Prometheus instrumentation library for JVM applications
Apache License 2.0

Prometheus JVM Client

It supports Java, Clojure, Scala, JRuby, and anything else that runs on the JVM.

Build Status

Table of Contents



If you use Maven, you can simply reference the assets below. The latest version can be found on in the maven repository for io.prometheus.

<!-- The client -->
<!-- Hotspot JVM metrics-->
<!-- Exposition HTTPServer-->
<!-- Pushgateway exposition-->


There are canonical examples defined in the class definition Javadoc of the client packages.

Documentation can be found at the Java Client Github Project Page.

Disabling _created metrics

By default, counters, histograms, and summaries export an additional series suffixed with _created and a value of the unix timestamp for when the metric was created. If this information is not helpful, it can be disabled by setting the environment variable PROMETHEUS_DISABLE_CREATED_SERIES=true.


Four types of metrics are offered: Counter, Gauge, Summary and Histogram. See the documentation on metric types and instrumentation best practices on how to use them.


Counters go up, and reset when the process restarts.

import io.prometheus.client.Counter;
class YourClass {
  static final Counter requests =
     .name("requests_total").help("Total requests.").register();

  void processRequest() {;
    // Your code here.


Gauges can go up and down.

class YourClass {
  static final Gauge inprogressRequests =
     .name("inprogress_requests").help("Inprogress requests.").register();

  void processRequest() {;
    // Your code here.

There are utilities for common use cases:

gauge.setToCurrentTime(); // Set to current unixtime.

As an advanced use case, a Gauge can also take its value from a callback by using the setChild() method. Keep in mind that the default inc(), dec() and set() methods on Gauge take care of thread safety, so when using this approach ensure the value you are reporting accounts for concurrency.


Summaries and Histograms can both be used to monitor distributions, like latencies or request sizes.

An overview of when to use Summaries and when to use Histograms can be found on

The following example shows how to measure latencies and request sizes:

class YourClass {

  private static final Summary requestLatency =
      .help("request latency in seconds")

  private static final Summary receivedBytes =
      .help("request size in bytes")

  public void processRequest(Request req) {
    Summary.Timer requestTimer = requestLatency.startTimer();
    try {
      // Your code here.
    } finally {

The Summary class provides different utility methods for observing values, like observe(double), startTimer(); timer.observeDuration(), time(Callable), etc.

By default, Summary metrics provide the count and the sum. For example, if you measure latencies of a REST service, the count will tell you how often the REST service was called, and the sum will tell you the total aggregated response time. You can calculate the average response time using a Prometheus query dividing sum / count.

In addition to count and sum, you can configure a Summary to provide quantiles:

Summary requestLatency =
    .help("Request latency in seconds.")
    .quantile(0.5, 0.01)    // 0.5 quantile (median) with 0.01 allowed error
    .quantile(0.95, 0.005)  // 0.95 quantile with 0.005 allowed error
    // ...

As an example, a 0.95 quantile of 120ms tells you that 95% of the calls were faster than 120ms, and 5% of the calls were slower than 120ms.

Tracking exact quantiles require a large amount of memory, because all observations need to be stored in a sorted list. Therefore, we allow an error to significantly reduce memory usage.

In the example, the allowed error of 0.005 means that you will not get the exact 0.95 quantile, but anything between the 0.945 quantile and the 0.955 quantile.

Experiments show that the Summary typically needs to keep less than 100 samples to provide that precision, even if you have hundreds of millions of observations.

There are a few special cases:

  • You can set an allowed error of 0, but then the Summary will keep all observations in memory.
  • You can track the minimum value with .quantile(0, 0). This special case will not use additional memory even though the allowed error is 0.
  • You can track the maximum value with .quantile(1, 0). This special case will not use additional memory even though the allowed error is 0.

Typically, you don't want to have a Summary representing the entire runtime of the application, but you want to look at a reasonable time interval. Summary metrics implement a configurable sliding time window:

Summary requestLatency =
    .help("Request latency in seconds.")
    .maxAgeSeconds(10 * 60)
    // ...

The default is a time window of 10 minutes and 5 age buckets, i.e. the time window is 10 minutes wide, and * we slide it forward every 2 minutes.


Like Summaries, Histograms can be used to monitor latencies (or other things like request sizes).

An overview of when to use Summaries and when to use Histograms can be found on

class YourClass {
  static final Histogram requestLatency =
     .name("requests_latency_seconds").help("Request latency in seconds.").register();

  void processRequest(Request req) {
    Histogram.Timer requestTimer = requestLatency.startTimer();
    try {
      // Your code here.
    } finally {

The default buckets are intended to cover a typical web/rpc request from milliseconds to seconds. They can be overridden with the buckets() method on the Histogram.Builder.

There are utilities for timing code:

class YourClass {
  static final Histogram requestLatency =
     .name("requests_latency_seconds").help("Request latency in seconds.").register();

  void processRequest(Request req) {
    requestLatency.time(new Runnable() {
      public abstract void run() {
        // Your code here.

    // Or the Java 8 lambda equivalent
    requestLatency.time(() -> {
      // Your code here.


All metrics can have labels, allowing grouping of related time series.

See the best practices on naming and labels.

Taking a counter as an example:

class YourClass {
  static final Counter requests =
     .name("my_library_requests_total").help("Total requests.")

  void processGetRequest() {
    // Your code here.

Registering Metrics

The best way to register a metric is via a static final class variable as is common with loggers.

static final Counter requests =
   .name("my_library_requests_total").help("Total requests.").labelNames("path").register();

Using the default registry with variables that are static is ideal since registering a metric with the same name is not allowed and the default registry is also itself static. You can think of registering a metric, more like registering a definition (as in the TYPE and HELP sections). The metric 'definition' internally holds the samples that are reported and pulled out by Prometheus. Here is an example of registering a metric that has no labels.

class YourClass {
  static final Gauge activeTransactions =
     .help("Active transactions.")

  void processThatCalculates(String key) {;
    try {
        // Perform work.
    } finally{

To create timeseries with labels, include labelNames() with the builder. The labels() method looks up or creates the corresponding labelled timeseries. You might also consider storing the labelled timeseries as an instance variable if it is appropriate. It is thread safe and can be used multiple times, which can help performance.

class YourClass {
  static final Counter calculationsCounter =
     .name("my_library_calculations_total").help("Total calls.")

  void processThatCalculates(String key) {
    // Run calculations.


Exemplars are a feature of the OpenMetrics format that allows applications to link metrics to example traces. In order to see exemplars, you need to set the Accept header for the OpenMetrics format like this:

curl -H 'Accept: application/openmetrics-text; version=1.0.0; charset=utf-8' http://localhost:8080/metrics

Exemplars are supported since client_java version 0.11.0. Exemplars are supported for Counter and Histogram metrics.

Global Exemplar Samplers

An ExemplarSampler is used implicitly under the hood to add exemplars to your metrics. The idea is that you get exemplars without changing your code.

The DefaultExemplarSampler comes with built-in support for OpenTelemetry tracing, see built-in support for tracing systems below.

You can disable the default exemplar samplers globally with:


You can set your own custom implementation of ExemplarSampler as a global default like this:

ExemplarSampler myExemplarSampler = new MyExemplarSampler();

Per Metric Exemplar Samplers

The metric builders for Counter and Histogram have methods for setting the exemplar sampler for that individual metric. This takes precedence over the global setting in ExemplarConfig.

The following calls enable the default exemplar sampler for individual metrics. This is useful if you disabled the exemplar sampler globally with ExemplarConfig.disableExemplars().

Counter myCounter =
Histogram myHistogram =

The following calls disable exemplars for individual metrics:

Counter myCounter =
Histogram myHistogram =

The following calls enable exemplars and set a custom MyExemplarSampler for individual metrics:

// CounterExemplarSampler
Counter myCounter =
    .withExemplarSampler(new MyExemplarSampler())
// HistogramExemplarSampler
Histogram myHistogram =
    .withExemplarSampler(new MyExemplarSampler())

Per Observation Exemplars

You can explicitly provide an exemplar for an individual observation. This takes precedence over the exemplar sampler configured with the metric.

The following call will increment a counter, and create an exemplar with the specified span_id and trace_id labels:

myCounter.incWithExemplar("span_id", "abcdef", "trace_id", "123456");

The following call will observe a value of 0.12 in a histogram, and create an exemplar with the specified span_id and trace_id labels:

myHistogram.observeWithExemplar(0.12, "span_id", "abcdef", "trace_id", "123456");

All methods for observing and incrementing values have ...withExemplar equivalents. There are versions taking the exemplar labels as a String... as shown in the example, as well as versions taking the exemplar labels as a Map<String, String>.

Built-in Support for Tracing Systems

The DefaultExemplarSampler detects if a tracing library is found on startup, and provides exemplars for that tracing library by default. Currently, only OpenTelemetry tracing is supported. If you are a tracing vendor, feel free to open a PR and add support for your tracing library.

Documentation of the individual tracer integrations:

Included Collectors

The Java client includes collectors for garbage collection, memory pools, classloading, and thread counts. These can be added individually or just use the DefaultExports to conveniently register them.



There are logging collectors for log4j, log4j2 and logback.

To register the Logback collector can be added to the root level like so:

<?xml version="1.0" encoding="UTF-8"?>
    <include resource="org/springframework/boot/logging/logback/base.xml"/>

    <appender name="METRICS" class="io.prometheus.client.logback.InstrumentedAppender" />

    <root level="INFO">
        <appender-ref ref="METRICS" />


To register the log4j collector at root level:

<?xml version="1.0" encoding="UTF-8" ?>
<!DOCTYPE log4j:configuration SYSTEM "log4j.dtd">
<log4j:configuration xmlns:log4j="">
    <appender name="METRICS" class="io.prometheus.client.log4j.InstrumentedAppender"/>
        <priority value ="info" />
        <appender-ref ref="METRICS" />

To register the log4j2 collector at root level:

<?xml version="1.0" encoding="UTF-8"?>
<Configuration packages="io.prometheus.client.log4j2">
        <Prometheus name="METRICS"/>
        <Root level="info">
            <AppenderRef ref="METRICS"/>

See ./integration_tests/it_log4j2/ for a log4j2 example.


To register the Guava cache collector, be certain to add recordStats() when building the cache and adding it to the registered collector.

CacheMetricsCollector cacheMetrics = new CacheMetricsCollector().register();

Cache<String, String> cache = CacheBuilder.newBuilder().recordStats().build();
cacheMetrics.addCache("myCacheLabel", cache);

The Caffeine equivalent is nearly identical. Again, be certain to call recordStats() when building the cache so that metrics are collected.

CacheMetricsCollector cacheMetrics = new CacheMetricsCollector().register();

Cache<String, String> cache = Caffeine.newBuilder().recordStats().build();
cacheMetrics.addCache("myCacheLabel", cache);


There is a collector for Hibernate which allows to collect metrics from one or more SessionFactory instances.

If you want to collect metrics from a single SessionFactory, you can register the collector like this:

new HibernateStatisticsCollector(sessionFactory, "myapp").register();

In some situations you may want to collect metrics from multiple factories. In this case just call add() on the collector for each of them.

new HibernateStatisticsCollector()
    .add(sessionFactory1, "myapp1")
    .add(sessionFactory2, "myapp2")

If you are using Hibernate in a JPA environment and only have access to the EntityManager or EntityManagerFactory, you can use this code to access the underlying SessionFactory:

SessionFactory sessionFactory = entityManagerFactory.unwrap(SessionFactory.class);


SessionFactory sessionFactory = entityManager.unwrap(Session.class).getSessionFactory();


There is a collector for recording various Jetty server metrics. You can do it by registering the collector like this:

// Configure StatisticsHandler.
StatisticsHandler stats = new StatisticsHandler();
// Register collector.
new JettyStatisticsCollector(stats).register();

Also, you can collect QueuedThreadPool metrics. If there is a single QueuedThreadPool to keep track of, use the following:

new QueuedThreadPoolStatisticsCollector(queuedThreadPool, "myapp").register();

If you want to collect multiple QueuedThreadPool metrics, also you can achieve it like this:

new QueuedThreadPoolStatisticsCollector()
    .add(queuedThreadPool1, "myapp1")
    .add(queuedThreadPool2, "myapp2")

Servlet Filter

There is a servlet filter available for measuring the duration taken by servlet requests:

  • javax version: io.prometheus.client.filter.MetricsFilter provided by simpleclient_servlet
  • jakarta version: io.prometheus.client.servlet.jakarta.filter.MetricsFilter provided by simpleclient_servlet_jakarta

Both versions implement the same functionality.

Configuration is as follows:

The metric-name init parameter is required, and is the name of the metric prometheus will expose for the timing metrics. Help text via the help init parameter is not required, although it is highly recommended. The number of buckets is overridable, and can be configured by passing a comma-separated string of doubles as the buckets init parameter. The granularity of path measuring is also configurable, via the path-components init parameter. By default, the servlet filter will record each path differently, but by setting an integer here, you can tell the filter to only record up to the Nth slashes. That is, all requests with greater than N "/" characters in the servlet URI path will be measured in the same bucket, and you will lose that granularity. The init parameter strip-context-path can be used to strip the leading part of the URL which is part of the deployment context (i.e. the folder the servlet is deployed to), so that the same servlet deployed to different paths can lead to similar metrics.

The code below is an example of the XML configuration for the filter. You will need to place this (replace your own values) code in your webapp/WEB-INF/web.xml file.

  <!-- This example shows the javax version. For Jakarta you would use -->
  <!-- <filter-class>io.prometheus.client.filter.servlet.jakarta.MetricsFilter</filter-class> -->
  <!-- help is optional, defaults to the message below -->
    <param-value>This is the help for your metrics filter</param-value>
  <!-- buckets is optional, unless specified the default buckets from io.prometheus.client.Histogram are used -->
  <!-- path-components is optional, anything less than 1 (1 is the default) means full granularity -->
  <!-- strip-context-path is optional, defaults to false -->

<!-- You will most likely want this to be the first filter in the chain
(therefore the first <filter-mapping> in the web.xml file), so that you can get
the most accurate measurement of latency. -->

Additionally, you can instantiate your servlet filter directly in Java code. To do this, you just need to call the non-empty constructor. The first parameter, the metric name, is required. The second, help, is optional but highly recommended. The other parameters are optional and will default sensibly if omitted.

Spring AOP

There is a Spring AOP collector that allows you to annotate methods that you would like to instrument with a Summary, but without going through the process of manually instantiating and registering your metrics classes. To use the metrics annotations, simply add simpleclient_spring_web as a dependency, annotate a configuration class with @EnablePrometheusTiming, then annotate your Spring components as such:

public class MyController {
  @PrometheusTimeMethod(name = "my_controller_path_duration_seconds", help = "Some helpful info here")
  public Object handleMain() {
    // Do something


There are several options for exporting metrics.


Metrics are usually exposed over HTTP, to be read by the Prometheus server.

There are HTTPServer, Servlet, SpringBoot, and Vert.x integrations included in the client library. The simplest of these is the HTTPServer:

HTTPServer server = new HTTPServer.Builder()

The HTTPServer.Builder supports configuration of a SampleNameFilter which can be used to restrict the time series being exported by name.

To add Prometheus exposition to an existing HTTP server using servlets, see the MetricsServlet. It also serves as a simple example of how to write a custom endpoint.

To expose the metrics used in your code, you would add the Prometheus servlet to your Jetty server:

Server server = new Server(1234);
ServletContextHandler context = new ServletContextHandler();

context.addServlet(new ServletHolder(new MetricsServlet()), "/metrics");

Like the HTTPServer, the MetricsServlet can be configured with a SampleNameFilter which can be used to restrict the time series being exported by name. See integration_tests/servlet_jakarta_exporter_webxml/ for an example how to configure this in web.xml.

All HTTP exposition integrations support restricting which time series to return using ?name[]= URL parameters. Due to implementation limitations, this may have false negatives.

Exporting to a Pushgateway

The Pushgateway allows ephemeral and batch jobs to expose their metrics to Prometheus.

void executeBatchJob() throws Exception {
  CollectorRegistry registry = new CollectorRegistry();
  Gauge duration =
     .name("my_batch_job_duration_seconds").help("Duration of my batch job in seconds.").register(registry);
  Gauge.Timer durationTimer = duration.startTimer();
  try {
    // Your code here.

    // This is only added to the registry after success,
    // so that a previous success in the Pushgateway isn't overwritten on failure.
    Gauge lastSuccess =
       .name("my_batch_job_last_success").help("Last time my batch job succeeded, in unixtime.").register(registry);
  } finally {
    PushGateway pg = new PushGateway("");
    pg.pushAdd(registry, "my_batch_job");

A separate registry is used, as the default registry may contain other metrics such as those from the Process Collector. See the Pushgateway documentation for more information.

with Basic Auth

PushGateway pushgateway = new PushGateway("");
pushgateway.setConnectionFactory(new BasicAuthHttpConnectionFactory("my_user", "my_password"));

with Custom Connection Preparation Logic

PushGateway pushgateway = new PushGateway("");
pushgateway.setConnectionFactory(new MyHttpConnectionFactory());


class MyHttpConnectionFactory implements HttpConnectionFactory {
    public HttpURLConnection create(String url) throws IOException {
        HttpURLConnection connection = (HttpURLConnection) new URL(url).openConnection();
        // add any connection preparation logic you need
        return connection;


It is also possible to expose metrics to systems other than Prometheus. This allows you to take advantage of Prometheus instrumentation even if you are not quite ready to fully transition to Prometheus yet.


Metrics are pushed over TCP in the Graphite plaintext format.

Graphite g = new Graphite("localhost", 2003);
// Push the default registry once.

// Push the default registry every 60 seconds.
Thread thread = g.start(CollectorRegistry.defaultRegistry, 60);
// Stop pushing.

Custom Collectors

Sometimes it is not possible to directly instrument code, as it is not in your control. This requires you to proxy metrics from other systems.

To do so you need to create a custom collector (which will need to be registered as a normal metric), for example:

class YourCustomCollector extends Collector {
  public List<MetricFamilySamples> collect() {
    List<MetricFamilySamples> mfs = new ArrayList<MetricFamilySamples>();
    // With no labels.
    mfs.add(new GaugeMetricFamily("my_gauge", "help", 42));
    // With labels
    GaugeMetricFamily labeledGauge = new GaugeMetricFamily("my_other_gauge", "help", Arrays.asList("labelname"));
    labeledGauge.addMetric(Arrays.asList("foo"), 4);
    labeledGauge.addMetric(Arrays.asList("bar"), 5);
    return mfs;

// Registration
static final YourCustomCollector requests = new YourCustomCollector().register()

SummaryMetricFamily works similarly.

A collector may implement a describe method which returns metrics in the same format as collect (though you don't have to include the samples). This is used to predetermine the names of time series a CollectorRegistry exposes and thus to detect collisions and duplicate registrations.

Usually custom collectors do not have to implement describe. If describe is not implemented and the CollectorRegistry was created with auto_describe=True (which is the case for the default registry) then collect will be called at registration time instead of describe. If this could cause problems, either implement a proper describe, or if that's not practical have describe return an empty list.

DropwizardExports Collector

DropwizardExports collector is available to proxy metrics from Dropwizard.

// Dropwizard MetricRegistry
MetricRegistry metricRegistry = new MetricRegistry();
new DropwizardExports(metricRegistry).register();

By default Dropwizard metrics are translated to Prometheus sample sanitizing their names, i.e. replacing unsupported chars with _, for example:

Dropwizard metric name:
Prometheus metric:

It is also possible add custom labels and name to newly created Samples by using a CustomMappingSampleBuilder with custom MapperConfigs:

// Dropwizard MetricRegistry
MetricRegistry metricRegistry = new MetricRegistry();
MapperConfig config = new MapperConfig();
// The match field in MapperConfig is a simplified glob expression that only allows * wildcard.
// The new Sample's template name.
Map<String, String> labels = new HashMap<String,String>();
// ... more configs
// Labels to be extracted from the metric. Key=label name. Value=label template
labels.put("name", "${0}");
labels.put("status", "${1}");

SampleBuilder sampleBuilder = new CustomMappingSampleBuilder(Arrays.asList(config));
new DropwizardExports(metricRegistry, sampleBuilder).register();

When a new metric comes to the collector, MapperConfigs are scanned to find the first one that matches the incoming metric name. The name set in the configuration will be used and labels will be extracted. Using the CustomMappingSampleBuilder in the previous example leads to the following result:

Dropwizard metric name
Prometheus metric

Template with placeholders can be used both as names and label values. Placeholders are in the ${n} format where n is the zero based index of the Dropwizard metric name wildcard group we want to extract.


The Prometheus Users Mailinglist is the best place to ask questions.

Details for those wishing to develop the library can be found on the wiki