Metrics reporter, which reports to elasticsearch
Clone or download
Pull request Compare This branch is 18 commits ahead of elastic:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
samples
src
.gitignore
.travis.yml
LICENSE
README.md
pom.xml

README.md

Metrics Elasticsearch Reporter

Introduction

This is a reporter for the excellent Metrics library, similar to the Graphite or Ganglia reporters, except that it reports to an Elasticsearch server.

In case, you are worried, that you need to include the 20MB elasticsearch dependency in your project, you do not need to be. As this reporter is using HTTP for putting data into elasticsearch, the only library needed is the awesome Jackson JSON library, more exactly the Jackson Databind library to easily serialize the metrics objects.

If you want to see this in action, go to the samples/ directory and read the readme over there, to get up and running with a sample application using the Metrics library as well as a dashboard application to graph.

Compatibility

Metrics-elasticsearch-reporter elasticsearch Release date
6.0.0-SNAPSHOT 6.x NONE
2.3.0 2.3.0 -> 5.x NONE
2.2.0 2.2.0 -> 2.2.x 2016-02-10
2.0 1.0.0 -> 1.7.x 2014-02-16
1.0 0.90.7 -> 0.90.x 2014-02-05

Travis CI build status

Build status

Installation

You can simply add a dependency in your pom.xml (or whatever dependency resolution system you might have)

<dependency>
  <groupId>org.elasticsearch</groupId>
  <artifactId>metrics-elasticsearch-reporter</artifactId>
  <version>2.2.0</version>
</dependency>

Configuration

final MetricRegistry registry = new MetricRegistry();
ElasticsearchReporter reporter = ElasticsearchReporter.forRegistry(registry)
    .hosts("localhost:9200", "localhost:9201")
    .build();
reporter.start(60, TimeUnit.SECONDS);

Define your metrics and registries as usual

private final Meter incomingRequestsMeter = registry.meter("incoming-http-requests");

// in your app code
incomingRequestsMeter.mark(1);

Options

  • hosts(): A list of hosts used to connect to, must be in the format hostname:port, default is localhost:9200
  • timeout(): Milliseconds to wait for an established connections, before the next host in the list is tried. Defaults to 1000
  • bulkSize(): Defines how many metrics are sent per bulk requests, defaults to 2500
  • filter(): A MetricFilter to define which metrics written to the elasticsearch
  • percolationFilter(): A MetricFilter to define which metrics should be percolated against. See below for an example
  • percolationNotifier(): An implementation of the Notifier interface, which is executed upon a matching percolator. See below for an example.
  • index(): The name of the index to write to, defaults to metrics
  • indexDateFormat(): The date format to make sure to rotate to a new index, defaults to yyyy-MM
  • timestampFieldname(): The field name of the timestamp, defaults to @timestamp, which makes it easy to use with kibana

Mapping

Note: The reporter automatically checks for the existence of an index template called metrics_template. If this template does not exist, it is created. This template ensures that all strings used in metrics are set to not_analyzed and disables the _all field.

Notifications with percolations

ElasticsearchReporter reporter = ElasticsearchReporter.forRegistry(registry)
    .percolationNotifier(new PagerNotifier())
    .percolationFilter(MetricFilter.ALL)
    .build();
reporter.start(60, TimeUnit.SECONDS);

Write a custom notifier

public class PagerNotifier implements Notifier {

  @Override
  public void notify(JsonMetrics.JsonMetric jsonMetric, String percolateMatcher) {
    // send pager duty here
  }
}

Add a percolation

curl http://localhost:9200/metrics/.percolator/http-monitor -X PUT -d '{
  "query" : { 
    "bool" : { 
      "must": [
        { "term": { "name" : "incoming-http-requests" } },
        { "range": { "m1_rate": { "to" : "10" } } }
      ]
    }
  }
}'

JSON Format of metrics

This is how the serialized metrics looks like in elasticsearch

Counter

{
  "name": "usa-gov-heartbearts",
  "@timestamp": "2013-07-20T09:29:58.000+0000",
  "count": 18
}

Timer

{
  "name" : "bulk-request-timer",
  "@timestamp" : "2013-07-20T09:43:58.000+0000",
  "count" : 114,
  "max" : 109.681,
  "mean" : 5.439666666666667,
  "min" : 2.457,
  "p50" : 4.3389999999999995,
  "p75" : 5.0169999999999995,
  "p95" : 8.37175,
  "p98" : 9.6832,
  "p99" : 94.68429999999942,
  "p999" : 109.681,
  "stddev" : 9.956913151098842,
  "m15_rate" : 0.10779994503690074,
  "m1_rate" : 0.07283351433589833,
  "m5_rate" : 0.10101298115113727,
  "mean_rate" : 0.08251056571678642,
  "duration_units" : "milliseconds",
  "rate_units" : "calls/second"
}

Meter

{
  "name" : "usagov-incoming-requests",
  "@timestamp" : "2013-07-20T09:29:58.000+0000",
  "count" : 224,
  "m1_rate" : 0.3236309568191993,
  "m5_rate" : 0.45207208204948995,
  "m15_rate" : 0.5014348927301423,
  "mean_rate" : 0.4135529888278531,
  "units" : "events/second"
}

Histogram

{
  "name" : "my-histgram",
  "@timestamp" : "2013-07-20T09:29:58.000+0000",
  "count" : 114,
  "max" : 109.681,
  "mean" : 5.439666666666667,
  "min" : 2.457,
  "p50" : 4.3389999999999995,
  "p75" : 5.0169999999999995,
  "p95" : 8.37175,
  "p98" : 9.6832,
  "p99" : 94.68429999999942,
  "p999" : 109.681,
  "stddev" : 9.956913151098842,}
}

Gauge

{
  "name" : "usagov-incoming-requests",
  "@timestamp" : "2013-07-20T09:29:58.000+0000",
  "value" : 123
}

Spark & Beam Integration

An Apache Spark Sink implementation is provided. As part of this implementation support is added to grab Apache Beam metrics automatically within a Spark runner.

Enabling and Configuring Spark Sink