java版的Metric工具介绍-hustfxj-ChinaUnix博客

Metrics是一个给JAVA服务的各项指标提供度量工具的包，在JAVA代码中嵌入Metrics代码，可以方便的对业务代码的各个指标进行监控，同时，Metrics能够很好的跟Ganlia、Graphite结合，方便的提供图形化接口。基本使用方式直接将core包（目前稳定版本3.0.1）导入pom文件即可，配置如下：

<dependency> <groupId>com.codahale.metrics</groupId> <artifactId>metrics-core</artifactId> <version>3.0.1</version> </dependency>

core包主要提供如下核心功能：

• Metrics Registries类似一个metrics容器，维护一个Map，可以是一个服务一个实例。
• 支持五种metric类型：Gauges、Counters、Meters、Histograms和Timers。
• 可以将metrics值通过JMX、Console，CSV文件和SLF4J loggers发布出来。

五种Metrics类型：

1.       Gauges

Gauges是一个最简单的计量，一般用来统计瞬时状态的数据信息，比如系统中处于pending状态的job。测试代码
 

点击(此处)折叠或打开

1. package com.netease.test.metrics;
2.  
3. import com.codahale.metrics.ConsoleReporter;
4. import com.codahale.metrics.Gauge;
5. import com.codahale.metrics.JmxReporter;
6. import com.codahale.metrics.MetricRegistry;
7.  
8. import java.util.Queue;
9. import java.util.concurrent.LinkedBlockingDeque;
10. import java.util.concurrent.TimeUnit;
11.  
12. /**
13.  * User: hzwangxx
14.  * Date: 14-2-17
15.  * Time: 14:47
16.  * 测试Gauges，实时统计pending状态的job个数
17.  */
18. public class TestGauges {
19.     /**
20.      * 实例化一个registry，最核心的一个模块，相当于一个应用程序的metrics系统的容器，维护一个Map
21.      */
22.     private static final MetricRegistry metrics = new MetricRegistry();
23.  
24.     private static Queue<String> queue = new LinkedBlockingDeque<String>();
25.  
26.     /**
27.      * 在控制台上打印输出
28.      */
29.     private static ConsoleReporter reporter = ConsoleReporter.forRegistry(metrics).build();
30.  
31.     public static void main(String[] args) throws InterruptedException {
32.         reporter.start(3, TimeUnit.SECONDS);
33.  
34.         //实例化一个Gauge
35.         Gauge<Integer> gauge = new Gauge<Integer>() {
36.             @Override
37.             public Integer getValue() {
38.                 return queue.size();
39.             }
40.         };
41.  
42.         //注册到容器中
43.         metrics.register(MetricRegistry.name(TestGauges.class, "pending-job", "size"), gauge);
44.  
45.         //测试JMX
46.         JmxReporter jmxReporter = JmxReporter.forRegistry(metrics).build();
47.         jmxReporter.start();
48.  
49.         //模拟数据
50.         for (int i=0; i<20; i++){
51.             queue.add("a");
52.             Thread.sleep(1000);
53.         }
54.  
55.     }
56. }
57.  
58. /*
59. console output:
60. 14-2-17 15:29:35 ===============================================================
61.  
62. -- Gauges ----------------------------------------------------------------------
63. com.netease.test.metrics.TestGauges.pending-job.size
64.              value = 4
65.  
66.  
67. 14-2-17 15:29:38 ===============================================================
68.  
69. -- Gauges ----------------------------------------------------------------------
70. com.netease.test.metrics.TestGauges.pending-job.size
71.              value = 6
72.  
73.  
74. 14-2-17 15:29:41 ===============================================================
75.  
76. -- Gauges ----------------------------------------------------------------------
77. com.netease.test.metrics.TestGauges.pending-job.size
78.              value = 9
79.  */

 

通过以上步骤将会向MetricsRegistry容器中注册一个名字为com.netease.test.metrics .TestGauges.pending-job.size的metrics，实时获取队列长度的指标。另外，Core包种还扩展了几种特定的Gauge：

• JMX Gauges—提供给第三方库只通过JMX将指标暴露出来。
• Ratio Gauges—简单地通过创建一个gauge计算两个数的比值。
• Cached Gauges—对某些计量指标提供缓存

Derivative Gauges—提供Gauge的值是基于其他Gauge值的接口。

2.       Counter

Counter是Gauge的一个特例，维护一个计数器，可以通过inc()和dec()方法对计数器做修改。使用步骤与Gauge基本类似，在MetricRegistry中提供了静态方法可以直接实例化一个Counter。

 

点击(此处)折叠或打开

1. package com.netease.test.metrics;
2.  
3. import com.codahale.metrics.ConsoleReporter;
4. import com.codahale.metrics.Counter;
5. import com.codahale.metrics.MetricRegistry;
6.  
7. import java.util.LinkedList;
8. import java.util.Queue;
9. import java.util.concurrent.TimeUnit;
10. import static com.codahale.metrics.MetricRegistry.*;
11. /**
12.  * User: hzwangxx
13.  * Date: 14-2-14
14.  * Time: 14:02
15.  * 测试Counter
16.  */
17. public class TestCounter {
18.  
19.     /**
20.      * 实例化一个registry，最核心的一个模块，相当于一个应用程序的metrics系统的容器，维护一个Map
21.      */
22.     private static final MetricRegistry metrics = new MetricRegistry();
23.  
24.     /**
25.      * 在控制台上打印输出
26.      */
27.     private static ConsoleReporter reporter = ConsoleReporter.forRegistry(metrics).build();
28.  
29.     /**
30.      * 实例化一个counter,同样可以通过如下方式进行实例化再注册进去
31.      * pendingJobs = new Counter();
32.      * metrics.register(MetricRegistry.name(TestCounter.class, "pending-jobs"), pendingJobs);
33.      */
34.     private static Counter pendingJobs = metrics.counter(name(TestCounter.class, "pedding-jobs"));
35. // private static Counter pendingJobs = metrics.counter(MetricRegistry.name(TestCounter.class, "pedding-jobs"));
36.  
37.  
38.  
39.     private static Queue<String> queue = new LinkedList<String>();
40.  
41.     public static void add(String str) {
42.         pendingJobs.inc();
43.         queue.offer(str);
44.     }
45.  
46.     public String take() {
47.         pendingJobs.dec();
48.         return queue.poll();
49.     }
50.  
51.     public static void main(String[]args) throws InterruptedException {
52.         reporter.start(3, TimeUnit.SECONDS);
53.         while(true){
54.             add("1");
55.             Thread.sleep(1000);
56.         }
57.  
58.     }
59. }
60.  
61. /*
62. console output：
63. 14-2-17 17:52:34 ===============================================================
64.  
65. -- Counters --------------------------------------------------------------------
66. com.netease.test.metrics.TestCounter.pedding-jobs
67.              count = 4
68.  
69.  
70. 14-2-17 17:52:37 ===============================================================
71.  
72. -- Counters --------------------------------------------------------------------
73. com.netease.test.metrics.TestCounter.pedding-jobs
74.              count = 6
75.  
76.  
77. 14-2-17 17:52:40 ===============================================================
78.  
79. -- Counters --------------------------------------------------------------------
80. com.netease.test.metrics.TestCounter.pedding-jobs
81.              count = 9
82.  
83.  */

 

3.       Meters

Meters用来度量某个时间段的平均处理次数（request per second），每1、5、15分钟的TPS。比如一个service的请求数，通过metrics.meter()实例化一个Meter之后，然后通过meter.mark()方法就能将本次请求记录下来。统计结果有总的请求数，平均每秒的请求数，以及最近的1、5、15分钟的平均TPS。

 

点击(此处)折叠或打开

1. package com.netease.test.metrics;
2.  
3. import com.codahale.metrics.ConsoleReporter;
4. import com.codahale.metrics.Meter;
5. import com.codahale.metrics.MetricRegistry;
6.  
7. import java.util.concurrent.TimeUnit;
8.  
9. import static com.codahale.metrics.MetricRegistry.*;
10.  
11. /**
12.  * User: hzwangxx
13.  * Date: 14-2-17
14.  * Time: 18:34
15.  * 测试Meters
16.  */
17. public class TestMeters {
18.     /**
19.      * 实例化一个registry，最核心的一个模块，相当于一个应用程序的metrics系统的容器，维护一个Map
20.      */
21.     private static final MetricRegistry metrics = new MetricRegistry();
22.  
23.     /**
24.      * 在控制台上打印输出
25.      */
26.     private static ConsoleReporter reporter = ConsoleReporter.forRegistry(metrics).build();
27.  
28.     /**
29.      * 实例化一个Meter
30.      */
31.     private static final Meter requests = metrics.meter(name(TestMeters.class, "request"));
32.  
33.     public static void handleRequest() {
34.         requests.mark();
35.     }
36.  
37.     public static void main(String[] args) throws InterruptedException {
38.         reporter.start(3, TimeUnit.SECONDS);
39.         while(true){
40.             handleRequest();
41.             Thread.sleep(100);
42.         }
43.     }
44.  
45. }
46.  
47. /*
48. 14-2-17 18:43:08 ===============================================================
49.  
50. -- Meters ----------------------------------------------------------------------
51. com.netease.test.metrics.TestMeters.request
52.              count = 30
53.          mean rate = 9.95 events/second
54.      1-minute rate = 0.00 events/second
55.      5-minute rate = 0.00 events/second
56.     15-minute rate = 0.00 events/second
57.  
58.  
59. 14-2-17 18:43:11 ===============================================================
60.  
61. -- Meters ----------------------------------------------------------------------
62. com.netease.test.metrics.TestMeters.request
63.              count = 60
64.          mean rate = 9.99 events/second
65.      1-minute rate = 10.00 events/second
66.      5-minute rate = 10.00 events/second
67.     15-minute rate = 10.00 events/second
68.  
69.  
70. 14-2-17 18:43:14 ===============================================================
71.  
72. -- Meters ----------------------------------------------------------------------
73. com.netease.test.metrics.TestMeters.request
74.              count = 90
75.          mean rate = 9.99 events/second
76.      1-minute rate = 10.00 events/second
77.      5-minute rate = 10.00 events/second
78.     15-minute rate = 10.00 events/second
79. */

 

4.       Histograms

Histograms主要使用来统计数据的分布情况，最大值、最小值、平均值、中位数，百分比（75%、90%、95%、98%、99%和99.9%）。例如，需要统计某个页面的请求响应时间分布情况，可以使用该种类型的Metrics进行统计。具体的样例代码如下：

 

点击(此处)折叠或打开

1. package com.netease.test.metrics;
2.  
3. import com.codahale.metrics.ConsoleReporter;
4. import com.codahale.metrics.Histogram;
5. import com.codahale.metrics.MetricRegistry;
6.  
7. import java.util.Random;
8. import java.util.concurrent.TimeUnit;
9.  
10. import static com.codahale.metrics.MetricRegistry.name;
11.  
12. /**
13.  * User: hzwangxx
14.  * Date: 14-2-17
15.  * Time: 18:34
16.  * 测试Histograms
17.  */
18. public class TestHistograms {
19.     /**
20.      * 实例化一个registry，最核心的一个模块，相当于一个应用程序的metrics系统的容器，维护一个Map
21.      */
22.     private static final MetricRegistry metrics = new MetricRegistry();
23.  
24.     /**
25.      * 在控制台上打印输出
26.      */
27.     private static ConsoleReporter reporter = ConsoleReporter.forRegistry(metrics).build();
28.  
29.     /**
30.      * 实例化一个Histograms
31.      */
32.     private static final Histogram randomNums = metrics.histogram(name(TestHistograms.class, "random"));
33.  
34.     public static void handleRequest(double random) {
35.         randomNums.update((int) (random*100));
36.     }
37.  
38.     public static void main(String[] args) throws InterruptedException {
39.         reporter.start(3, TimeUnit.SECONDS);
40.         Random rand = new Random();
41.         while(true){
42.             handleRequest(rand.nextDouble());
43.             Thread.sleep(100);
44.         }
45.     }
46.  
47. }
48.  
49. /*
50. 14-2-17 19:39:11 ===============================================================
51.  
52. -- Histograms ------------------------------------------------------------------
53. com.netease.test.metrics.TestHistograms.random
54.              count = 30
55.                min = 1
56.                max = 97
57.               mean = 45.93
58.             stddev = 29.12
59.             median = 39.50
60.               75% <= 71.00
61.               95% <= 95.90
62.               98% <= 97.00
63.               99% <= 97.00
64.             99.9% <= 97.00
65.  
66.  
67. 14-2-17 19:39:14 ===============================================================
68.  
69. -- Histograms ------------------------------------------------------------------
70. com.netease.test.metrics.TestHistograms.random
71.              count = 60
72.                min = 0
73.                max = 97
74.               mean = 41.17
75.             stddev = 28.60
76.             median = 34.50
77.               75% <= 69.75
78.               95% <= 92.90
79.               98% <= 96.56
80.               99% <= 97.00
81.             99.9% <= 97.00
82.  
83.  
84. 14-2-17 19:39:17 ===============================================================
85.  
86. -- Histograms ------------------------------------------------------------------
87. com.netease.test.metrics.TestHistograms.random
88.              count = 90
89.                min = 0
90.                max = 97
91.               mean = 44.67
92.             stddev = 28.47
93.             median = 43.00
94.               75% <= 71.00
95.               95% <= 91.90
96.               98% <= 96.18
97.               99% <= 97.00
98.             99.9% <= 97.00
99. */

5.       Timers

 

Timers主要是用来统计某一块代码段的执行时间以及其分布情况，具体是基于Histograms和Meters来实现的。样例代码如下：
 

点击(此处)折叠或打开

1. package com.netease.test.metrics;
2.  
3. import com.codahale.metrics.ConsoleReporter;
4. import com.codahale.metrics.MetricRegistry;
5. import com.codahale.metrics.Timer;
6.  
7. import java.util.Random;
8. import java.util.concurrent.TimeUnit;
9.  
10. import static com.codahale.metrics.MetricRegistry.name;
11.  
12. /**
13.  * User: hzwangxx
14.  * Date: 14-2-17
15.  * Time: 18:34
16.  * 测试Timers
17.  */
18. public class TestTimers {
19.     /**
20.      * 实例化一个registry，最核心的一个模块，相当于一个应用程序的metrics系统的容器，维护一个Map
21.      */
22.     private static final MetricRegistry metrics = new MetricRegistry();
23.  
24.     /**
25.      * 在控制台上打印输出
26.      */
27.     private static ConsoleReporter reporter = ConsoleReporter.forRegistry(metrics).build();
28.  
29.     /**
30.      * 实例化一个Meter
31.      */
32. // private static final Timer requests = metrics.timer(name(TestTimers.class, "request"));
33.     private static final Timer requests = metrics.timer(name(TestTimers.class, "request"));
34.  
35.     public static void handleRequest(int sleep) {
36.         Timer.Context context = requests.time();
37.         try {
38.             //some operator
39.             Thread.sleep(sleep);
40.         } catch (InterruptedException e) {
41.             e.printStackTrace();
42.         } finally {
43.             context.stop();
44.         }
45.  
46.     }
47.  
48.     public static void main(String[] args) throws InterruptedException {
49.         reporter.start(3, TimeUnit.SECONDS);
50.         Random random = new Random();
51.         while(true){
52.             handleRequest(random.nextInt(1000));
53.         }
54.     }
55.  
56. }
57.  
58. /*
59. 14-2-18 9:31:54 ================================================================
60.  
61. -- Timers ----------------------------------------------------------------------
62. com.netease.test.metrics.TestTimers.request
63.              count = 4
64.          mean rate = 1.33 calls/second
65.      1-minute rate = 0.00 calls/second
66.      5-minute rate = 0.00 calls/second
67.     15-minute rate = 0.00 calls/second
68.                min = 483.07 milliseconds
69.                max = 901.92 milliseconds
70.               mean = 612.64 milliseconds
71.             stddev = 196.32 milliseconds
72.             median = 532.79 milliseconds
73.               75% <= 818.31 milliseconds
74.               95% <= 901.92 milliseconds
75.               98% <= 901.92 milliseconds
76.               99% <= 901.92 milliseconds
77.             99.9% <= 901.92 milliseconds
78.  
79.  
80. 14-2-18 9:31:57 ================================================================
81.  
82. -- Timers ----------------------------------------------------------------------
83. com.netease.test.metrics.TestTimers.request
84.              count = 8
85.          mean rate = 1.33 calls/second
86.      1-minute rate = 1.40 calls/second
87.      5-minute rate = 1.40 calls/second
88.     15-minute rate = 1.40 calls/second
89.                min = 41.07 milliseconds
90.                max = 968.19 milliseconds
91.               mean = 639.50 milliseconds
92.             stddev = 306.12 milliseconds
93.             median = 692.77 milliseconds
94.               75% <= 885.96 milliseconds
95.               95% <= 968.19 milliseconds
96.               98% <= 968.19 milliseconds
97.               99% <= 968.19 milliseconds
98.             99.9% <= 968.19 milliseconds
99.  
100.  
101. 14-2-18 9:32:00 ================================================================
102.  
103. -- Timers ----------------------------------------------------------------------
104. com.netease.test.metrics.TestTimers.request
105.              count = 15
106.          mean rate = 1.67 calls/second
107.      1-minute rate = 1.40 calls/second
108.      5-minute rate = 1.40 calls/second
109.     15-minute rate = 1.40 calls/second
110.                min = 41.07 milliseconds
111.                max = 968.19 milliseconds
112.               mean = 591.35 milliseconds
113.             stddev = 302.96 milliseconds
114.             median = 650.56 milliseconds
115.               75% <= 838.07 milliseconds
116.               95% <= 968.19 milliseconds
117.               98% <= 968.19 milliseconds
118.               99% <= 968.19 milliseconds
119.             99.9% <= 968.19 milliseconds
120.  
121. */

 

6  Health Checks

Metrics提供了一个独立的模块：Health Checks，用于对Application、其子模块或者关联模块的运行是否正常做检测。该模块是独立metrics-core模块的，使用时则导入metrics-healthchecks包。

<dependency> <groupId>com.codahale.metrics</groupId> <artifactId>metrics-healthchecks</artifactId> <version>3.0.1</version> </dependency>

使用起来和与上述几种类型的Metrics有点类似，但是需要重新实例化一个Metrics容器HealthCheckRegistry，待检测模块继承抽象类HealthCheck并实现check()方法即可，然后将该模块注册到HealthCheckRegistry中，判断的时候通过isHealthy()接口即可。如下示例代码：

 

点击(此处)折叠或打开

1. package com.netease.test.metrics;
2.  
3. import com.codahale.metrics.health.HealthCheck;
4. import com.codahale.metrics.health.HealthCheckRegistry;
5.  
6. import java.util.Map;
7. import java.util.Random;
8.  
9. /**
10.  * User: hzwangxx
11.  * Date: 14-2-18
12.  * Time: 9:57
13.  */
14. public class DatabaseHealthCheck extends HealthCheck{
15.     private final Database database;
16.  
17.     public DatabaseHealthCheck(Database database) {
18.         this.database = database;
19.     }
20.  
21.     @Override
22.     protected Result check() throws Exception {
23.         if (database.ping()) {
24.             return Result.healthy();
25.         }
26.         return Result.unhealthy("Can't ping database.");
27.     }
28.  
29.     /**
30.      * 模拟Database对象
31.      */
32.     static class Database {
33.         /**
34.          * 模拟database的ping方法
35.          * @return 随机返回boolean值
36.          */
37.         public boolean ping() {
38.             Random random = new Random();
39.             return random.nextBoolean();
40.         }
41.     }
42.  
43.     public static void main(String[] args) {
44. // MetricRegistry metrics = new MetricRegistry();
45. // ConsoleReporter reporter = ConsoleReporter.forRegistry(metrics).build();
46.         HealthCheckRegistry registry = new HealthCheckRegistry();
47.         registry.register("database1", new DatabaseHealthCheck(new Database()));
48.         registry.register("database2", new DatabaseHealthCheck(new Database()));
49.         while (true) {
50.             for (Map.Entry<String, Result> entry : registry.runHealthChecks().entrySet()) {
51.                 if (entry.getValue().isHealthy()) {
52.                     System.out.println(entry.getKey() + ": OK");
53.                 } else {
54.                     System.err.println(entry.getKey() + ": FAIL, error message: " + entry.getValue().getMessage());
55.                     final Throwable e = entry.getValue().getError();
56.                     if (e != null) {
57.                         e.printStackTrace();
58.                     }
59.                 }
60.             }
61.             try {
62.                 Thread.sleep(1000);
63.             } catch (InterruptedException e) {
64.  
65.             }
66.         }
67.     }
68. }
69.  
70. /*
71. console output:
72. database1: OK
73. database2: FAIL, error message: Can't ping database.
74. database1: FAIL, error message: Can't ping database.
75. database2: OK
76. database1: OK
77. database2: FAIL, error message: Can't ping database.
78. database1: FAIL, error message: Can't ping database.
79. database2: OK
80. database1: FAIL, error message: Can't ping database.
81. database2: FAIL, error message: Can't ping database.
82. database1: FAIL, error message: Can't ping database.
83. database2: FAIL, error message: Can't ping database.
84. database1: OK
85. database2: OK
86. database1: OK
87. database2: FAIL, error message: Can't ping database.
88. database1: FAIL, error message: Can't ping database.
89. database2: OK
90. database1: OK
91. database2: OK
92. database1: FAIL, error message: Can't ping database.
93. database2: OK
94. database1: OK
95. database2: OK
96. database1: OK
97. database2: OK
98. database1: OK
99. database2: FAIL, error message: Can't ping database.
100. database1: FAIL, error message: Can't ping database.
101. database2: FAIL, error message: Can't ping database.
102.  
103.  */

 

其他支持

metrics提供了对Ehcache、Apache HttpClient、JDBI、Jersey、Jetty、Log4J、Logback、JVM等的集成，可以方便地将Metrics输出到Ganglia、Graphite中，供用户图形化展示。

阅读全文……