metrics-datadog
is a simple reporting bridge between Dropwizard Metrics and the Datadog service. It includes support for:
- Datadog's tagging feature
- Metric reporting via either UDP (dogstatsd) or the Datadog HTTP API
- Tight integration with the Dropwizard framework via the
dropwizard-metrics-datadog
sub-project.
Datadog supports two main metric ingestion methods:
- POSTing metrics via their HTTP API
- Sending metrics via UDP (using a statsd-like protocol) to the local dogstatsd agent
Datadog recommends the dogstatsd
UDP-based approach, but some may prefer the HTTP-based approach
for various reasons e.g. a general adversity to running agents, the additional memory required by the agent and
forwarder (though this is configurable), stability, security or other environment/platform-level
conflicts.
Note that, in the event of a delivery failure, the HTTP-based transport does not buffer metrics in memory. It will attempt a handful of retries and then give up. Hence, when faced with an extended network partition window or a Datadog ingestion outage, some metrics will certainly be lost using this transport. That said, note that the UDP-based reporter also cannot buffer metrics forever due to memory constraints.
import org.coursera.metrics.datadog.DatadogReporter
import org.coursera.metrics.datadog.DatadogReporter.Expansion._
import org.coursera.metrics.datadog.transport.Transport
import org.coursera.metrics.datadog.transport.HttpTransport
import org.coursera.metrics.datadog.transport.UdpTransport
...
val expansions = EnumSet.of(COUNT, RATE_1_MINUTE, RATE_15_MINUTE, MEDIAN, P95, P99)
val httpTransport = new HttpTransport.Builder().withApiKey(apiKey).build()
val reporter = DatadogReporter.forRegistry(registry)
.withEC2Host()
.withTransport(httpTransport)
.withExpansions(expansions)
.build()
reporter.start(10, TimeUnit.SECONDS)
Example of using UDP transport:
...
val udpTransport = new UdpTransport.Builder().build()
val reporter =
...
.withTransport(udpTransport)
...
Datadog supports powerful tagging
functionality while the Metrics API does not. Thus, metrics-datadog
utilizes
a special, overloaded metric naming syntax that enables tags to piggyback on
metric names while passing through the Metrics library. The tags are unpacked
by metrics-datadog
at reporting time and are sent along to Datadog via the
configured transport layer. Here's the metric name syntax:
metricName[tagName:tagValue,tagName:tagValue,...]
metrics-datadog
is mainly a reporting library and doesn't currently
implement a tag-aware decorator on top of the core Metrics
API. It
does, however, expose a TaggedName
class that helps you encode/decode tags in
metric names using the syntax above. You can utilize this helper class
methods when registering and recording metrics. Note that in order for tag
propagation to work, you'll need to use our DefaultMetricNameFormatter
(or a formatter with compatible parsing logic).
We also support the notion of static, "additional tags". This feature allows
you to define a set of tags that are appended to all metrics sent through
the reporter. It's useful for setting static tags such as the
environment, service name or version. Additional tags are configured via
the DatadogReporter
constructor.
Finally, we support the notion of "dynamic tags". By implementing and
registering a DynamicTagsCallback
with DatadogReporter
, you can control
the values of "additional tags" at runtime. Dynamic tags are merged with
and override any additional tags set.
Performance note: Heavy use of tagging, especially tags values with high cardinality, can dramatically increase memory usage, as all tag permutations are tracked and counted in-memory by the Metrics library. Also note that some MetricRegistry APIs do defensive copies on the entire metrics set, which can be prohibitively expensive CPU and memory-wise if you have a huge, heavily tagged metric set.
If you have a dropwizard project and have at least dropwizard-core
0.7.X,
then you can perform the following steps to automatically report metrics to
datadog.
First, add the dropwizard-metrics-datadog
dependency in your POM:
<dependency>
<groupId>org.coursera</groupId>
<artifactId>dropwizard-metrics-datadog</artifactId>
<version>1.1.6</version>
</dependency>
Then just add the following to your dropwizard
YAML config file.
metrics:
frequency: 1 minute # Default is 1 second.
reporters:
- type: datadog
host: <host> # Optional with UDP Transport
tags: # Optional. Defaults to (empty)
includes: # Optional. Defaults to (all).
excludes: # Optional. Defaults to (none).
prefix: # Optional. Defaults to (none).
expansions: # Optional. Defaults to (all).
metricNameFormatter: # Optional. Default is "default".
dynamicTagsCallback: # Optional. Defaults to (none).
transport:
type: http
apiKey: <apiKey>
connectTimeout: <duration> # Optional. Default is 5 seconds
socketTimeout: <duration> # Optional. Default is 5 seconds
Once your dropwizard
application starts, your metrics should start appearing
in Datadog.
HTTP Transport:
metrics:
frequency: 1 minute # Default is 1 second.
reporters:
- type: datadog
host: <host>
transport:
type: http
apiKey: <apiKey>
connectTimeout: <duration> # Optional. Default is 5 seconds
socketTimeout: <duration> # Optional. Default is 5 seconds
UDP Transport:
metrics:
frequency: 1 minute # Default is 1 second.
reporters:
- type: datadog
transport:
type: udp
prefix: # Optional. Default is (empty)
statsdHost: "localhost" # Optional. Default is "localhost"
port: 8125 # Optional. Default is 8125
If you want to filter only a few metrics, you can use the includes
or
excludes
key to create a set of metrics to include or exclude respectively.
metrics:
frequency: 1 minute # Default is 1 second.
reporters:
- type: datadog
host: <host>
includes:
- jvm.
- ch.
The check is very simplistic so be as specific as possible. For example, if you have "jvm.", the filter will check if the includes has that value in any part of the metric name (not just the beginning).
If you want to limit the set of expansions applied to each metric, you can specify a custom set.
The full set of expansions can be found in the Expansion enum.
metrics:
reporters:
- type: datadog
expansions:
- COUNT
- RATE_1_MINUTE
- MAX
- P95
By default, the metric names are sent as-is (e.g. io.dropwizard.jetty.MutableServletContextHandler.2xx-responses
)
The prefix option adds a custom prefix to each metric name:
metrics:
reporters:
- type: datadog
prefix: custom.prefix
would produce: custom.prefix.io.dropwizard.jetty.MutableServletContextHandler.2xx-responses
The metricNameFormatter option can be used to add custom logic when processing each metric's name. By default it will use the DefaultMetricNameFormatter which handles Datadog tags but does not modify the metric name.
metrics:
reporters:
- type: datadog
metricNameFormatter: custom
Adding a custom formatter requires a few things:
public class CustomMetricNameFormatter extends DefaultMetricNameFormatter {
@Override
public String format(String name, String... path) {
// Make response metrics names less verbose
String newName = name.replace("io.dropwizard.jetty.MutableServletContextHandler", "");
// Call DefaultMetricNameFormatter to handle tags
return super.format(newName, path);
}
}
@JsonTypeName("custom") // This must match the name specified in the configuration
public class CustomMetricNameFormatterFactory implements MetricNameFormatterFactory {
@Override
public MetricNameFormatter build() {
return new CustomMetricNameFormatter();
}
}
We need to make sure our CustomMetricNameFormatterFactory
is added to the list of subTypes
for MetricNameFormatterFactory
, otherwise the "custom"
in our config won't be recognized.
Add a file called org.coursera.metrics.datadog.MetricNameFormatterFactory
to
src/main/resources/META-INF/services
and add the full path to your class to the file
(e.g. com.company.CustomMetricNameFormatterFactory
)
See: http://www.dropwizard.io/1.0.0/docs/manual/configuration.html#polymorphic-configuration for details
Similar to the MetricNameFormatter
steps, we need to:
- Create a DynamicTagsCallback
- Create a DynamicTagsCallbackFactory with
@JsonTypeName
annotation - Add the Factory to
org.coursera.metrics.datadog.DynamicTagsCallbackFactory
file
See above instructions for details.
Metrics datadog reporter is available as an artifact on Maven Central
- Group: org.coursera
- Artifact: metrics-datadog
- Version: 1.1.6
Dropwizard datadog reporter is available as an artifact on Maven Central
- Group: org.coursera
- Artifact: dropwizard-metrics-datadog
- Version: 1.1.6
We follow Google's Java Code Style