[processor/transform] unit is copied to count metric when using `extract_count_metric()`

[pichlermarc]

Component(s)

processor/transform

What happened?

Description

Using extract_count_metric() will retain the unit of the original metric. While this makes sense for extracted sum metrics, it may not be accurate in most (all?) cases of extracted count metrics.

Example (http.server.request.duration Histogram):

Its unit is s (seconds). In an app, we record how many seconds a request takes. The sum will be in seconds, as we just add those seconds together, so we should keep the unit there. However, for count the unit cannot be seconds as it's actually the number of how many requests were recorded. I think (if I'm not missing anything) that the unit for extracted count should always be 1.

Steps to Reproduce

Prerequisite files

collector-config.yaml

• See config below

request.json

{
  "resourceMetrics": [
    {
      "resource": {
        "attributes": [
          {
            "key": "service.name",
            "value": {
              "stringValue": "basic-metric-service"
            }
          }
        ],
        "droppedAttributesCount": 0
      },
      "scopeMetrics": [
        {
          "metrics": [
            {
              "description": "Example of a Histogram",
              "name": "example.histogram",
              "histogram": {
                "aggregationTemporality": 2,
                "dataPoints": [
                  {
                    "startTimeUnixNano": "1544712660300000000",
                    "timeUnixNano": "1544712660300000000",
                    "count": 2,
                    "sum": 2,
                    "bucketCounts": [1,1],
                    "explicitBounds": [1],
                    "min": 0,
                    "max": 2,
                    "attributes": []
                  }
                ]
              },
              "unit": "s"
            }
          ],
          "scope": {
            "name": "example",
            "version": ""
          }
        }
      ]
    }
  ]
}

Steps:

• create the two files above
• run otelcol-contrib --config collector-config.yaml
• run curl -X POST -H "Content-Type: application/json" -d @request.json -i localhost:4318/v1/metrics
• inspect output (see below, it includes same unit as the source metric)

Expected Result

• Unit is not set to s on the extracted metric

Actual Result

• Unit is set to s the extracted metric

Collector version

v0.95.0

Environment information

Environment

OS: Ubuntu 22.04

OpenTelemetry Collector configuration

receivers:
  otlp:
    protocols:
      http:
        endpoint: 0.0.0.0:4318

processors:
  transform/histogram:
    metric_statements:
      - context: metric
        statements:
          # Get count from the histogram. The new metric name will be <histogram_name>_count
          - extract_count_metric(true) where type == METRIC_DATA_TYPE_HISTOGRAM

exporters:
  debug:
    verbosity: detailed

service:
  pipelines:
    metrics/otlp:
      receivers: [otlp]
      processors: [transform/histogram]
      exporters: [debug]

Log output

Metric #1
Descriptor:
     -> Name: example.histogram_count
     -> Description: Example of a Histogram
     -> Unit: s
     -> DataType: Sum
     -> IsMonotonic: true
     -> AggregationTemporality: Cumulative
NumberDataPoints #0
StartTimestamp: 2018-12-13 14:51:00.3 +0000 UTC
Timestamp: 2018-12-13 14:51:00.3 +0000 UTC
Value: 2

Additional context

• Unit is copied from the source metric at
  

  opentelemetry-collector-contrib/processor/transformprocessor/internal/metrics/func_extract_count_metric.go

  Line 47 in 0bd3c27

  countMetric.SetUnit(metric.Unit())

• Unit is not explicitly mentioned in the README.md (perhaps it was not intended to be copied?)
  

  opentelemetry-collector-contrib/processor/transformprocessor/README.md

  Line 236 in 0bd3c27

  The name for the new metric will be `<original metric name>_count`. The fields that are copied are: `timestamp`, `starttimestamp`, `attibutes`, `description`, and `aggregation_temporality`. As metrics of type Summary don't have an `aggregation_temporality` field, this field will be set to `AGGREGATION_TEMPORALITY_CUMULATIVE` for those metrics.

[github-actions]

Pinging code owners:

• processor/transform: @TylerHelmuth @kentquirk @bogdandrutu @evan-bradley

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

[pichlermarc]

I'd be happy to open a PR if code-owners agree that we should change it. 🙂

[evan-bradley]

Thanks for the detailed report! This is a good catch, I agree that a count metric should have a counter unit and not the unit of the original histogram.