7D%5B2m%5D%29%29%29+%3E+80\u0026g0.tab=1"}]
```

另外，根据上面的输出我们还可以发现，当“解除告警”发出去以后，Prometheus还坚持把“解除告警”发送了好多次，而且时间间隔也为75秒。在实验中我们发现大概发送了十几次后，Prometheus就不再发送了。

## 附录

alertmanager.go的代码

```
package main

import (
    "time"
    "io/ioutil"
    "net/http"
    "fmt"
)

type MyHandler struct{}

func (mh *MyHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
    body, err := ioutil.ReadAll(r.Body)
    if err != nil {
        fmt.Printf("read body err, %v\n", err)
        return
    }
    fmt.Println(time.Now())
    fmt.Printf("%s\n\n", string(body))
}

func main() {
    http.Handle("/api/v1/alerts", &MyHandler{})
    http.ListenAndServe(":19093", nil)
}
```