processor_metric.go

package pure

import (
	"errors"
	"fmt"
	"sort"
	"strconv"
	"strings"
	"time"

	"github.com/benthosdev/benthos/v4/internal/bloblang/field"
	"github.com/benthosdev/benthos/v4/internal/bundle"
	"github.com/benthosdev/benthos/v4/internal/component/metrics"
	"github.com/benthosdev/benthos/v4/internal/component/processor"
	"github.com/benthosdev/benthos/v4/internal/docs"
	"github.com/benthosdev/benthos/v4/internal/log"
	"github.com/benthosdev/benthos/v4/internal/message"
)

func init() {
	err := bundle.AllProcessors.Add(func(conf processor.Config, mgr bundle.NewManagement) (processor.V1, error) {
		return newMetricProcessor(conf, mgr, mgr.Logger(), mgr.Metrics())
	}, docs.ComponentSpec{
		Name: "metric",
		Categories: []string{
			"Utility",
		},
		Summary: "Emit custom metrics by extracting values from messages.",
		Description: `
This processor works by evaluating an [interpolated field ` + "`value`" + `](/docs/configuration/interpolation#bloblang-queries) for each message and updating a emitted metric according to the [type](#types).

Custom metrics such as these are emitted along with Benthos internal metrics, where you can customize where metrics are sent, which metric names are emitted and rename them as/when appropriate. For more information check out the [metrics docs here](/docs/components/metrics/about).`,
		Config: docs.FieldComponent().WithChildren(
			docs.FieldString("type", "The metric [type](#types) to create.").HasOptions(
				"counter",
				"counter_by",
				"gauge",
				"timing",
			),
			docs.FieldString("name", "The name of the metric to create, this must be unique across all Benthos components otherwise it will overwrite those other metrics."),
			docs.FieldString(
				"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.",
				map[string]string{
					"type":  "${! json(\"doc.type\") }",
					"topic": "${! meta(\"kafka_topic\") }",
				},
			).IsInterpolated().Map(),
			docs.FieldString("value", "For some metric types specifies a value to set, increment.").IsInterpolated(),
		).ChildDefaultAndTypesFromStruct(processor.NewMetricConfig()),
		Examples: []docs.AnnotatedExample{
			{
				Title:   "Counter",
				Summary: "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 Benthos metrics to emit.",
				Config: `
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
`,
			},
			{
				Title:   "Gauge",
				Summary: "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.",
				Config: `
pipeline:
  processors:
    - metric:
        name: FooSize
        type: gauge
        labels:
          topic: ${! meta("kafka_topic") }
        value: ${! json("foo.size") }

metrics:
  mapping: 'if this != "FooSize" { deleted() }'
  prometheus: {}
`,
			},
		},
		Footnotes: `
## 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`" + `:

` + "```yaml" + `
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`" + `:

` + "```yaml" + `
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 Benthos timing metrics, as in some cases these values are automatically converted into other units such as when exporting timings as histograms with Prometheus metrics.`,
	})
	if err != nil {
		panic(err)
	}
}

type metricProcessor struct {
	conf  processor.Config
	log   log.Modular
	stats metrics.Type

	value  *field.Expression
	labels labels

	mCounter metrics.StatCounter
	mGauge   metrics.StatGauge
	mTimer   metrics.StatTimer

	mCounterVec metrics.StatCounterVec
	mGaugeVec   metrics.StatGaugeVec
	mTimerVec   metrics.StatTimerVec

	handler func(string, int, *message.Batch) error
}

type labels []label
type label struct {
	name  string
	value *field.Expression
}

func (l *label) val(index int, msg *message.Batch) string {
	return l.value.String(index, msg)
}

func (l labels) names() []string {
	var names []string
	for i := range l {
		names = append(names, l[i].name)
	}
	return names
}

func (l labels) values(index int, msg *message.Batch) []string {
	var values []string
	for i := range l {
		values = append(values, l[i].val(index, msg))
	}
	return values
}

func newMetricProcessor(conf processor.Config, mgr bundle.NewManagement, log log.Modular, stats metrics.Type) (processor.V1, error) {
	value, err := mgr.BloblEnvironment().NewField(conf.Metric.Value)
	if err != nil {
		return nil, fmt.Errorf("failed to parse value expression: %v", err)
	}

	m := &metricProcessor{
		conf:  conf,
		log:   log,
		stats: stats,
		value: value,
	}

	name := conf.Metric.Name
	if name == "" {
		return nil, errors.New("metric name must not be empty")
	}

	labelNames := make([]string, 0, len(conf.Metric.Labels))
	for n := range conf.Metric.Labels {
		labelNames = append(labelNames, n)
	}
	sort.Strings(labelNames)

	for _, n := range labelNames {
		v, err := mgr.BloblEnvironment().NewField(conf.Metric.Labels[n])
		if err != nil {
			return nil, fmt.Errorf("failed to parse label '%v' expression: %v", n, err)
		}
		m.labels = append(m.labels, label{
			name:  n,
			value: v,
		})
	}

	switch strings.ToLower(conf.Metric.Type) {
	case "counter":
		if len(m.labels) > 0 {
			m.mCounterVec = stats.GetCounterVec(name, m.labels.names()...)
		} else {
			m.mCounter = stats.GetCounter(name)
		}
		m.handler = m.handleCounter
	case "counter_by":
		if len(m.labels) > 0 {
			m.mCounterVec = stats.GetCounterVec(name, m.labels.names()...)
		} else {
			m.mCounter = stats.GetCounter(name)
		}
		m.handler = m.handleCounterBy
	case "gauge":
		if len(m.labels) > 0 {
			m.mGaugeVec = stats.GetGaugeVec(name, m.labels.names()...)
		} else {
			m.mGauge = stats.GetGauge(name)
		}
		m.handler = m.handleGauge
	case "timing":
		if len(m.labels) > 0 {
			m.mTimerVec = stats.GetTimerVec(name, m.labels.names()...)
		} else {
			m.mTimer = stats.GetTimer(name)
		}
		m.handler = m.handleTimer
	default:
		return nil, fmt.Errorf("metric type unrecognised: %v", conf.Metric.Type)
	}

	return m, nil
}

func (m *metricProcessor) handleCounter(val string, index int, msg *message.Batch) error {
	if len(m.labels) > 0 {
		m.mCounterVec.With(m.labels.values(index, msg)...).Incr(1)
	} else {
		m.mCounter.Incr(1)
	}
	return nil
}

func (m *metricProcessor) handleCounterBy(val string, index int, msg *message.Batch) error {
	i, err := strconv.ParseInt(val, 10, 64)
	if err != nil {
		return err
	}
	if i < 0 {
		return errors.New("value is negative")
	}
	if len(m.labels) > 0 {
		m.mCounterVec.With(m.labels.values(index, msg)...).Incr(i)
	} else {
		m.mCounter.Incr(i)
	}
	return nil
}

func (m *metricProcessor) handleGauge(val string, index int, msg *message.Batch) error {
	i, err := strconv.ParseInt(val, 10, 64)
	if err != nil {
		return err
	}
	if i < 0 {
		return errors.New("value is negative")
	}
	if len(m.labels) > 0 {
		m.mGaugeVec.With(m.labels.values(index, msg)...).Set(i)
	} else {
		m.mGauge.Set(i)
	}
	return nil
}

func (m *metricProcessor) handleTimer(val string, index int, msg *message.Batch) error {
	i, err := strconv.ParseInt(val, 10, 64)
	if err != nil {
		return err
	}
	if i < 0 {
		return errors.New("value is negative")
	}
	if len(m.labels) > 0 {
		m.mTimerVec.With(m.labels.values(index, msg)...).Timing(i)
	} else {
		m.mTimer.Timing(i)
	}
	return nil
}

func (m *metricProcessor) ProcessMessage(msg *message.Batch) ([]*message.Batch, error) {
	_ = msg.Iter(func(i int, p *message.Part) error {
		value := m.value.String(i, msg)
		if err := m.handler(value, i, msg); err != nil {
			m.log.Errorf("Handler error: %v\n", err)
		}
		return nil
	})
	return []*message.Batch{msg}, nil
}

func (m *metricProcessor) CloseAsync() {
}

func (m *metricProcessor) WaitForClose(timeout time.Duration) error {
	return nil
}