metric

Available in: Cloud, Self-Managed

Emit custom metrics by extracting values from messages.

# Config fields, showing default values
label: ""
metric:
  type: "" # No default (required)
  name: "" # No default (required)
  labels: {} # No default (optional)
  value: ""

This processor works by evaluating an interpolated field value for each message and updating a emitted metric according to the type.

Custom metrics such as these are emitted along with Redpanda Connect internal metrics, where you can customize where metrics are sent, which metric names are emitted and rename them as/when appropriate. For more information see the metrics docs.

Fields

type

The metric type to create.

Type: string

Options: counter , counter_by , gauge , timing .

name

The name of the metric to create, this must be unique across all Redpanda Connect components otherwise it will overwrite those other metrics.

Type: string

labels

A map of label names and values that can be used to enrich metrics. Labels are not supported by some metric destinations, in which case the metrics series are combined. This field supports interpolation functions.

Type: object

# Examples

labels:
  topic: ${! meta("kafka_topic") }
  type: ${! json("doc.type") }

value

For some metric types specifies a value to set, increment. Certain metrics exporters such as Prometheus support floating point values, but those that do not will cast a floating point value into an integer. This field supports interpolation functions.

Type: string

Default: ""

Examples

  • Counter

  • Gauge

In this example we emit a counter metric called Foos, which increments for every message processed, and we label the metric with some metadata about where the message came from and a field from the document that states what type it is. We also configure our metrics to emit to CloudWatch, and explicitly only allow our custom metric and some internal Redpanda Connect metrics to emit.

pipeline:
  processors:
    - metric:
        name: Foos
        type: counter
        labels:
          topic: ${! meta("kafka_topic") }
          partition: ${! meta("kafka_partition") }
          type: ${! json("document.type").or("unknown") }

metrics:
  mapping: |
    root = if ![
      "Foos",
      "input_received",
      "output_sent"
    ].contains(this) { deleted() }
  aws_cloudwatch:
    namespace: ProdConsumer

In this example we emit a gauge metric called FooSize, which is given a value extracted from JSON messages at the path foo.size. We then also configure our Prometheus metric exporter to only emit this custom metric and nothing else. We also label the metric with some metadata.

pipeline:
  processors:
    - metric:
        name: FooSize
        type: gauge
        labels:
          topic: ${! meta("kafka_topic") }
        value: ${! json("foo.size") }

metrics:
  mapping: 'if this != "FooSize" { deleted() }'
  prometheus: {}

Types

counter

Increments a counter by exactly 1, the contents of value are ignored by this type.

counter_by

If the contents of value can be parsed as a positive integer value then the counter is incremented by this value.

For example, the following configuration will increment the value of the count.custom.field metric by the contents of field.some.value:

pipeline:
  processors:
    - metric:
        type: counter_by
        name: CountCustomField
        value: ${!json("field.some.value")}

gauge

If the contents of value can be parsed as a positive integer value then the gauge is set to this value.

For example, the following configuration will set the value of the gauge.custom.field metric to the contents of field.some.value:

pipeline:
  processors:
    - metric:
        type: gauge
        name: GaugeCustomField
        value: ${!json("field.some.value")}

timing

Equivalent to gauge where instead the metric is a timing. It is recommended that timing values are recorded in nanoseconds in order to be consistent with standard Redpanda Connect timing metrics, as in some cases these values are automatically converted into other units such as when exporting timings as histograms with Prometheus metrics.