【问题标题】:How to instrument Prometheus Gauge using OpenCensus?如何使用 OpenCensus 检测 Prometheus Gauge?
【发布时间】:2021-01-23 03:45:22
【问题描述】:

我正在尝试找到一种在 Golang 中使用 OpenCencus 来检测 Prometheus Gauge 指标的方法。目标是跟踪活动会话的数量。所以值可以增加和减少,也可以在服务器重启时重置为 0。

他们有一个例子https://opencensus.io/quickstart/go/metrics/,但我无法将任何与 Gauge 相关联并重置为 0。

您能否建议我应该使用哪个 Measure 和 View 来检测可以增加、减少和重置为 0 的 Gauge?

【问题讨论】:

    标签: go prometheus gauge opencensus


    【解决方案1】:

    https://opencensus.io/stats/view/

    我没有尝试过,但LastValue 可能(!?)转换为 Prometheus Gauge。

    Count 为您提供测量次数并产生一个(增加的)计数器。所以,这对你没有帮助。

    唯一的其他选择是SumDistribution

    如果LastValue 没有产生一个仪表,您可能需要使用Distribution

    更新:LastValue == Gauge

    破解了给出的例子:

    package main
    
    import (
        "context"
        "fmt"
        "log"
        "math/rand"
        "net/http"
        "os"
        "time"
    
        "contrib.go.opencensus.io/exporter/prometheus"
        "go.opencensus.io/stats"
        "go.opencensus.io/stats/view"
        "go.opencensus.io/tag"
    )
    
    var (
        MLatencyMs = stats.Float64("latency", "The latency in milliseconds", "ms")
    )
    var (
        KeyMethod, _ = tag.NewKey("method")
    )
    
    func main() {
    
        port := os.Getenv("PORT")
        if port == "" {
            port = "8080"
        }
    
        view1 := &view.View{
            Name:        "dist",
            Measure:     MLatencyMs,
            Description: "The dist of the latencies",
            TagKeys:     []tag.Key{KeyMethod},
            Aggregation: view.Distribution(0, 10, 100, 1000, 10000, 100000),
        }
    
        view2 := &view.View{
            Name:        "last",
            Measure:     MLatencyMs,
            Description: "The last of the latencies",
            TagKeys:     []tag.Key{KeyMethod},
            Aggregation: view.LastValue(),
        }
    
        if err := view.Register(view1, view2); err != nil {
            log.Fatalf("Failed to register the views: %v", err)
        }
    
        pe, err := prometheus.NewExporter(prometheus.Options{
            Namespace: "distlast",
        })
        if err != nil {
            log.Fatalf("Failed to create the Prometheus stats exporter: %v", err)
        }
    
        go func() {
            mux := http.NewServeMux()
            mux.Handle("/metrics", pe)
            log.Fatal(http.ListenAndServe(fmt.Sprintf(":%s", port), mux))
        }()
    
        rand.Seed(time.Now().UnixNano())
        ctx := context.Background()
    
        for {
            n := rand.Intn(100)
            log.Printf("[loop] n=%d\n", n)
            stats.Record(ctx, MLatencyMs.M(float64(time.Duration(n))))
            time.Sleep(1 * time.Second)
        }
    
    }
    

    然后go run . 产生:

    2020/10/15 14:03:25 [loop] n=77
    2020/10/15 14:03:26 [loop] n=62
    2020/10/15 14:03:27 [loop] n=48
    2020/10/15 14:03:28 [loop] n=76
    2020/10/15 14:03:29 [loop] n=20
    2020/10/15 14:03:30 [loop] n=46
    2020/10/15 14:03:31 [loop] n=47
    2020/10/15 14:03:32 [loop] n=64
    2020/10/15 14:03:33 [loop] n=15
    2020/10/15 14:03:34 [loop] n=8
    

    localhost:8080/metrics 上的指标收益率:

    # HELP distlast_dist The dist of the latencies
    # TYPE distlast_dist histogram
    distlast_dist_bucket{method="",le="10"} 1
    distlast_dist_bucket{method="",le="100"} 10
    distlast_dist_bucket{method="",le="1000"} 10
    distlast_dist_bucket{method="",le="10000"} 10
    distlast_dist_bucket{method="",le="100000"} 10
    distlast_dist_bucket{method="",le="+Inf"} 10
    distlast_dist_sum{method=""} 463.00000000000006
    distlast_dist_count{method=""} 10
    # HELP distlast_last The last of the latencies
    # TYPE distlast_last gauge
    distlast_last{method=""} 8
    

    注意distlast_last 的值为8,对应于n=8distlast_dist_sum 的值为463

    【讨论】: