SignalFx client libraries

This repository contains libraries for instrumenting Java applications and
reporting metrics to SignalFx. You will need a SignalFx account and organization
API token to use them. For more information on SignalFx and to create an
account, go to http://www.signalfx.com.

We recommend sending metrics with Java using Codahale Metrics version 3.0+. You
can also use Yammer Metrics 2.0.x (an earlier version of Codahale Metrics). More
information on the Codahale Metrics library can be found on the
Codahale Metrics website.

You can also use the module signalfx-java to send metrics directly to SignalFx
using protocol buffers, without using Codahale or Yammer metrics.

Supported languages

Java 6+ with signalfx-metrics.

Using this library in your project

With Maven

If you're using Maven, add the following to your project's pom.xml file.

To work with Codahale 3.0.x:

<dependency>
  <groupId>com.signalfx.public</groupId>
  <artifactId>signalfx-codahale</artifactId>
  <version>0.0.44</version>
</dependency>

To work with Yammer Metrics 2.0.x:

<dependency>
<groupId>com.signalfx.public</groupId>
  <artifactId>signalfx-yammer</artifactId>
  <version>0.0.44</version>
</dependency>

With SBT

If you're using SBT, add the following to your project's build.sbt file.

To work with Codahale 3.0.x:

libraryDependencies += "com.signalfx.public" % "signalfx-codahale" % "0.0.44"

To work with Yammer Metrics 2.0.x:

libraryDependencies += "com.signalfx.public" % "signalfx-yammer" % "0.0.44"

From source

You can also install this library from source by cloning the repo and using
mvn install as follows. However, we strongly recommend using the automated
mechanisms described above.

$ git clone https://github.com/signalfx/signalfx-java.git
Cloning into 'signalfx-java'...
remote: Counting objects: 930, done.
remote: Compressing objects: 100% (67/67), done.
remote: Total 930 (delta 20), reused 0 (delta 0)
Receiving objects: 100% (930/930), 146.79 KiB | 0 bytes/s, done.
Resolving deltas: 100% (289/289), done.
Checking connectivity... done.
$ cd signalfx-java
$ mvn install
[INFO] Scanning for projects...
...
...
...
[INFO] SignalFx parent .................................. SUCCESS [  2.483 s]
[INFO] SignalFx Protocol Buffer definitions ............. SUCCESS [  5.503 s]
[INFO] SignalFx Protobuf Utilities ...................... SUCCESS [  2.269 s]
[INFO] SignalFx java libraries .......................... SUCCESS [  3.728 s]
[INFO] Codahale to SignalFx ............................. SUCCESS [  2.910 s]
[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 17.120 s
[INFO] ------------------------------------------------------------------------

Sending metrics

Codahale Metrics 3.0.x

1. Set up the Codahale reporter

final MetricRegistry metricRegistry = new MetricRegistry();
final SignalFxReporter signalfxReporter = new SignalFxReporter.Builder(
    metricRegistry,
    "SIGNALFX_AUTH_TOKEN"
).build();
signalfxReporter.start(1, TimeUnit.SECONDS);
final MetricMetadata metricMetadata = signalfxReporter.getMetricMetadata();

2. Send a metric

// This will send the current time in ms to SignalFx as a gauge
metricRegistry.register("gauge", new Gauge<Long>() {
    public Long getValue() {
        return System.currentTimeMillis();
    }
});

3. Add existing dimensions and metadata to metrics

You can add SignalFx specific metadata to Codahale metrics by first gathering
available metadata using getMetricMetadata(), then attaching the
MetricMetadata to the metric.

When you use MetricMetadata, call the .register() method you get from the call
forMetric() rather than registering your metric directly with the
metricRegistry. This will construct a unique Codahale string for your metric.

/*
 * This will send the size of a queue as a gauge, and attach dimension
 * 'queue_name' to the gauge.
 */
final Queue customerQueue = new ArrayBlockingQueue(100);
metricMetadata.forMetric(new Gauge<Long>() {
    @Override
    public Long getValue() {
        return customerQueue.size();
    }
}).withDimension("queue_name", "customer_backlog")
  .register(metricRegistry);

4. (optional) Add dimensions without knowing if they already exist

We recommend creating your Codahale object as a field of your class, as a
counter or gauge, then using that field to increment values. If you don't want
to maintain this for reasons of code cleanliness, you can create it on the fly
with our builders.

For example, if you wanted a timer that included a dimension indicating which
store it is from, you could use code like this.

Timer t = metricMetadata
    .forBuilder(MetricBuilder.TIMERS)
    .withMetricName("request_time")
    .withDimension("storename", "electronics")
    .createOrGet(metricRegistery);

Timer.Context c = t.time();
try {
    System.out.println("Doing store things");
} finally {
    c.close();
}

/*
 * Java 7 alternative:
 *
 * try (Timer.Context ignored = t.time()) {
 *     System.out.println("Doing store things");
 * }
 */

After setting up Codahale

After setting up a SignalFxReporter, you can use Codahale metrics as you
normally would, reported at the frequency configured by the SignalFxReporter.

Yammer Metrics

You can also use this library with Yammer metrics 2.0.x as shown in the
following examples.

1. Set up Yammer metrics

final MetricRegistry metricRegistry = new MetricRegistry();
final SignalFxReporter signalfxReporter = new SignalFxReporter.Builder(
    metricRegistery,
    "SIGNALFX_AUTH_TOKEN"
).build();
signalfxReporter.start(1, TimeUnit.SECONDS);
final MetricMetadata metricMetadata = signalfxReporter.getMetricMetadata();

2. Send a metric with Yammer metrics

// This will send the current time in ms to SignalFx as a gauge
MetricName gaugeName = new MetricName("group", "type", "gauge");
Metric gauge = metricRegistry.newGauge(gaugeName, new Gauge<Long>() {
    @Override
    public Long value() {
        return System.currentTimeMillis();
    }
});

3. Add Dimensions and SignalFx metadata to Yammer metrics

Use the MetricMetadata of the reporter as shown.

final Queue customerQueue = new ArrayBlockingQueue(100);

MetricName gaugeName = new MetricName("group", "type", "gauge");
Metric gauge = metricRegistry.newGauge(gaugeName, new Gauge<Integer>() {
    @Override
    public Integer value() {
        return customerQueue.size();
    }
});

metricMetadata.forMetric(gauge)
    .withDimension("queue_name", "customer_backlog");

4. Adding dimensions without knowing if they already exist

This is not supported in Yammer Metrics 2.0.x.

Changing the default source

The default source name for metrics is discovered by [SourceNameHelper]
(signalfx-java/src/main/java/com/signalfx/metrics/SourceNameHelper.java).
If you want to override the default behavior, you can pass a third parameter to
your Builder and that String is then used as the source.

For example:

final SignalFxReporter signalfxReporter = new SignalFxReporter.Builder(
    metricRegistry,
    "SIGNALFX_AUTH_TOKEN",
    "MYHOST1"
).build();

Default dimensions

Sometimes there is a desire to set one or more dimension key/value pairs
on every datapoint that is reported by this library. In order to do this
call addDimension(String key, String value) or
addDimensions(Map<String,String> dimensions) on the
SignalFxReporter.Builder object.

Note that if IncrementalCounter is used to create a distributed
counter you will want to make sure that none of the dimensions passed to
addDimension()/addDimensions() are unique to the reporting source
(e.g. hostname, AWSUniqueId) as this will make make the counter
non-distributed. For such dimensions use
addUniqueDimension()/addUniqueDimensions() on the
SignalFxReporter.Builder object.

AWS Integration

To enable AWS integration in SignalFx (i.e aws tag/property syncing) to a metric
you can use com.signalfx.metrics.aws.AWSInstanceInfo. And either add it as
a dimension in MetricMetadata or add it as a default dimension.

String instanceInfo = AWSInstanceInfo.get()
Timer t = metricMetadata
    .forBuilder(MetricBuilder.TIMERS)
    .withMetricName("request_time")
    .withDimension(AWSInstanceInfo.DIMENSION_NAME, instanceInfo)
    .createOrGet(metricRegistery);

/**
 * As default dimension
 */
final SignalFxReporter signalfxReporter = new SignalFxReporter.Builder(
    metricRegistry,
    "SIGNALFX_AUTH_TOKEN"
).addUniqueDimension(AWSInstanceInfo.DIMENSION_NAME, instanceInfo).build();

Sending metrics without using Codahale

We recommend sending metrics using Codahale as shown above. You can also
interact with our Java library directly if you do not want to use Codahale. To
do this, you will need to build the metric manually using protocol buffers as
shown in the following example. Sending both datapoints and events are now
supported using protocol buffers.

SignalFxReceiverEndpoint signalFxEndpoint = new SignalFxEndpoint();
AggregateMetricSender mf = new AggregateMetricSender("test.SendMetrics",
    new HttpDataPointProtobufReceiverFactory(signalFxEndpoint).setVersion(2),
    new HttpEventProtobufReceiverFactory(signalFxEndpoint),
    new StaticAuthToken(auth_token),
    Collections.<OnSendErrorHandler> singleton(new OnSendErrorHandler() {
        @Override
        public void handleError(MetricError metricError) {
            System.out.println("Unable to POST metrics: " + metricError.getMessage());
        }
    }));

try (AggregateMetricSender.Session i = mf.createSession()) {
    i.setDatapoint(
        SignalFxProtocolBuffers.DataPoint.newBuilder()
            .setMetric("curtime")
            .setMetricType(SignalFxProtocolBuffers.MetricType.GAUGE)
            .setValue(
                SignalFxProtocolBuffers.Datum.newBuilder()
                    .setIntValue(System.currentTimeMillis()))
            .addDimensions(
                SignalFxProtocolBuffers.Dimension.newBuilder()
                    .setKey("source")
                    .setValue("java"))
            .build());

    i.setEvent(
            SignalFxProtocolBuffers.Event.newBuilder()
                  .setEventType("Deployments")
                  .setCategory(SignalFxProtocolBuffers.EventCategory.USER_DEFINED)
                  .setTimestamp(System.currentTimeMillis())
                  .addDimensions(
                    SignalFxProtocolBuffers.Dimension.newBuilder()
                                      .setKey("source")
                                      .setValue("java"))
                  .addProperties(
                    SignalFxProtocolBuffers.Property.newBuilder()
                                    .setKey("version")
                                    .setValue(
                                            SignalFxProtocolBuffers.PropertyValue.newBuilder()
                                                    .setIntValue(2)
                                                    .build())
                                    .build())
                  .build());
}

Sending metrics through a HTTP proxy

To send metrics through a HTTP proxy one can set the standard java system
properties used to control HTTP protocol handling. There are 3 properties you
can set to specify the proxy that will be used by the HTTP protocol handler:

http.proxyHost: the host name of the proxy server
http.proxyPort: the port number, the default value being 80.
http.nonProxyHosts: a list of hosts that should be reached directly, bypassing
the proxy. This is a list of regular expressions separated by |. Any host
matching one of these regular expressions will be reached through a direct
connection instead of through a proxy.

Basic example:

$ java -Dhttp.proxyHost=webcache.mydomain.com -Dhttp.proxyPort=8080”

Example with directive to bypass proxy for localhost and host.mydomain.com:

$ java -Dhttp.proxyHost=webcache.mydomain.com -Dhttp.proxyPort=8080 -Dhttp.noProxyHosts=”localhost|host.mydomain.com”

Disabling compression when sending datapoints

By default, the Java library compresses datapoint payloads when sending them to
SignalFx. This can provide significant egress volume savings when sending data
to SignalFx's ingest API. This behavior can be disabled by setting the
com.signalfx.public.java.disableHttpCompression system property to true:

$ java -Dcom.signalfx.public.java.disableHttpCompression=true ...

Example Project

You can find a full-stack example project called "signalfx-java-examples" in
the repo.

Run it as follows:

Download the code and create an "auth" file in the "signalfx-java-examples"
directory. The auth file should contain the following:
```
auth=<signalfx API Token>
host=https://ingest.signalfx.com
```

Run the following commands in your terminal to install and run the example
project, replacing path/to/signalfx-java-examples with the location of the
example project code in your environment. You must have Maven installed.

cd path/to/signalfx-java-examples
mvn install
# an example for Yammer 2.x metrics
mvn exec:java -Dexec.mainClass="com.signalfx.example.YammerExample"
# an example for sending datapoints and events using protocol buffers
mvn exec:java -Dexec.mainClass="com.signalfx.example.ProtobufExample"

New metrics and events from the example project should appear in SignalFx.

Executing SignalFlow computations

SignalFlow is SignalFx's real-time analytics computation language. The
SignalFlow API allows SignalFx users to execute real-time streaming analytics
computations on the SignalFx platform. For more information, head over to our
Developers documentation:

Executing a SignalFlow program is very simple with this client library:

String program = "data('cpu.utilization').mean().publish()";
SignalFlowClient flow = new SignalFlowClient("MY_TOKEN");
System.out.println("Executing " + program);
Computation computation = flow.execute(program);
for (ChannelMessage message : computation) {
    switch (message.getType()) {
    case DATA_MESSAGE:
        DataMessage dataMessage = (DataMessage) message;
        System.out.printf("%d: %s%n",
                dataMessage.getLogicalTimestampMs(), dataMessage.getData());
        break;

    case EVENT_MESSAGE:
        EventMessage eventMessage = (EventMessage) message;
        System.out.printf("%d: %s%n",
                eventMessage.getTimestampMs(),
                eventMessage.getProperties());
        break;
    }
}

Metadata about the timeseries is received from the iterable stream, and it
is also automatically intercepted by the client library and made available through
the Computation object returned by execute():

case DATA_MESSAGE:
    DataMessage dataMessage = (DataMessage) message;
    for (Map<String, Number> datum : dataMessage.getData()) {
        Map<String,Object> metadata = computation.getMetadata(datum.getKey());
        // ...
    }

tlisonbee/signalfx-java