Metrics: Real-time monitoring tool for JVM

1. Background

In the past two months of work, I have been monitoring the situation of online applications, and when troubleshooting online problems, I found that there are too many online machines. In order to monitor online applications in an automated and platform-based manner, I chose Metrics. . Metrics is a toolkit for monitoring various metrics of Java services.

2. Introduction

In combination with the project team, the application deployment method of SpringBoot and the project management method of Maven are used. The method of introducing Metrics is as follows:

<dependency>
    <groupId>io.dropwizard.metrics</groupId>
    <artifactId>metrics-core</artifactId>
    <version>3.1.2</version>
</dependency>

1. Basic tools of Metrics

Metrics provides five basic metric types:

Gauges: Metrics

Counters: counters

Histograms: Histogram data

Meters: TPS counters

Timers: counters

MetricRegistry in Metrics is the central container, which is the container of all metrics in the program. All new metrics must be registered in a MetricRegistry instance before they can be used. What needs to be explained here is that in an application, try to keep MetricRegistry as a singleton .

2, Metric Registry container

The code to configure the MetricRegistry container is as follows:

@Bean
public MetricRegistry metrics() {
    return new MetricRegistry();
}

3、Meters

其实将Meters称作TPS计数器并不是那么准确，Meters工具会帮助我们统计系统中某个事件的速率。比如，每秒请求数TPS，每秒查询数QPS等等。这个指标能反应出当时系统的处理能力，帮助我们判断资源是否已经不足等等。Meters本身是一个计数器，并且是自增的。而获取Meters的一个对象meter，或者是实例，如下所示：

@Bean
public Meter requestMeter(MetricRegistry metrics) {
    return metrics.meter("request");
}

在请求中调用mark()方法，来增加计数，代码如下所示。当然，我们可以在不同的请求中添加不同的meter，针对自己的系统完成定制的监控任务。

@RequestMapping("/hello")
@ResponseBody
public String helloWorld() {
    requestMeter.mark();
    return "Hello World";
}

引入meter后，运行应用，会在控制台输出如下信息：

-- Meters ----------------------------------------------------------------------
request
             count = 21055
         mean rate = 133.35 events/second
     1-minute rate = 121.66 events/second
     5-minute rate = 36.99 events/second
    15-minute rate = 13.33 events/second

从上面的console打印出的信息，可以看出meter是为我们提供平均速率，以及采样后的1分钟，5分钟，15分钟。

4、Histogram

直方图是一种很常见的统计图表，Metrics通过Histogram这个类型提供的方便实时的数据绘制成数据直方图。和之前的Meters一样，我们通过在MetricRegistry中注册一个Histogram对象来获取一个对象，代码如下所示：

@Bean
public Histogram responseSizes(MetricRegistry metrics) {
    return metrics.histogram("response-sizes");
}

在应用中，只要在需要统计的地方调用Histogram的update()方法。例如，我们需要统计某个网站的某个方法的流量情况：

responseSizes.update(new Random().nextInt(10));

在console上输出的信息如下：

-- Histograms ------------------------------------------------------------------
response-sizes
             count = 21051
               min = 0
               max = 9
              mean = 4.55
            stddev = 2.88
            median = 4.00
              75% <= 7.00
              95% <= 9.00
              98% <= 9.00
              99% <= 9.00
            99.9% <= 9.00

Histogram提供了最小值、最大值和平均值等数据，利用这些数据，就可以绘制自定义的数据直方图了。

5、Counter

Counter的本质是一个AtomicLong实例，可以增加或者减少值，可以用来统计队列中Job的总数。通过MetricRegistry注册一个Counter对象，如下所示：

@Bean
public Counter pendingJobs(MetricRegistry metrics) {
    return metrics.counter("requestCount");
}

在需要统计数据的位置调用inc()方法和dec()方法，同样地，在console中也会输出具体的信息，如下所示：

// 增加计数
pendingJobs.inc();
// 减去计数
pendingJobs.dec();

-- Counters --------------------------------------------------------------------
requestCount
             count = 21051

这里只是输出了当前度量的值。

6、Timer

Timer其实是一个Meter和Histogram的组合。这个度量单位可以比较方便地统计请求的速率和处理时间。对于接口中调用的延迟等信息的统计就比较方便了。如果发现一个方法的RPS（请求速率）很低，而且平均的处理时间很长，那么这个方法八成出问题了。

同样，在MetricRegistry中注册，获取一个Timer对象，如下所示：

@Bean
public Timer responses(MetricRegistry metrics) {
    return metrics.timer("executeTime");
}

在需要统计信息的位置使用这样的代码：

final Timer.Context context = responses.time();
try {
    // handle request
} finally {
    context.stop();
}

console中就会实时返回这个Timer的信息：

-- Timers ----------------------------------------------------------------------
executeTime
             count = 21061
         mean rate = 133.39 calls/second
     1-minute rate = 122.22 calls/second
     5-minute rate = 37.11 calls/second
    15-minute rate = 13.37 calls/second
               min = 0.00 milliseconds
               max = 0.01 milliseconds
              mean = 0.00 milliseconds
            stddev = 0.00 milliseconds
            median = 0.00 milliseconds
              75% <= 0.00 milliseconds
              95% <= 0.00 milliseconds
              98% <= 0.00 milliseconds
              99% <= 0.00 milliseconds
            99.9% <= 0.01 milliseconds

7、Gauges

除了Metrics提供的几个度量类型，我们可以通过Gauges完成自定义的度量类型。比方说很简单的，我们想看我们缓存里面的数据大小，就可以自己定义一个Gauges，如下所示：

metrics.register(
        MetricRegistry.name(ListManager.class, "cache", "size"),
         (Gauge<Integer>) () -> cache.size()
);

这样Metrics就会一直监控Cache的大小。除此之外有时候，我们需要计算自己定义的一直单位，比如消息队列里面消费者(consumers)消费的速率和生产者(producers)的生产速率的比例，这也是一个度量，可以看下面的代码片段：

public class CompareRatio extends RatioGauge {

    private final Meter consumers;
    private final Meter producers;

    public CacheHitRatio(Meter consumers, Meter producers) {
        this.consumers = consumers;
        this.producers = producers;
    }

    @Override
    protected Ratio getRatio() {
        return Ratio.of(consumers.getOneMinuteRate(),
                producers.getOneMinuteRate());
    }
}

把这个类也注册到Metrics容器里面：

@Bean
public CompareRatio cacheHitRatio(MetricRegistry metrics, Meter requestMeter, 
    Meter producers) {
    CompareRatio compareRatio = new CompareRatio(consumers, producers);
    metrics.register("生产者消费者比率", compareRatio);
    return cacheHitRatio;
}

8、Reporter

Metrics通过报表，将采集的数据展现到不同的位置,这里比如我们注册一个ConsoleReporter到MetricRegistry中，那么console中就会打印出对应的信息。

@Bean
public ConsoleReporter consoleReporter(MetricRegistry metrics) {
    return ConsoleReporter.forRegistry(metrics)
            .convertRatesTo(TimeUnit.SECONDS)
            .convertDurationsTo(TimeUnit.MILLISECONDS)
            .build();
}

除此之外Metrics还支持JMX、HTTP、Slf4j等等，可以访问 http://metrics.dropwizard.io/3.1.0/manual/core/#reporters 来查看Metrics提供的报表，如果还是不能满足自己的业务，也可以自己继承Metrics提供的ScheduledReporter类完成自定义的报表类。原文链接：http://www.jianshu.com/p/e4f70ddbc287。

三、demo

1、配置类MetricConfig

import com.codahale.metrics.*;
import org.slf4j.LoggerFactory;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;

import java.util.concurrent.TimeUnit;

@Configuration
public class MetricConfig {

    @Bean
    public MetricRegistry metrics() {
        return new MetricRegistry();
    }

    /**
     * Reporter 数据的展现位置
     *
     * @param metrics
     * @return
     */
    @Bean
    public ConsoleReporter consoleReporter(MetricRegistry metrics) {
        return ConsoleReporter.forRegistry(metrics)
                .convertRatesTo(TimeUnit.SECONDS)
                .convertDurationsTo(TimeUnit.MILLISECONDS)
                .build();
    }

    @Bean
    public Slf4jReporter slf4jReporter(MetricRegistry metrics) {
        return Slf4jReporter.forRegistry(metrics)
                .outputTo(LoggerFactory.getLogger("demo.metrics"))
                .convertRatesTo(TimeUnit.SECONDS)
                .convertDurationsTo(TimeUnit.MILLISECONDS)
                .build();
    }

    @Bean
    public JmxReporter jmxReporter(MetricRegistry metrics) {
        return JmxReporter.forRegistry(metrics).build();
    }

    /**
     * 自定义单位
     *
     * @param metrics
     * @return
     */
    @Bean
    public ListManager listManager(MetricRegistry metrics) {
        return new ListManager(metrics);
    }

    /**
     * TPS 计算器
     *
     * @param metrics
     * @return
     */
    @Bean
    public Meter requestMeter(MetricRegistry metrics) {
        return metrics.meter("request");
    }

    /**
     * 直方图
     *
     * @param metrics
     * @return
     */
    @Bean
    public Histogram responseSizes(MetricRegistry metrics) {
        return metrics.histogram("response-sizes");
    }

    /**
     * 计数器
     *
     * @param metrics
     * @return
     */
    @Bean
    public Counter pendingJobs(MetricRegistry metrics) {
        return metrics.counter("requestCount");
    }

    /**
     * 计时器
     *
     * @param metrics
     * @return
     */
    @Bean
    public Timer responses(MetricRegistry metrics) {
        return metrics.timer("executeTime");
    }
}

2、接收请求的类MainController

import com.codahale.metrics.Counter;
import com.codahale.metrics.Histogram;
import com.codahale.metrics.Meter;
import com.codahale.metrics.Timer;
import demo.metrics.config.ListManager;
import org.springframework.stereotype.Controller;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.ResponseBody;
import java.util.Random;
import javax.annotation.Resource;

@Controller
@RequestMapping("/")
public class MainController {

    @Resource
    private Meter requestMeter;

    @Resource
    private Histogram responseSizes;

    @Resource
    private Counter pendingJobs;

    @Resource
    private Timer responses;

    @Resource
    private ListManager listManager;

    @RequestMapping("/hello")
    @ResponseBody
    public String helloWorld() {

        requestMeter.mark();

        pendingJobs.inc();

        responseSizes.update(new Random().nextInt(10));

        listManager.getList().add(1);

        final Timer.Context context = responses.time();
        try {
            return "Hello World";
        } finally {
            context.stop();
        }
    }
}

3、应用运行类DemoApplication

import com.codahale.metrics.ConsoleReporter;
import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;
import org.springframework.context.ApplicationContext;
import java.util.concurrent.TimeUnit;

@SpringBootApplication
public class DemoApplication {
    public static void main(String[] args) {
        ApplicationContext ctx = SpringApplication.run(DemoApplication.class, args);

        // 启动Reporter
        ConsoleReporter reporter = ctx.getBean(ConsoleReporter.class);
        reporter.start(1, TimeUnit.SECONDS);

    }
}

Metrics: Real-time monitoring tool for JVM

Guess you like