Metrics.NET实践(1)

起因：对应用的监控和测量是WEB应用的一个重要话题，尤其在监控错误率，并发量，以及框架库中的动态值。于是，在性能优化的时候找到了metrics.net。

• 简介
• 开始使用
• 度量
• Gauges
• Counters
• Meters
• Histograms
• Timers

1. 簡介

• 項目主頁：https://github.com/etishor/Metrics.NET
• WIKI：https://github.com/etishor/Metrics.NET/wiki
• 可視化項目：https://github.com/etishor/Metrics.NET.FlotVisualization
• 可視化Demo：http://www.erata.net/Metrics.NET/demo/#/
• Java版地址：https://github.com/dropwizard/metrics

Metrics.NET - a .NET Port, with lots of additional functionality, of the awesome Java metrics library by Coda Hale.

2. Getting Started

Install-Package Metrics.NET
Install-Package Metrics.NET.ElasticSearch -Version 0.5.0 # with ES
Install-Package Metrics.NET.Graphite -Version 0.5.0 #with Graphite

other:

• Metrics.NET.RabbitMQ
• Metrics.NET.InfluxDB
• Metrics.NET.SignalFX
• Metrics.NET.InfluxDbReporting
• Metrics.NET.CloudWatch

Application_Start：

using Metrics;

Metric.Config
    .WithHttpEndpoint("http://localhost:1234/")
    .WithAllCounters();

访问：http://localhost:1234/ 即可看到效果。

3. 度量

• Meters record the rate at which an event occurs。某事件发生的比率
• Histograms measure the distribution of values in a stream of data。数据流的分布
• Timers keep a histogram of the duration of a type of event and a meter of the rate of its occurrence。Meters和Histograms的结合。
• Counters 64 bit integers that can be incremented or decremented。64位计数器
• Gauges instantaneous values。简单值

3.1 Gauges

最简单的度量方式。代表一个瞬时值。

代码片段：

// gauge from Func<double>
Metric.Gauge("MyValue", () => ComputeMagicValue(), Unit.Items);

// gauge that reads its value from a performance counter
Metric.PerformanceCounter("CPU Usage", "Processor", "% Processor Time",
    "_Total", Unit.Custom("%"));

// gauge that transforms the value of another gauge
Metric.Advanced.Gauge("Kbytes gauge",
    () => new DerivedGauge(gaugeReturningValueInBytes, v => v / 1024.0 ),
    Unit.KiloBytes);

Metric.Context("[LogPool]").Gauge("dict.Count_Gauge",
    () => { return dict.Count; }, Unit.Custom("個"), "log");

3.2 Counters

代表可以增減的64位整數。

代碼片段——緩存的數量

public class Cache
{
    private static readonly Counter counter =
        Metric.Counter("ItemsInCache", Unit.Items);

    private void AddItems(object[] items)
    {
        counter.Increment(items.Length);
    }

    private void AddItem(object item)
    {
        counter.Increment();
    }

    private void RemoveItem(object item)
    {
        counter.Decrement();
    }
}

Counter提供分组计数的能力，针对标记接口可以实现：

public class SetCounterSample
{
    private readonly Counter commandCounter =
        Metric.Counter("Command Counter", Unit.Custom("Commands"));

    public interface Command { }
    public class SendEmail : Command { }
    public class ShipProduct : Command { }
    public class BillCustomer : Command { }
    public class MakeInvoice : Command { }
    public class MarkAsPreffered : Command { }

    public void Process(Command command)
    {
        this.commandCounter.Increment(command.GetType().Name);

        // do actual command processing
    }
}

输出：

 Command Counter
             Count = 2550 Commands
       Total Items = 5
            Item 0 = 20.90%   533 Commands [BillCustomer]
            Item 1 = 19.22%   490 Commands [MakeInvoice]
            Item 2 = 19.41%   495 Commands [MarkAsPreffered]
            Item 3 = 20.98%   535 Commands [SendEmail]
            Item 4 = 19.49%   497 Commands [ShipProduct]

应用场景：可以用在WEB API中统计各种请求的数量。

3.3 Meters

A meter measures the rate at which an event occurs.meter测量一个事件发生的比率。

示例代码：请求异常的比率

public class RequestProcessor
{
    private readonly Meter meter =
        Metric.Meter("Errors", Unit.Requests, TimeUnit.Seconds);

    public void ProcessRequest()
    {
        try
        {
            // do actual processing
        }
        catch
        {
            meter.Mark(); // records an error
            throw;
        }
    }
}

同样，也可以支持多态的分组：

public class SetMeterSample
{
    private readonly Meter errorMeter = Metric.Meter("Errors", Unit.Errors);

    public interface Command { }
    public class SendEmail : Command { }
    public class ShipProduct : Command { }
    public class BillCustomer : Command { }
    public class MakeInvoice : Command { }
    public class MarkAsPreffered : Command { }

    public void Process(Command command)
    {
        try
        {
            ActualCommandProcessing(command);
        }
        catch
        {
            errorMeter.Mark(command.GetType().Name);
        }
    }
}

输出：

 Errors
             Count = 450 Errors
        Mean Value = 35.68 Errors/s
     1 Minute Rate = 25.44 Errors/s
     5 Minute Rate = 24.30 Errors/s
    15 Minute Rate = 24.10 Errors/s
       Total Items = 5
            Item 0 = 19.56%    88 Errors [BillCustomer]
             Count = 88 Errors
        Mean Value = 6.98 Errors/s
     1 Minute Rate = 6.05 Errors/s
     5 Minute Rate = 6.01 Errors/s
    15 Minute Rate = 6.00 Errors/s
            Item 1 = 18.67%    84 Errors [MakeInvoice]
             Count = 84 Errors
        Mean Value = 6.66 Errors/s
     1 Minute Rate = 4.23 Errors/s
     5 Minute Rate = 3.89 Errors/s
    15 Minute Rate = 3.83 Errors/s
            Item 2 = 20.22%    91 Errors [MarkAsPreffered]
             Count = 91 Errors
        Mean Value = 7.22 Errors/s
     1 Minute Rate = 5.38 Errors/s
     5 Minute Rate = 5.24 Errors/s
    15 Minute Rate = 5.21 Errors/s
            Item 3 = 19.78%    89 Errors [SendEmail]
             Count = 89 Errors
        Mean Value = 7.06 Errors/s
     1 Minute Rate = 4.92 Errors/s
     5 Minute Rate = 4.67 Errors/s
    15 Minute Rate = 4.62 Errors/s
            Item 4 = 21.78%    98 Errors [ShipProduct]
             Count = 98 Errors
        Mean Value = 7.77 Errors/s
     1 Minute Rate = 4.86 Errors/s
     5 Minute Rate = 4.50 Errors/s
    15 Minute Rate = 4.43 Errors/s

3.4 Histograms

代码片段：搜索结果的分布。

开箱即用的三种抽样方法：

• Exponentially Decaying Reservoir - 最近五分钟数据的分位数。
• Uniform Reservoir - 产生整个週期有效的分位数
• Sliding Window Reservoir - 产生代表过去N次测量的分位数

 private readonly Histogram histogram = Metric.Histogram("Search Results", Unit.Items);
    public void Search(string keyword)
    {
        var results = ActualSearch(keyword);
        histogram.Update(results.Length);
    }

// The histogram has the ability to track for which user value a Min, Max or Last Value has been recorded.
// The user value can be any string value (documentId, operationId, etc).
public class UserValueHistogramSample
{
    private readonly Histogram histogram =
        Metric.Histogram("Results", Unit.Items);

    public void Process(string documentId)
    {
        var results = GetResultsForDocument(documentId);
        this.histogram.Update(results.Length, documentId);
    }
}

输出：

    Results
             Count = 90 Items
              Last = 46.00 Items
   Last User Value = document-3
               Min = 2.00 Items
    Min User Value = document-7
               Max = 98.00 Items
    Max User Value = document-4
              Mean = 51.52 Items
            StdDev = 30.55 Items
            Median = 50.00 Items
              75% <= 80.00 Items
              95% <= 97.00 Items
              98% <= 98.00 Items
              99% <= 98.00 Items
            99.9% <= 98.00 Items

3.5 Timers

示例代碼：

private readonly Timer timer =
    Metric.Timer("HTTP Requests",Unit.Requests);

public void ProcessRequest()
{
    using(timer.NewContext())
    {
        // Actual Processing of the request
    }
}

private readonly Timer timer =
    Metric.Timer("Requests", Unit.Requests);

public void Process(string documentId)
{
    using (var context = timer.NewContext(documentId))
    {
        ActualProcessingOfTheRequest(documentId);

        // if needed elapsed time is available in context.Elapsed
    }
}

輸出：

  Requests
             Count = 14 Requests
        Mean Value = 1.86 Requests/s
     1 Minute Rate = 1.80 Requests/s
     5 Minute Rate = 1.80 Requests/s
    15 Minute Rate = 1.80 Requests/s
             Count = 14 Requests
              Last = 869.03 ms
   Last User Value = document-1
               Min = 59.90 ms
    Min User Value = document-6
               Max = 869.03 ms
    Max User Value = document-1
              Mean = 531.81 ms
            StdDev = 212.98 ms
            Median = 594.83 ms
              75% <= 670.18 ms
              95% <= 869.03 ms
              98% <= 869.03 ms
              99% <= 869.03 ms
            99.9% <= 869.03 ms