[receiver/statsd]Add support for DataDog `distribution` metrics type

[anuprout]

Component(s)

receiver/statsd

Is your feature request related to a problem? Please describe.

The DataDog supported statsd metrics have an additional metric type distribution which is denoted by d in the statsd protocol. The Otel statsd receivers currently recognizes metric type GAUGE(g), TIMER(ms),COUNTER(c) & HISTOGRAM(h). This causes the distribution metrics sent from a DataDog statsd client to be dropped by the receiver.

Below are some examples of the distribution metrics sent from a DataDog statsd client.
test_distribution_metric:1|d|#tag1-key:tag1-value,tag2-key:tag2-value
test_distribution_metric:2|d|#tag1-key:tag1-value,tag2-key:tag2-value
test_distribution_metric:3|d|#tag1-key:tag1-value,tag2-key:tag2-value

Describe the solution you'd like

The distribution metrics can be parsed by the receiver as histograms . The DataDog exporter converts the Otel histograms to DataDog distributions by default.

When the metric payload is parsed, the metrics with type d can be treated as histograms which would wrap all the functionalities provided for histogram metrics.

Describe alternatives you've considered

The distribution metrics can be parsed by the receiver similar to the histograms where all the configs exposed for a histogram metric can be exposed for distribution metrics as well with default observer_type as histogram.

Sample receiver config.

receivers:
  statsd:
  statsd/2:
    endpoint: "localhost:8127"
    aggregation_interval: 70s
    enable_metric_type: true
    is_monotonic_counter: false
    timer_histogram_mapping:
      - statsd_type: "distribution"
        observer_type: "histogram"
        histogram: 
          max_size: 100 

Additional context

No response

[github-actions]

Pinging code owners:

• receiver/statsd: @jmacd @dmitryax

See Adding Labels via Comments if you do not have permissions to add labels yourself.

[mx-psi]

My recommendation (speaking as a Datadog exporter codeowner here) would be to map these to an OTLP exponential histogram (or at least have the option to do so).

[anuprout]

@mx-psi the statsd receiver, translates the histograms to OTLP exponential histogram with observer_type: "histogram". The same can be used for distribution type metrics as well. I would wait for the comments from the receiver code owners on this.

[pedro-stanaka]

I would like to see this implemented as well and can try taking a stab at it, if maintainers think it is okay to move forward with the solution proposed by @anuprout.

cc @jmacd