CloudWatch Metrics (OTel)

Selection criteria determine which OTel metrics enter a pipeline for processing. Each criterion is an expression of the form <path> == "<value>" that matches against a specific attribute of the incoming metric. At least one selection criterion is required.

Criteria are grouped inside a match_all block with AND semantics — a metric must match every expression in the group to enter the pipeline.

Supported paths

The following OTTL paths are supported in selection criteria:

Selection criteria paths

Path	Description
resource.attributes["key"]	Resource-level attribute
instrumentation_scope.name	Instrumentation scope name
instrumentation_scope.version	Instrumentation scope version
instrumentation_scope.attributes["key"]	Instrumentation scope attribute
metric.name	Metric name
datapoint.attributes["key"]	Datapoint-level attribute
attributes["key"]	Short form for datapoint.attributes["key"]

Configuration

Selection criteria are defined in the source section of the pipeline configuration:

pipeline:
  source:
    cloudwatch_metrics:
      format: otlp
      selection_criteria:
        - match_all:
            - 'resource.attributes["service.name"] == "my-service"'
            - 'metric.name == "http.server.request.duration"'
  processor:
    - add_attributes:
        attributes:
          - key: resource.attributes["team"]
            value: "platform-engineering"
  sink:
    - cloudwatch_metrics: {}

The following example uses all supported path types:

pipeline:
  source:
    cloudwatch_metrics:
      format: otlp
      selection_criteria:
        - match_all:
            - 'resource.attributes["service.name"] == "my-service"'
            - 'instrumentation_scope.name == "my-scope"'
            - 'instrumentation_scope.version == "1.0.0"'
            - 'instrumentation_scope.attributes["library"] == "otel-java"'
            - 'metric.name == "http.server.request.duration"'
            - 'datapoint.attributes["status_code"] == "200"'
            - 'attributes["environment"] == "production"'
  processor:
    - add_attributes:
        attributes:
          - key: resource.attributes["team"]
            value: "observability"
  sink:
    - cloudwatch_metrics: {}

Verifying with PromQL

CloudWatch maps OTLP attribute scopes to PromQL labels using the @ prefix convention. Use this mapping to verify that your pipeline is processing metrics as expected in Query Studio:

OTTL path to PromQL label mapping

Pipeline OTTL path	PromQL label prefix	Example
resource.attributes["key"]	@resource.	@resource.service.name
instrumentation_scope.name	@instrumentation.@name	@instrumentation.@name
instrumentation_scope.attributes["key"]	@instrumentation.	@instrumentation.library
datapoint.attributes["key"] / attributes["key"]	@datapoint. or bare	status_code

For example, if your pipeline adds resource.attributes["team"] with value "platform-engineering", you can confirm it was applied:

{"CPUUtilization", "@resource.team"="platform-engineering"}

Requirements and limitations

Single match_all group

Each pipeline supports exactly one match_all group in selection_criteria. You cannot define multiple match_all groups in a single pipeline.

Minimum criteria

At least one expression is required in the match_all group.

Maximum criteria

A match_all group can contain at most 20 expressions.

AND semantics

All expressions in the match_all group must match for a metric to enter the pipeline.

Exact string matching

Values must be static strings. Wildcards, regular expressions, and partial matches are not supported. Each expression must use the == operator.

No overlapping criteria across pipelines

Each metric datapoint must match at most one pipeline. If a datapoint matches both new and existing pipeline criteria, pipeline creation fails. To prevent overlap, include at least one attribute path with distinct values in each pipeline's selection criteria.

For example, the following two selection criteria overlap because Pipeline A selects all metrics from payment-service, which includes the specific metric that Pipeline B targets:

# Pipeline A — selects ALL metrics from payment-service
selection_criteria:
  - match_all:
      - 'resource.attributes["service.name"] == "payment-service"'

# Pipeline B — FAILS: a datapoint with service.name="payment-service"
# and metric.name="http.server.request.duration" matches both pipelines
selection_criteria:
  - match_all:
      - 'metric.name == "http.server.request.duration"'