devbox
你好赵伟
这个屌丝很懒，什么也没留下！
关注作者
热门标签
热门文章
当前位置:   article > 正文

metrics指标采集

作者：你好赵伟 | 2024-08-09 13:28:14

赞

踩

metrics指标
官方参考资料：

http://metrics.dropwizard.io/4.0.0/

http://metrics.dropwizard.io/4.0.0/getting-started.html

http://metrics.dropwizard.io/4.0.0/manual/index.html


Metrics，英文就是度量、指标的意思，当我们需要为某个系统、服务做监控、做统计，就需要用到Metrics

举个例子，一个图片压缩服务：

  1. 每秒钟的请求数是多少（TPS TransactionPerSecond）
  2. 平均每个请求处理的时间
  3. 请求处理的最长耗时
  4. 等待处理的请求队列长度

又或者一个缓存服务：

  1. 缓存的命中率
  2. 平均查询缓存的时间

基本上每一个服务、应用都需要做一个监控系统，这需要尽量以少量的代码，实现统计某类数据的功能。

以 Java 为例，目前最为流行的 metrics 库是来自 Coda Hale 的 dropwizard/metrics ，该库被广泛地应用于各个知名的开源项目中。例如 Hadoop，Kafka，Spark，JStorm 中。


添加Maven依赖

  1.         <dependency>
  2.             <groupId>io.dropwizard.metrics</groupId>
  3.             <artifactId>metrics-core</artifactId>
  4.             <version>4.0.0</version>
  5.         </dependency>

Metric Registries

MetricRegistry类是Metrics的核心，它是存放应用中所有metrics的容器，也是我们使用 Metrics 库的起点 
MetricRegistry registry = new MetricRegistry();

每一个 metric 都有它独一无二的名字，Metrics 中使用句点名字，如 com.example.Queue.size。当你在 com.example.Queue 下有两个 metric 实例，如request和response，可以命名为：com.example.Queue.request.size 和 com.example.Queue.response.size 。使用MetricRegistry类的静态方法name()，可以非常方便地生成这种格式的名字，例如：

  1. MetricRegistry.name(Queue.class,"request","size");
  2. MetricRegistry.name(Queue.class,"response","size");

Metrics 数据展示

Metircs 提供了 Report 接口，用于展示 metrics 获取到的统计数据。metrics-core中主要实现了四种 reporter：JMX, console, SLF4J, 和 CSV。

1、console

可看下面“Metrics类型”的例子

2、JMX

看官方文档

3、SLF4J

看官方文档

4、CSV

看官方文档

5、除了metrics-core提供的4种数据输出外，metrics-servlets还提供了通过HTTP的输出方式

看官方文档

五种 Metrics 类型

1、Gauges

最简单的度量指标，只有一个简单的返回值，例如，我们想衡量一个待处理队列中任务的个数，代码如下：

  1. package com.metrics;
  2.  
  3. import com.codahale.metrics.*;
  4. import java.util.LinkedList;
  5. import java.util.Queue;
  6. import java.util.concurrent.TimeUnit;
  7.  
  8. public class GaugeTest {
  9.     public static void main(String[] args) throws InterruptedException {
  10.     	final Queue<String> q = new LinkedList<String>();
  11.         MetricRegistry registry = new MetricRegistry();
  12.         ConsoleReporter reporter = ConsoleReporter.forRegistry(registry).build();
  13.         reporter.start(1, TimeUnit.SECONDS);
  14. 		registry.register(
  15. 				MetricRegistry.name(GaugeTest.class, "queue", "size"),
  16. 				new Gauge<Integer>() {
  17. 					public Integer getValue() {
  18. 						return q.size();
  19. 					}
  20. 				});
  21.         while(true){
  22.             Thread.sleep(200);
  23.             q.add("Job-xxx");
  24.         }
  25.     }
  26. }

reporter.start(1, TimeUnit.SECONDS);用于启动ConsoleReporter（启动时会新建一个线程放到线程池中执行，线程在线程池中被执行会有延迟，所以与main函数后面的while部分没有顺序关系），且每秒钟从中读取一次数据输出到控制台，输出结果如下：

18-1-8 11:26:10 ================================================================

-- Gauges ----------------------------------------------------------------------
com.metrics.GaugeTest.queue.size
             value = 5


18-1-8 11:26:11 ================================================================

-- Gauges ----------------------------------------------------------------------
com.metrics.GaugeTest.queue.size
             value = 9


18-1-8 11:26:12 ================================================================

-- Gauges ----------------------------------------------------------------------
com.metrics.GaugeTest.queue.size
             value = 14

但是对于大多数队列数据结构，我们并不想简单地返回q.size()，因为java.util和java.util.concurrent中实现的size()方法很多都是O(n) 的复杂度，这会影响 Gauge 的性能。

2、Counters

Counter 就是计数器，Counter 只是用 Gauge 封装了 AtomicLong 。我们可以使用如下的方法，使得获得队列大小更加高效。

  1. package com.metrics;
  2.  
  3. import com.codahale.metrics.*;
  4. import java.util.Queue;
  5. import java.util.Random;
  6. import java.util.concurrent.LinkedBlockingQueue;
  7. import java.util.concurrent.TimeUnit;
  8.  
  9. public class CounterTest {
  10.  
  11.     public static Queue<String> q = new LinkedBlockingQueue<String>();
  12.  
  13.     public static Counter pendingJobs;
  14.  
  15.     public static Random random = new Random();
  16.  
  17.     public static void addJob(String job) {
  18.         pendingJobs.inc();
  19.         q.offer(job);
  20.     }
  21.  
  22.     public static String takeJob() {
  23.         String result =  q.poll();
  24.         if(result != null)
  25.         	pendingJobs.dec();
  26.         return result;
  27.     }
  28.  
  29.     public static void main(String[] args) throws InterruptedException {
  30.         MetricRegistry registry = new MetricRegistry();
  31.         ConsoleReporter reporter = ConsoleReporter.forRegistry(registry).build();
  32.         reporter.start(1, TimeUnit.SECONDS);
  33.  
  34.         pendingJobs = registry.counter(MetricRegistry.name(Queue.class,"pending-jobs","size"));
  35.  
  36.         int num = 0;
  37.         while(true){
  38.             Thread.sleep(200);
  39.             if (random.nextDouble() > 0.7){
  40.                 String job = takeJob();
  41.                 System.out.println("take job : "+job);
  42.             }else{
  43.             	num++;
  44.                 String job = "Job-"+num;
  45.                 addJob(job);
  46.                 System.out.println("add job : "+job);
  47.             }
  48.         }
  49.     }
  50. }

输出结果如下：

take job : null
take job : null
add job : Job-1
add job : Job-2
add job : Job-3
18-1-8 14:25:03 ================================================================

-- Counters --------------------------------------------------------------------
java.util.Queue.pending-jobs.size
             count = 3


take job : Job-1
add job : Job-4
add job : Job-5
add job : Job-6
18-1-8 14:25:04 ================================================================

-- Counters --------------------------------------------------------------------
java.util.Queue.pending-jobs.size
             count = 5

3、Meters

Meter度量一系列事件发生的速率(rate)，例如TPS。Meters会统计每秒，最近1分钟，5分钟，15分钟，还有全部时间的速率。

  1. package com.metrics;
  2.  
  3. import com.codahale.metrics.ConsoleReporter;
  4. import com.codahale.metrics.Meter;
  5. import com.codahale.metrics.MetricRegistry;
  6. import java.util.Random;
  7. import java.util.concurrent.TimeUnit;
  8.  
  9. public class MeterTest {
  10.     public static Random random = new Random();
  11.     public static void request(Meter meter){
  12.         System.out.println("request");
  13.         meter.mark();
  14.     }
  15.     public static void request(Meter meter, int n){
  16.         while(n > 0){
  17.             request(meter);
  18.             n--;
  19.         }
  20.     }
  21.     public static void main(String[] args) throws InterruptedException {
  22.         MetricRegistry registry = new MetricRegistry();
  23.         ConsoleReporter reporter = ConsoleReporter.forRegistry(registry).build();
  24.         reporter.start(1, TimeUnit.SECONDS);
  25.         Meter meterTps = registry.meter(MetricRegistry.name(MeterTest.class,"request","tps"));
  26.         while(true){
  27.             request(meterTps,random.nextInt(5));
  28.             Thread.sleep(1000);
  29.         }
  30.     }
  31. }

输出结果：

request
18-1-8 15:09:56 ================================================================

-- Meters ----------------------------------------------------------------------
com.metrics.MeterTest.request.tps
             count = 1
         mean rate = 0.97 events/second
     1-minute rate = 0.00 events/second
     5-minute rate = 0.00 events/second
    15-minute rate = 0.00 events/second


request
request
request
18-1-8 15:09:57 ================================================================

-- Meters ----------------------------------------------------------------------
com.metrics.MeterTest.request.tps
             count = 4
         mean rate = 1.99 events/second
     1-minute rate = 0.00 events/second
     5-minute rate = 0.00 events/second
    15-minute rate = 0.00 events/second


18-1-8 15:09:58 ================================================================

-- Meters ----------------------------------------------------------------------
com.metrics.MeterTest.request.tps
             count = 4
         mean rate = 1.33 events/second
     1-minute rate = 0.00 events/second
     5-minute rate = 0.00 events/second
    15-minute rate = 0.00 events/second


18-1-8 15:09:59 ================================================================

-- Meters ----------------------------------------------------------------------
com.metrics.MeterTest.request.tps
             count = 4
         mean rate = 1.00 events/second
     1-minute rate = 0.00 events/second
     5-minute rate = 0.00 events/second
    15-minute rate = 0.00 events/second


request
request
request
request
18-1-8 15:10:00 ================================================================

-- Meters ----------------------------------------------------------------------
com.metrics.MeterTest.request.tps
             count = 8
         mean rate = 1.60 events/second
     1-minute rate = 1.60 events/second
     5-minute rate = 1.60 events/second
    15-minute rate = 1.60 events/second


request
request
18-1-8 15:10:01 ================================================================

-- Meters ----------------------------------------------------------------------
com.metrics.MeterTest.request.tps
             count = 10
         mean rate = 1.66 events/second
     1-minute rate = 1.60 events/second
     5-minute rate = 1.60 events/second
    15-minute rate = 1.60 events/second

4、Histograms

Histogram统计数据的分布情况。比如最小值，最大值，中间值，还有中位数，75百分位, 90百分位, 95百分位, 98百分位, 99百分位, 和 99.9百分位的值(percentiles)。

  1. package com.metrics;
  2.  
  3. import com.codahale.metrics.ConsoleReporter;
  4. import com.codahale.metrics.ExponentiallyDecayingReservoir;
  5. import com.codahale.metrics.Histogram;
  6. import com.codahale.metrics.MetricRegistry;
  7. import java.util.Random;
  8. import java.util.concurrent.TimeUnit;
  9. public class HistogramTest {
  10.     public static Random random = new Random();
  11.     public static void main(String[] args) throws InterruptedException {
  12.         MetricRegistry registry = new MetricRegistry();
  13.         ConsoleReporter reporter = ConsoleReporter.forRegistry(registry).build();
  14.         reporter.start(1, TimeUnit.SECONDS);
  15.         Histogram histogram = new Histogram(new ExponentiallyDecayingReservoir());
  16.         registry.register(MetricRegistry.name(HistogramTest.class, "request", "histogram"), histogram);
  17.         while(true){
  18.             Thread.sleep(1000);
  19.             histogram.update(random.nextInt(100000));
  20.         }
  21.     }
  22. }

输出结果：

18-1-8 15:33:20 ================================================================

-- Histograms ------------------------------------------------------------------
com.metrics.HistogramTest.request.histogram
             count = 1
               min = 0
               max = 0
              mean = 0.00
            stddev = 0.00
            median = 0.00
              75% <= 0.00
              95% <= 0.00
              98% <= 0.00
              99% <= 0.00
            99.9% <= 0.00


18-1-8 15:33:21 ================================================================

-- Histograms ------------------------------------------------------------------
com.metrics.HistogramTest.request.histogram
             count = 1
               min = 23833
               max = 23833
              mean = 23833.00
            stddev = 0.00
            median = 23833.00
              75% <= 23833.00
              95% <= 23833.00
              98% <= 23833.00
              99% <= 23833.00
            99.9% <= 23833.00


18-1-8 15:33:22 ================================================================

-- Histograms ------------------------------------------------------------------
com.metrics.HistogramTest.request.histogram
             count = 2
               min = 23833
               max = 91252
              mean = 57795.32
            stddev = 33708.55
            median = 91252.00
              75% <= 91252.00
              95% <= 91252.00
              98% <= 91252.00
              99% <= 91252.00
            99.9% <= 91252.00


18-1-8 15:33:23 ================================================================

-- Histograms ------------------------------------------------------------------
com.metrics.HistogramTest.request.histogram
             count = 3
               min = 23833
               max = 91252
              mean = 54671.26
            stddev = 27765.11
            median = 48562.00
              75% <= 91252.00
              95% <= 91252.00
              98% <= 91252.00
              99% <= 91252.00
            99.9% <= 91252.00

5、Timers

  1. package com.metrics;
  2.  
  3. import com.codahale.metrics.ConsoleReporter;
  4. import com.codahale.metrics.MetricRegistry;
  5. import com.codahale.metrics.Timer;
  6. import java.util.Random;
  7. import java.util.concurrent.TimeUnit;
  8. public class TimerTest {
  9.     public static Random random = new Random();
  10.     public static void main(String[] args) throws InterruptedException {
  11.         MetricRegistry registry = new MetricRegistry();
  12.         ConsoleReporter reporter = ConsoleReporter.forRegistry(registry).build();
  13.         reporter.start(1, TimeUnit.SECONDS);
  14.         Timer timer = registry.timer(MetricRegistry.name(TimerTest.class,"get-latency"));
  15.         Timer.Context ctx;
  16.         while(true){
  17.             ctx = timer.time();//计时开始
  18.             Thread.sleep(random.nextInt(1000));//模拟操作耗时
  19.             //计时结束，meter：统计总数、每秒、分等调用的次数，histogram：记录每次操作耗时的分布情况
  20.             ctx.stop();
  21.         }
  22.     }
  23. }

输出结果：

18-1-8 15:56:55 ================================================================

-- Timers ----------------------------------------------------------------------
com.metrics.TimerTest.get-latency
             count = 2
         mean rate = 1.99 calls/second
     1-minute rate = 0.00 calls/second
     5-minute rate = 0.00 calls/second
    15-minute rate = 0.00 calls/second
               min = 88.08 milliseconds
               max = 693.13 milliseconds
              mean = 390.60 milliseconds
            stddev = 302.52 milliseconds
            median = 693.13 milliseconds
              75% <= 693.13 milliseconds
              95% <= 693.13 milliseconds
              98% <= 693.13 milliseconds
              99% <= 693.13 milliseconds
            99.9% <= 693.13 milliseconds


18-1-8 15:56:56 ================================================================

-- Timers ----------------------------------------------------------------------
com.metrics.TimerTest.get-latency
             count = 3
         mean rate = 1.50 calls/second
     1-minute rate = 0.00 calls/second
     5-minute rate = 0.00 calls/second
    15-minute rate = 0.00 calls/second
               min = 88.08 milliseconds
               max = 848.11 milliseconds
              mean = 544.63 milliseconds
            stddev = 327.80 milliseconds
            median = 693.13 milliseconds
              75% <= 848.11 milliseconds
              95% <= 848.11 milliseconds
              98% <= 848.11 milliseconds
              99% <= 848.11 milliseconds
            99.9% <= 848.11 milliseconds


18-1-8 15:56:57 ================================================================

-- Timers ----------------------------------------------------------------------
com.metrics.TimerTest.get-latency
             count = 5
         mean rate = 1.67 calls/second
     1-minute rate = 0.00 calls/second
     5-minute rate = 0.00 calls/second
    15-minute rate = 0.00 calls/second
               min = 88.08 milliseconds
               max = 848.11 milliseconds
              mean = 519.35 milliseconds
            stddev = 323.80 milliseconds
            median = 693.13 milliseconds
              75% <= 796.57 milliseconds
              95% <= 848.11 milliseconds
              98% <= 848.11 milliseconds
              99% <= 848.11 milliseconds
            99.9% <= 848.11 milliseconds















声明：本文内容由网友自发贡献，不代表【wpsshop博客】立场，版权归原作者所有，本站不承担相应法律责任。如您发现有侵权的内容，请联系我们。转载请注明出处：https://www.wpsshop.cn/w/你好赵伟/article/detail/953520
推荐阅读
相关标签

Copyright © 2003-2013 www.wpsshop.cn 版权所有，并保留所有权利。

  

闽ICP备14008679号