Datadog DogStatsD implements the StatsD protocol with some differences. DogStatsD enables you to send metrics and monitor your application code without blocking it. Data is transmitted from your application through UDP to the local DogStatsD server (embedded in the Datadog Agent), which aggregates and then sends it to Datadog’s API endpoint. Read more about the DogStatsD setup.

This article describes why and how the aggregation is performed over your data.

Why aggregate metrics?

Aggregation improves performance by reducing the number of API calls, each of which takes a certain amount of time.

Consider a COUNT metric that is incremented 1,000 times (+1 each time) over a short amount of time. Instead of making 1,000 separate API calls, the DogStatsD server aggregates it into a few API calls. Depending on the situation (see below), the library may submit—for instance—1 datapoint with value 1,000 or X aggregate datapoints with a cumulated value of 1,000.

How is aggregation performed with the DogStatsD server?

DogStatsD uses a flush interval of 10 seconds. Every 10 seconds, DogStatsD checks all data received since the last flush. All values that correspond to the same metric name and the same tags are aggregated together into a single value.

Note: With the StatsD protocol, the StatsD client doesn’t send metrics with timestamps. The timestamp is added at the flush time. So for a flush occurring at 10:00:10, all data received by the DogStatsD server (embedded in the Datadog Agent) between 10:00:00 and 10:00:10 is rolled up in a single datapoint that gets 10:00:00 as timestamp.

Aggregation rules per metric type

Among all values received during the same flush interval, the aggregated value send depends on the metric type:

Further Reading

Additional helpful documentation, links, and articles:

Introduction to DogStatsDDOCUMENTATION Official and Community created API and DogStatsD client librariesDOCUMENTATION