本篇内容有点长，代码有点多。有兴趣的可以坚持看下去，并动手实践，没兴趣的可以划走。本文分两大块，一是搞清楚prometheus四种类型的指标Counter，Gauge，Histogram，Summary用golang语言如何构造这4种类型对应的指标，二是搞清楚修改指标值的场景和方式。
| 指标类型 | 类别 | 描述 | 可应用场景 |
| Counter | 计数类 | 使用在累计指标单调递增或递减情况下，只能在目标重启后自动归零 | 服务请求处理数量、已完成任务数量、错误数量 |
| Guage | 测量类 | 使用可增可减的的数据情况下 | 当前内存/CPU使用情况、并发请求数量 |
| Histogram | 直方图类 | 使用统计指标信息在不同区间内的统计数量 | 延迟时间、响应大小。例如：0-1秒内的延迟时间、、0-5秒内的延迟时间、例如0-1kb之内的响应大小、0-5kb之内的响应大小 |
| Summary | 摘要类 | 类似于直方图，在客户端对百分位进行统计 | 延迟时间、响应大小。例如：超过百分之多少的人要满足需求的话，需要多长时间完成。 |
1. Gauge指标类型
1.1 不带label的基本例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 cpuUsage := prometheus.NewGauge(prometheus.GaugeOpts{ Name: "cpu_usage", // 指标名称 Help: "this is test metrics cpu usage", // 帮助信息 }) // 给指标设置值 cpuUsage.Set(89.56) // 注册指标 prometheus.MustRegister(cpuUsage) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
1.2 带有固定label指标的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 cpuUsage := prometheus.NewGauge(prometheus.GaugeOpts{ Name: "cpu_usage", // 指标名称 Help: "this is test metrics cpu usage", // 帮助信息 ConstLabels: prometheus.Labels{"MachineType": "host"}, // label }) // 给指标设置值 cpuUsage.Set(89.56) // 注册指标 prometheus.MustRegister(cpuUsage) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
1.3 带有非固定label指标的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { //定义带有不固定label的指标 mtu := prometheus.NewGaugeVec(prometheus.GaugeOpts{ Name: "interface_mtu", Help: "网卡接口MTU", }, []string{"interface", "Machinetype"}) // 给指标设置值 mtu.WithLabelValues("lo", "host").Set(1500) mtu.WithLabelValues("ens32", "host").Set(1500) mtu.WithLabelValues("eth0", "host").Set(1500) // 注册指标 prometheus.MustRegister(mtu) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
2. Counter指标类型
2.1 不带label的基本例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqTotal := prometheus.NewCounter(prometheus.CounterOpts{ Name: "current_request_total", Help: "当前请求总数", }) // 注册指标 prometheus.MustRegister(reqTotal) // 设置值 reqTotal.Add(10) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
2.2 带有固定label指标的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 suceReqTotal := prometheus.NewCounter(prometheus.CounterOpts{ Name: "sucess_request_total", Help: "请求成功的总数", ConstLabels: prometheus.Labels{"StatusCode": "200"}, }) // 注册指标 prometheus.MustRegister(suceReqTotal) // 设置值 suceReqTotal.Add(5675) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
2.3 带有非固定label指标的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 pathReqTotal := prometheus.NewCounterVec(prometheus.CounterOpts{ Name: "path_request_total", Help: "path请求总数", }, []string{"path"}) // 注册指标 prometheus.MustRegister(pathReqTotal) // 设置值 pathReqTotal.WithLabelValues("/token").Add(37) pathReqTotal.WithLabelValues("/auth").Add(23) pathReqTotal.WithLabelValues("/user").Add(90) pathReqTotal.WithLabelValues("/api").Add(67) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
3. Histogram指标类型
3.1 不带label的基本例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqDelay := prometheus.NewHistogram(prometheus.HistogramOpts{ Name: "request_delay", Help: "请求延迟，单位秒", Buckets: prometheus.LinearBuckets(0, 3, 6), // 调用LinearBuckets生成区间，从0开始，宽度3，共6个Bucket }) // 注册指标 prometheus.MustRegister(reqDelay) // 设置值 reqDelay.Observe(6) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
3.2 带固定label的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqDelay := prometheus.NewHistogram(prometheus.HistogramOpts{ Name: "request_delay", Help: "请求延迟，单位秒", Buckets: prometheus.LinearBuckets(0, 3, 6), // 调用LinearBuckets生成区间，从0开始，宽度3，共6个Bucket ConstLabels: prometheus.Labels{"path": "/api"}, }) // 注册指标 prometheus.MustRegister(reqDelay) // 设置值 reqDelay.Observe(6) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
3.3 带有非固定label的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqDelay := prometheus.NewHistogramVec(prometheus.HistogramOpts{ Name: "request_delay", Help: "请求延迟，单位秒", Buckets: prometheus.LinearBuckets(0, 3, 6), // 调用LinearBuckets生成区间，从0开始，宽度3，共6个Bucket }, []string{"path"}) // 注册指标 prometheus.MustRegister(reqDelay) // 设置值 reqDelay.WithLabelValues("/api").Observe(6) reqDelay.WithLabelValues("/user").Observe(3) reqDelay.WithLabelValues("/delete").Observe(2) reqDelay.WithLabelValues("/get_token").Observe(13) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
4. Summary指标类型
4.1 不带label的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqDelay := prometheus.NewSummary(prometheus.SummaryOpts{ Name: "request_delay", Help: "请求延迟", Objectives: map[float64]float64{0.5: 0.05, 0.90: 0.01, 0.99: 0.001}, // 百分比:精度 }) // 注册指标 prometheus.MustRegister(reqDelay) // 设置值 reqDelay.Observe(4) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
4.2 带有固定label的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqDelay := prometheus.NewSummary(prometheus.SummaryOpts{ Name: "request_delay", Help: "请求延迟", Objectives: map[float64]float64{0.5: 0.05, 0.90: 0.01, 0.99: 0.001}, // 百分比:精度 ConstLabels: prometheus.Labels{"path": "/api"}, }) // 注册指标 prometheus.MustRegister(reqDelay) // 设置值 reqDelay.Observe(4) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
4.3 带有非固定label的例子
package main import ( "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { // 定义指标 reqDelay := prometheus.NewSummaryVec(prometheus.SummaryOpts{ Name: "request_delay", Help: "请求延迟", Objectives: map[float64]float64{0.5: 0.05, 0.90: 0.01, 0.99: 0.001}, // 百分比:精度 }, []string{"path"}) // 注册指标 prometheus.MustRegister(reqDelay) // 设置值 reqDelay.WithLabelValues("/api").Observe(4) reqDelay.WithLabelValues("/token").Observe(2) reqDelay.WithLabelValues("/user").Observe(3) // 暴露指标 http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
5. 值的修改
5.1 基于事件的触发来修改值，比如每访问1次/api就增1
基于事件的触发对指标的值进行修改，通常大多数是来自业务方面的指标需求，如自研的应用需要暴露相关指标给promethus进行监控、展示，那么指标采集的代码（指标定义、设置值）就要嵌入到自研应用代码里了。
package main import ( "fmt" "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) func main() { urlRequestTotal := prometheus.NewCounterVec(prometheus.CounterOpts{ Name: "urlRequestTotal", Help: "PATH请求累计 单位 次", }, []string{"path"}) http.HandleFunc("/api", func(w http.ResponseWriter, r *http.Request) { urlRequestTotal.WithLabelValues(r.URL.Path).Inc() // 使用Inc函数进行增1 fmt.Fprintf(w, "Welcome to the api") }) prometheus.MustRegister(urlRequestTotal) http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
5.2 基于时间周期的触发来修改值，比如下面的示例中，是每间隔1秒获取cpu负载指标
基于时间周期的触发，可以是多少秒、分、时、日、月、周。
package main import ( "net/http" "time" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" "github.com/shirou/gopsutil/load" ) func main() { cpuUsage := prometheus.NewGaugeVec(prometheus.GaugeOpts{ Name: "CpuLoad", Help: "CPU负载", }, []string{"time"}) // 开启一个子协程执行该匿名函数内的逻辑来给指标设置值，且每秒获取一次 go func() { for range time.Tick(time.Second) { info, _ := load.Avg() cpuUsage.WithLabelValues("Load1").Set(float64(info.Load1)) cpuUsage.WithLabelValues("Load5").Set(float64(info.Load5)) cpuUsage.WithLabelValues("Load15").Set(float64(info.Load15)) } }() prometheus.MustRegister(cpuUsage) http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
5.3 基于每访问一次获取指标的URI才修改值，比如每次访问/metrics才去修改某些指标的值
下面的这个示例是，每访问一次/metrics就获取一次内存总容量
package main import ( "fmt" "net/http" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" "github.com/shirou/gopsutil/mem" ) func main() { MemTotal := prometheus.NewGaugeFunc(prometheus.GaugeOpts{ Name: "MemTotal", Help: "内存总容量 单位 GB", }, func() float64 { fmt.Println("call MemTotal ...") info, _ := mem.VirtualMemory() return float64(info.Total / 1024 / 1024 / 1024) }) prometheus.MustRegister(MemTotal) http.Handle("/metrics", promhttp.Handler()) http.ListenAndServe(":9900", nil) }
目前只有Gauge和Counter的指标类型有对应的函数，分别是NewGaugeFunc和NewCounterFunc，而且是固定的label。
