README.md

Fast Java Statsd Client

A Statsd client library implemented in Java. Allows Java applications to be easily and efficiently monitored by publishing metrics via statsd protocol. The client implements all metric types defined by the statsd spec and few extra to support some extensions like DataDog.

Our main goal was to be fast and do no allocations, and indeed, the library literally produces zero garbage in runtime.

Installation

The client jar is distributed via Maven central, and can be imported like this:

<dependency>
    <groupId>tech.energyit</groupId>
    <artifactId>fast-java-statsd-client</artifactId>
    <version>1.0</version>
</dependency>

or for Gradle :

dependencies {
        implementation "tech.energyit:fast-java-statsd-client:1.0"
}

Usage

// create a sender - provides lot's of options how to send data via UDP, but this is likely the best for you:
SynchronousSender sender = SynchronousSender.builder()
                          .withHostAndPort(STATSD_SERVER_HOST, STATSD_SERVER_PORT)
                          .build();
// create client and set prefix
StatsDClient client = new FastStatsDClient("tradeApp", sender);

// monitor your app
client.count(ORDER_COUNT, count, productTag);
client.time(PROCESSING_TIME, durationInMs, productTag);
// ...  

// close the sender with your app
sender.close();

For more details see SampleMonitorApp.java

How Fast Is It ?

You best find for yourself. To make it easy for you, we prepared StatsdClientBenchmark.java which compares different setups of FastStatsDClient.java and also with the most used java client from DataDog. Just make sure you have some statsd server listening on the port (e.g. by starting DummyStatsDServer.java )

If no statsd server is listening on the port, DatagramChannels generate IOExceptions which is expensive. If the client uses an ErrorHandler that also generates a stacktrace, the impact can be big, and the benched results will be misleading.

This is the result running with openjdk version "11.0.7":

Benchmark                                                                                       Mode  Cnt     Score   Error   Units
StatsdClientBenchmark.countLongsViaSyncFastClient                                               avgt          3.078           us/op
StatsdClientBenchmark.countLongsViaSyncFastClient:·gc.alloc.rate.norm                           avgt          ≈ 10⁻³            B/op
StatsdClientBenchmark.countLongsViaSyncFastClient:·gc.count                                     avgt            ≈ 0          counts

StatsdClientBenchmark.countDoublesViaSyncFastClient                                             avgt          3.083           us/op
StatsdClientBenchmark.countDoublesViaSyncFastClient:·gc.alloc.rate.norm                         avgt          ≈ 10⁻³            B/op
StatsdClientBenchmark.countDoublesViaSyncFastClient:·gc.count                                   avgt            ≈ 0          counts

StatsdClientBenchmark.countDoublesViaSyncFastClientUsingExactDoubles                            avgt          3.503           us/op
StatsdClientBenchmark.countDoublesViaSyncFastClientUsingExactDoubles:·gc.alloc.rate.norm        avgt         80.009            B/op
StatsdClientBenchmark.countDoublesViaSyncFastClientUsingExactDoubles:·gc.count                  avgt          1.000          counts

StatsdClientBenchmark.countLongsViaAsyncFastClient                                              avgt          0.119           us/op
StatsdClientBenchmark.countLongsViaAsyncFastClient:·gc.alloc.rate.norm                          avgt          0.029            B/op
StatsdClientBenchmark.countLongsViaAsyncFastClient:·gc.count                                    avgt          7.000          counts

StatsdClientBenchmark.countLongsViaAsyncFastClientWithFallbackToSyncSender                      avgt          5.674           us/op
StatsdClientBenchmark.countLongsViaAsyncFastClientWithFallbackToSyncSender:·gc.alloc.rate.norm  avgt          2.191            B/op
StatsdClientBenchmark.countLongsViaAsyncFastClientWithFallbackToSyncSender:·gc.count            avgt            ≈ 0          counts

StatsdClientBenchmark.countLongsViaDatadogClient                                                avgt          0.559           us/op
StatsdClientBenchmark.countLongsViaDatadogClient:·gc.alloc.rate.norm                            avgt        400.000            B/op
StatsdClientBenchmark.countLongsViaDatadogClient:·gc.count                                      avgt         25.000          counts