Introducción a las métricas de Dropwizard

1. Introducción

Metrics es una biblioteca de Java que proporciona instrumentos de medición para aplicaciones Java.

Tiene varios módulos, y en este artículo, elaboraremos el módulo de métricas-núcleo, el módulo de métricas-comprobaciones de estado, el módulo de métricas-servlets y el módulo de métricas-servlet, y bosquejaremos el resto, para su referencia.

2. Módulo de métricas básicas

2.1. Dependencias de Maven

Para usar el módulo metrics-core , solo se requiere una dependencia que debe agregarse al archivo pom.xml :

 io.dropwizard.metrics metrics-core 3.1.2  

Y puedes encontrar su última versión aquí.

2.2. MetricRegistry

En pocas palabras, usaremos la clase MetricRegistry para registrar una o varias métricas.

Podemos usar un registro de métricas para todas nuestras métricas, pero si queremos usar diferentes métodos de informes para diferentes métricas, también podemos dividir nuestras métricas en grupos y usar diferentes registros de métricas para cada grupo.

Creemos un MetricRegistry ahora:

MetricRegistry metricRegistry = new MetricRegistry();

Y luego podemos registrar algunas métricas con este MetricRegistry :

Meter meter1 = new Meter(); metricRegistry.register("meter1", meter1); Meter meter2 = metricRegistry.meter("meter2"); 

Hay dos formas básicas de crear una nueva métrica: crear una instancia usted mismo u obtener una del registro de métricas. Como puede ver, usamos ambos en el ejemplo anterior, estamos instanciando el objeto Meter "meter1" y obtenemos otro objeto Meter "meter2" que es creado por metricRegistry .

En un registro de métricas, cada métrica tiene un nombre único, ya que usamos "meter1" y "meter2" como nombres de métricas anteriores. MetricRegistry también proporciona un conjunto de métodos auxiliares estáticos para ayudarnos a crear nombres de métricas adecuados:

String name1 = MetricRegistry.name(Filter.class, "request", "count"); String name2 = MetricRegistry.name("CustomFilter", "response", "count"); 

Si necesitamos administrar un conjunto de registros de métricas, podemos usar la clase SharedMetricRegistries , que es un singleton y seguro para subprocesos. Podemos agregarle un registro de métricas, recuperar este registro de métricas y eliminarlo:

SharedMetricRegistries.add("default", metricRegistry); MetricRegistry retrievedMetricRegistry = SharedMetricRegistries.getOrCreate("default"); SharedMetricRegistries.remove("default"); 

3. Conceptos de métricas

El módulo metrics-core proporciona varios tipos de métricas de uso común: medidor , indicador , contador , histograma y temporizador , y reportero para generar valores de métricas .

3.1. Metro

Un medidor mide el recuento y la tasa de ocurrencias de eventos:

Meter meter = new Meter(); long initCount = meter.getCount(); assertThat(initCount, equalTo(0L)); meter.mark(); assertThat(meter.getCount(), equalTo(1L)); meter.mark(20); assertThat(meter.getCount(), equalTo(21L)); double meanRate = meter.getMeanRate(); double oneMinRate = meter.getOneMinuteRate(); double fiveMinRate = meter.getFiveMinuteRate(); double fifteenMinRate = meter.getFifteenMinuteRate(); 

El método getCount () devuelve el recuento de sucesos y el método mark () agrega 1 on al recuento de sucesos. El objeto Medidor proporciona cuatro tasas que representan tasas promedio para toda la vida útil del medidor , para el último minuto, para los últimos cinco minutos y para el trimestre reciente, respectivamente.

3.2. Calibre

Gauge es una interfaz que se usa simplemente para devolver un valor particular. El módulo de metrics-core proporciona varias implementaciones del mismo: RatioGauge , CachedGauge , DerivativeGauge y JmxAttributeGauge .

RatioGauge es una clase abstracta y mide la relación de un valor a otro.

Veamos cómo usarlo. Primero, implementamos una clase AttendanceRatioGauge :

public class AttendanceRatioGauge extends RatioGauge { private int attendanceCount; private int courseCount; @Override protected Ratio getRatio() { return Ratio.of(attendanceCount, courseCount); } // standard constructors } 

Y luego lo probamos:

RatioGauge ratioGauge = new AttendanceRatioGauge(15, 20); assertThat(ratioGauge.getValue(), equalTo(0.75)); 

CachedGauge es otra clase abstracta que puede almacenar en caché el valor, por lo tanto, es bastante útil cuando los valores son costosos de calcular. Para usarlo, necesitamos implementar una clase ActiveUsersGauge :

public class ActiveUsersGauge extends CachedGauge
    
      { @Override protected List loadValue() { return getActiveUserCount(); } private List getActiveUserCount() { List result = new ArrayList(); result.add(12L); return result; } // standard constructors }
    

Luego lo probamos para ver si funciona como se esperaba:

Gauge
    
      activeUsersGauge = new ActiveUsersGauge(15, TimeUnit.MINUTES); List expected = new ArrayList(); expected.add(12L); assertThat(activeUsersGauge.getValue(), equalTo(expected)); 
    

Establecemos el tiempo de vencimiento de la caché en 15 minutos al crear una instancia de ActiveUsersGauge .

DerivativeGauge también es una clase abstracta y le permite derivar un valor de otro Gauge como su valor.

Veamos un ejemplo:

public class ActiveUserCountGauge extends DerivativeGauge
    
      { @Override protected Integer transform(List value) { return value.size(); } // standard constructors }
    

This Gauge derives its value from an ActiveUsersGauge, so we expect it to be the value from the base list's size:

Gauge
    
      activeUsersGauge = new ActiveUsersGauge(15, TimeUnit.MINUTES); Gauge activeUserCountGauge = new ActiveUserCountGauge(activeUsersGauge); assertThat(activeUserCountGauge.getValue(), equalTo(1)); 
    

JmxAttributeGauge is used when we need to access other libraries' metrics exposed via JMX.

3.3. Counter

The Counter is used for recording incrementations and decrementations:

Counter counter = new Counter(); long initCount = counter.getCount(); assertThat(initCount, equalTo(0L)); counter.inc(); assertThat(counter.getCount(), equalTo(1L)); counter.inc(11); assertThat(counter.getCount(), equalTo(12L)); counter.dec(); assertThat(counter.getCount(), equalTo(11L)); counter.dec(6); assertThat(counter.getCount(), equalTo(5L));

3.4. Histogram

Histogram is used for keeping track of a stream of Long values and it analyzes their statistical characteristics such as max, min, mean, median, standard deviation, 75th percentile and so on:

Histogram histogram = new Histogram(new UniformReservoir()); histogram.update(5); long count1 = histogram.getCount(); assertThat(count1, equalTo(1L)); Snapshot snapshot1 = histogram.getSnapshot(); assertThat(snapshot1.getValues().length, equalTo(1)); assertThat(snapshot1.getValues()[0], equalTo(5L)); histogram.update(20); long count2 = histogram.getCount(); assertThat(count2, equalTo(2L)); Snapshot snapshot2 = histogram.getSnapshot(); assertThat(snapshot2.getValues().length, equalTo(2)); assertThat(snapshot2.getValues()[1], equalTo(20L)); assertThat(snapshot2.getMax(), equalTo(20L)); assertThat(snapshot2.getMean(), equalTo(12.5)); assertEquals(10.6, snapshot2.getStdDev(), 0.1); assertThat(snapshot2.get75thPercentile(), equalTo(20.0)); assertThat(snapshot2.get999thPercentile(), equalTo(20.0)); 

Histogram samples the data by using reservoir sampling, and when we instantiate a Histogram object, we need to set its reservoir explicitly.

Reservoir is an interface and metrics-core provides four implementations of them: ExponentiallyDecayingReservoir, UniformReservoir, SlidingTimeWindowReservoir, SlidingWindowReservoir.

In the section above, we mentioned that a metric can also be created by MetricRegistry, besides using a constructor. When we use metricRegistry.histogram(), it returns a Histogram instance with ExponentiallyDecayingReservoir implementation.

3.5. Timer

Timer is used for keeping track of multiple timing durations which are represented by Context objects, and it also provides their statistical data:

Timer timer = new Timer(); Timer.Context context1 = timer.time(); TimeUnit.SECONDS.sleep(5); long elapsed1 = context1.stop(); assertEquals(5000000000L, elapsed1, 1000000); assertThat(timer.getCount(), equalTo(1L)); assertEquals(0.2, timer.getMeanRate(), 0.1); Timer.Context context2 = timer.time(); TimeUnit.SECONDS.sleep(2); context2.close(); assertThat(timer.getCount(), equalTo(2L)); assertEquals(0.3, timer.getMeanRate(), 0.1); 

3.6. Reporter

When we need to output our measurements, we can use Reporter. This is an interface, and the metrics-core module provides several implementations of it, such as ConsoleReporter, CsvReporter, Slf4jReporter, JmxReporter and so on.

Here we use ConsoleReporter as an example:

MetricRegistry metricRegistry = new MetricRegistry(); Meter meter = metricRegistry.meter("meter"); meter.mark(); meter.mark(200); Histogram histogram = metricRegistry.histogram("histogram"); histogram.update(12); histogram.update(17); Counter counter = metricRegistry.counter("counter"); counter.inc(); counter.dec(); ConsoleReporter reporter = ConsoleReporter.forRegistry(metricRegistry).build(); reporter.start(5, TimeUnit.MICROSECONDS); reporter.report(); 

Here is the sample output of the ConsoleReporter:

-- Histograms ------------------------------------------------------------------ histogram count = 2 min = 12 max = 17 mean = 14.50 stddev = 2.50 median = 17.00 75% <= 17.00 95% <= 17.00 98% <= 17.00 99% <= 17.00 99.9% <= 17.00 -- Meters ---------------------------------------------------------------------- meter count = 201 mean rate = 1756.87 events/second 1-minute rate = 0.00 events/second 5-minute rate = 0.00 events/second 15-minute rate = 0.00 events/second 

4. Module metrics-healthchecks

Metrics has an extension metrics-healthchecks module for dealing with health checks.

4.1. Maven Dependencies

To use the metrics-healthchecks module, we need to add this dependency to the pom.xml file:

 io.dropwizard.metrics metrics-healthchecks 3.1.2 

And you can find its latest version here.

4.2. Usage

First, we need several classes which are responsible for specific health check operations, and these classes must implement HealthCheck.

For example, we use DatabaseHealthCheck and UserCenterHealthCheck:

public class DatabaseHealthCheck extends HealthCheck { @Override protected Result check() throws Exception { return Result.healthy(); } } 
public class UserCenterHealthCheck extends HealthCheck { @Override protected Result check() throws Exception { return Result.healthy(); } } 

Then, we need a HealthCheckRegistry (which is just like MetricRegistry), and register the DatabaseHealthCheck and UserCenterHealthCheck with it:

HealthCheckRegistry healthCheckRegistry = new HealthCheckRegistry(); healthCheckRegistry.register("db", new DatabaseHealthCheck()); healthCheckRegistry.register("uc", new UserCenterHealthCheck()); assertThat(healthCheckRegistry.getNames().size(), equalTo(2)); 

We can also unregister the HealthCheck:

healthCheckRegistry.unregister("uc"); assertThat(healthCheckRegistry.getNames().size(), equalTo(1)); 

We can run all the HealthCheck instances:

Map results = healthCheckRegistry.runHealthChecks(); for (Map.Entry entry : results.entrySet()) { assertThat(entry.getValue().isHealthy(), equalTo(true)); } 

Finally, we can run a specific HealthCheck instance:

healthCheckRegistry.runHealthCheck("db"); 

5. Module metrics-servlets

Metrics provides us a handful of useful servlets which allow us to access metrics related data through HTTP requests.

5.1. Maven Dependencies

To use the metrics-servlets module, we need to add this dependency to the pom.xml file:

 io.dropwizard.metrics metrics-servlets 3.1.2 

And you can find its latest version here.

5.2. HealthCheckServlet Usage

HealthCheckServlet provides health check results. First, we need to create a ServletContextListener which exposes our HealthCheckRegistry:

public class MyHealthCheckServletContextListener extends HealthCheckServlet.ContextListener { public static HealthCheckRegistry HEALTH_CHECK_REGISTRY = new HealthCheckRegistry(); static { HEALTH_CHECK_REGISTRY.register("db", new DatabaseHealthCheck()); } @Override protected HealthCheckRegistry getHealthCheckRegistry() { return HEALTH_CHECK_REGISTRY; } } 

Then, we add both this listener and HealthCheckServlet into the web.xml file:

 com.baeldung.metrics.servlets.MyHealthCheckServletContextListener   healthCheck com.codahale.metrics.servlets.HealthCheckServlet   healthCheck /healthcheck 

Now we can start the web application, and send a GET request to “//localhost:8080/healthcheck” to get health check results. Its response should be like this:

{ "db": { "healthy": true } }

5.3. ThreadDumpServlet Usage

ThreadDumpServlet provides information about all live threads in the JVM, their states, their stack traces, and the state of any locks they may be waiting for.

If we want to use it, we simply need to add these into the web.xml file:

 threadDump com.codahale.metrics.servlets.ThreadDumpServlet   threadDump /threaddump 

Thread dump data will be available at “//localhost:8080/threaddump”.

5.4. PingServlet Usage

PingServlet can be used to test if the application is running. We add these into the web.xml file:

 ping com.codahale.metrics.servlets.PingServlet   ping /ping 

And then send a GET request to “//localhost:8080/ping”. The response's status code is 200 and its content is “pong”.

5.5. MetricsServlet Usage

MetricsServlet provides metrics data. First, we need to create a ServletContextListener which exposes our MetricRegistry:

public class MyMetricsServletContextListener extends MetricsServlet.ContextListener { private static MetricRegistry METRIC_REGISTRY = new MetricRegistry(); static { Counter counter = METRIC_REGISTRY.counter("m01-counter"); counter.inc(); Histogram histogram = METRIC_REGISTRY.histogram("m02-histogram"); histogram.update(5); histogram.update(20); histogram.update(100); } @Override protected MetricRegistry getMetricRegistry() { return METRIC_REGISTRY; } } 

Both this listener and MetricsServlet need to be added into web.xml:

 com.codahale.metrics.servlets.MyMetricsServletContextListener   metrics com.codahale.metrics.servlets.MetricsServlet   metrics /metrics 

This will be exposed in our web application at “//localhost:8080/metrics”. Its response should contain various metrics data:

{ "version": "3.0.0", "gauges": {}, "counters": { "m01-counter": { "count": 1 } }, "histograms": { "m02-histogram": { "count": 3, "max": 100, "mean": 41.66666666666666, "min": 5, "p50": 20, "p75": 100, "p95": 100, "p98": 100, "p99": 100, "p999": 100, "stddev": 41.69998667732268 } }, "meters": {}, "timers": {} } 

5.6. AdminServlet Usage

AdminServlet aggregates HealthCheckServlet, ThreadDumpServlet, MetricsServlet, and PingServlet.

Let's add these into the web.xml:

 admin com.codahale.metrics.servlets.AdminServlet   admin /admin/* 

It can now be accessed at “//localhost:8080/admin”. We will get a page containing four links, one for each of those four servlets.

Note that, if we want to do health check and access metrics data, those two listeners are still needed.

6. Module metrics-servlet

The metrics-servlet module provides a Filter which has several metrics: meters for status codes, a counter for the number of active requests, and a timer for request duration.

6.1. Maven Dependencies

To use this module, let's first add the dependency into the pom.xml:

 io.dropwizard.metrics metrics-servlet 3.1.2 

And you can find its latest version here.

6.2. Usage

To use it, we need to create a ServletContextListener which exposes our MetricRegistry to the InstrumentedFilter:

public class MyInstrumentedFilterContextListener extends InstrumentedFilterContextListener { public static MetricRegistry REGISTRY = new MetricRegistry(); @Override protected MetricRegistry getMetricRegistry() { return REGISTRY; } } 

Then, we add these into the web.xml:

  com.baeldung.metrics.servlet.MyInstrumentedFilterContextListener    instrumentFilter  com.codahale.metrics.servlet.InstrumentedFilter    instrumentFilter /* 

Now the InstrumentedFilter can work. If we want to access its metrics data, we can do it through its MetricRegistryREGISTRY.

7. Other Modules

Except for the modules we introduced above, Metrics has some other modules for different purposes:

  • metrics-jvm: provides several useful metrics for instrumenting JVM internals
  • metrics-ehcache: provides InstrumentedEhcache, a decorator for Ehcache caches
  • metrics-httpclient: provides classes for instrumenting Apache HttpClient (4.x version)
  • metrics-log4j: provides InstrumentedAppender, a Log4j Appender implementation for log4j 1.x which records the rate of logged events by their logging level
  • metrics-log4j2: is similar to metrics-log4j, just for log4j 2.x
  • metrics-logback: provides InstrumentedAppender, a Logback Appender implementation which records the rate of logged events by their logging level
  • metrics-json : proporciona HealthCheckModule y MetricsModule para Jackson

Además, además de estos módulos principales del proyecto, algunas otras bibliotecas de terceros proporcionan integración con otras bibliotecas y marcos.

8. Conclusión

Instrumentar aplicaciones es un requisito común, por lo que en este artículo presentamos Metrics, con la esperanza de que pueda ayudarlo a resolver su problema.

Como siempre, el código fuente completo del ejemplo está disponible en GitHub.