The aggregation processor is based on the OpenTelemetry Collector Contrib interval processor. It aggregates a subset of metric types over a configurable interval and periodically forwards the result to the next component in the pipeline. Metrics that are not aggregated are passed through unchanged.

This implementation aggregates non-monotonic delta sums and histograms and keeps latest gauge values in the aggregation interval.

The behavior per metric type is:

Aggregation semantics:

• Gauges keep the value of the data point with the most recent timestamp ("last value wins").
• Non-monotonic delta sums add data point values together within a stream and keep the most recent timestamp.
• Delta histograms add bucket counts, total count, and sum, and merge min/max across data points in a stream. If two data points in the same interval have different explicit bucket boundaries, only the one with the most recent timestamp is kept.

The following settings can be optionally configured:

atlassian_aggregation:
  # The interval at which the processor exports the aggregated metrics. Must be > 0.
  [ interval: <duration> | default = 60s ]

The following metrics come into the processor to be handled:

The processor immediately passes the following metric to the next component in the pipeline, because monotonic sums are not aggregated:

At the next interval (60s by default), the processor exports the aggregated test_metric data points (values aggregated per attribute set, keeping the latest timestamp):

Note: After exporting, the internal aggregation state is cleared. If no new metrics arrive, the next interval exports nothing.