Reduce Resolution processor

This is a custom processor that reduces all received metrics within one time interval into a set of statics that describe what happened in that same period.

How are metrics reduced

This processor grabs a group of received metrics, and summarizes each metric with a group of statistics. The processor does not batch by itself. Therefore, to aggregate metrics for a window of 30 seconds, for example, it is necessary to add a batch processor first with a timeout of 30 seconds before the processor. For example:

...

processors:
  batch:
    timeout: 30s
  
  reduceresolution:

...

service:
  pipelines:
    metrics:
      receivers: [otlp]
      processors: [batch, reduceresolution]
      exporters: [debug]
...

Gauge

The end result of an aggregated gauge is the following metrics:

• _gauge_abs_max (the maximum absolute value found within the sample)
• _gauge_abs_min (the minimum absolute value found within the sample)

When a gauge is received, it is converted into 2 different gauges, represented above. Therefore, if a gauge called speed is emitted from a client, it will result in 5 gauges called speed_gauge_abs_max, and speed_gauge_abs_min.

If other types of aggregates are more suitable, like the real maximum, minimum or even other types, like average, counter, etc., it is possible to add the name of the metric in the collector's options:

...
processors:
  batch:
    timeout: 30s
  
  reduceresolution:
    gauge-aggregations:
      metricA:
        - max
        - min
        - avg
      metricb:
        - avg
        - count
        - abs_max
...

This will change the name from _gauge_abs_max, and _gauge_abs_min to _gauge_max, _gauge_min, and _gauge_avg for metricA, and _gauge_avg, _gauge_count, and _gauge_abs_max for metricb.

The possible values are:

• avg
• count
• sum
• max
• min
• abs_max
• abs_min

Counter and UpDownCounter

Both the Counter and the UpDownCounter are just summed together and emitted with a single value. The name of the counter or the UpDownCounter are not changed.

Linear Histogram

Just like the Counter, the Linear Histogram is summed together and emitted a single set of buckets. The name of the Histogram is not changed.

How to work locally

Install the go version 1.20.12 locally. Obtain the Open Telemetry Collector builder with Dockerfile.

How to build a collector

To build a collector run the following command:

./ocb --config manifest.yaml

How to run the built collector

To run the built collector, run the following command:

./_build/opentelemetry-collector --config config.yaml

How to run the unit tests

Change to the reduceresolution folder and run go test. For more details on the test run, execute go test -v.