prometheus

type ClientFilterOption ¶

type ClientFilterOption func(opt *clientFilterOption)

ClientFilterOption defines the configuration for the various elements of BatchProcessor

func WithClientFilterTraceConfig(enableDeferredSample, deferredSampleError bool,
	deferredSampleSlowDuration time.Duration) ClientFilterOption

type ServerFilterOption ¶

type ServerFilterOption func(opt *serverFilterOption)

ServerFilterOption defines the configuration for the various elements of BatchProcessor

func WithServerFilterTraceConfig(enableDeferredSample, deferredSampleError bool,
	deferredSampleSlowDuration time.Duration) ServerFilterOption

type Sink ¶

type Sink struct {
	// contains filtered or unexported fields
}

Sink implements metrics.Sink interface, convert trpc metrics API to Prometheus format. convert rules is different from trpc-metrics-prometheus plugin: trpc metric name => save original name in `_name` label, name use converted pinyin name trpc metric type => prometheus `_type` label 2021-09-10 update: as too many trpc metrics name is too long, use pinyin name to replace original name will make prometheus metadata api useless. Change into use same metric name for each type, use label to distinguish. 2021-11-29 update: for multi-dimension metrics, generate a collector for each metric, each collector has its own dimension information to prevent dimension explosion. Add counters and gauges fields to store multi-dimension counter and gauge 2023-03-28 update: the style of trpc metrics api multi-dimension metrics is multi-value, Prometheus is single-value. when mapping to Prometheus, use record.metric[*].name as metric name, not unique. Therefor, use record.name + record.metric[*].name as metric name, but it will break the panel. report case 1: metrics.ReportMultiDimensionMetricsX(

    record_name1,  // record.name
    []*metrics.Dimension{ {Name: "dim_1", Value: "val_11"} }, // labels values
    []*metrics.Metrics{
				metrics.NewMetrics("metric_name1", 1, metrics.PolicySUM), // record.metric[*].name
		},
	)

report case 2: metrics.ReportMultiDimensionMetricsX(

    record_name2,  // record.name
    []*metrics.Dimension{ {Name: "dim_1", Value: "val_11"}, {Name: "dim_2", Value: "val_11"} }, // labels values
    []*metrics.Metrics{
				metrics.NewMetrics("metric_name1", 1, metrics.PolicySUM), // record.metric[*].name
		},
	)

The correct way is to use record.name + record.metric[*].name as the Prometheus metric name, but it will change the metric name and break the panel. So use a compatible way to map without modifying the panel statement: 1. Add a label _name=record.name + record.metric[*].name. report case 1 mapping to metric_name1{_name="record_name1_metric_name1", dim_1="val_11"} report case 2 mapping to metric_name1{_name="record_name2_metric_name1", dim_1="val_11", dim_2="val_11"} 2. Implement a special composeMetricVec type, wrap the collectorVec with these two labels inconsistent. When Prometheus stores metrics, label value is empty and there is no label are equivalent.

func NewSink(registerer prometheus.Registerer) *Sink

func (*Sink) Name() string

func (s *Sink) Report(rec metrics.Record, opts ...metrics.Option) error