Distinguish data labels instead of identifying ones

Signed-off-by: Arve Knudsen <arve.knudsen@gmail.com>

model/labels/labels.go

@@ -412,12 +412,8 @@ func (b *ScratchBuilder) Reset() {
 
 // Add a name/value pair.
 // Note if you Add the same name twice you will get a duplicate label, which is invalid.
-func (b *ScratchBuilder) Add(name, value string, identifyingLabels ...string) {
-	identifying := false
-	if len(identifyingLabels) > 0 && slices.Contains(identifyingLabels, name) {
-		identifying = true
-	}
-	b.add = append(b.add, Label{Name: name, Value: value, Identifying: identifying})
+func (b *ScratchBuilder) Add(name, value string) {
+	b.add = append(b.add, Label{Name: name, Value: value})
 }
 
 // Add a name/value pair, using []byte instead of string.

model/labels/labels_common.go

@@ -16,7 +16,6 @@ package labels
 import (
 	"bytes"
 	"encoding/json"
-	"fmt"
 	"strconv"
 
 	"github.com/prometheus/common/model"
@@ -37,13 +36,6 @@ var seps = []byte{'\xff'}
 // Label is a key/value pair of strings.
 type Label struct {
 	Name, Value string
-	// Identifying signifies whether this is an identifying label for an info metric.
-	Identifying bool
-}
-
-// String returns l's string representation.
-func (l Label) String() string {
-	return fmt.Sprintf("%s=%s", l.Name, l.Value)
 }
 
 func (ls Labels) String() string {

model/labels/labels_stringlabels.go

@@ -620,12 +620,8 @@ func (b *ScratchBuilder) Reset() {
 
 // Add a name/value pair.
 // Note if you Add the same name twice you will get a duplicate label, which is invalid.
-func (b *ScratchBuilder) Add(name, value string, identifyingLabels ...string) {
-	identifying := false
-	if len(identifyingLabels) > 0 && slices.Contains(identifyingLabels, name) {
-		identifying = true
-	}
-	b.add = append(b.add, Label{Name: name, Value: value, Identifying: identifying})
+func (b *ScratchBuilder) Add(name, value string) {
+	b.add = append(b.add, Label{Name: name, Value: value})
 }
 
 // Add a name/value pair, using []byte instead of string to reduce memory allocations.

model/labels/labels_test.go

@@ -582,7 +582,7 @@ func TestBuilder(t *testing.T) {
 		},
 		{
 			base: FromStrings("aaa", "111", "bbb", "222", "ccc", "333"),
-			set:  []Label{{Name: "aaa", Value: "444"}, {Name: "bbb", Value: "555"}, {Name: "ccc", Value: "666"}},
+			set:  []Label{{"aaa", "444"}, {"bbb", "555"}, {"ccc", "666"}},
 			want: FromStrings("aaa", "444", "bbb", "555", "ccc", "666"),
 		},
 		{

prompb/types.proto

@@ -127,7 +127,8 @@ message TimeSeries {
   repeated Sample samples           = 2 [(gogoproto.nullable) = false];
   repeated Exemplar exemplars       = 3 [(gogoproto.nullable) = false];
   repeated Histogram histograms     = 4 [(gogoproto.nullable) = false];
-  repeated string identifyingLabels = 5;
+  // data labels for target_info metrics.
+  repeated string dataLabels        = 5;
 }
 
 message Label {

storage/remote/codec.go

@@ -760,10 +760,10 @@ func LabelProtosToMetric(labelPairs []*prompb.Label) model.Metric {
 	return metric
 }
 
-func labelProtosToLabels(labelPairs []prompb.Label, identifyingLabels ...string) labels.Labels {
+func labelProtosToLabels(labelPairs []prompb.Label) labels.Labels {
 	b := labels.ScratchBuilder{}
 	for _, l := range labelPairs {
-		b.Add(l.Name, l.Value, identifyingLabels...)
+		b.Add(l.Name, l.Value)
 	}
 	b.Sort()
 	return b.Labels()

storage/remote/otlptranslator/prometheusremotewrite/helper.go

@@ -72,8 +72,7 @@ func (a ByLabelName) Swap(i, j int)      { a[i], a[j] = a[j], a[i] }
 // creates a new TimeSeries in the map if not found and returns the time series signature.
 // tsMap will be unmodified if either labels or sample is nil, but can still be modified if the exemplar is nil.
 func addSample(tsMap map[string]*prompb.TimeSeries, sample *prompb.Sample, labels []prompb.Label,
-	datatype string, isTargetMetric bool) string {
-
+	datatype string) string {
 	if sample == nil || labels == nil || tsMap == nil {
 		return ""
 	}
@@ -84,18 +83,9 @@ func addSample(tsMap map[string]*prompb.TimeSeries, sample *prompb.Sample, label
 	if ok {
 		ts.Samples = append(ts.Samples, *sample)
 	} else {
-		var identifyingLabels []string
-		if isTargetMetric {
-			identifyingLabels = []string{
-				model.InstanceLabel,
-				model.JobLabel,
-			}
-			slices.Sort(identifyingLabels)
-		}
 		newTs := &prompb.TimeSeries{
-			Labels:            labels,
-			Samples:           []prompb.Sample{*sample},
-			IdentifyingLabels: identifyingLabels,
+			Labels:  labels,
+			Samples: []prompb.Sample{*sample},
 		}
 		tsMap[sig] = newTs
 	}
@@ -300,7 +290,7 @@ func addSingleHistogramDataPoint(pt pmetric.HistogramDataPoint, resource pcommon
 		}
 
 		sumlabels := createLabels(sumStr)
-		addSample(tsMap, sum, sumlabels, metric.Type().String(), false)
+		addSample(tsMap, sum, sumlabels, metric.Type().String())
 
 	}
 
@@ -314,7 +304,7 @@ func addSingleHistogramDataPoint(pt pmetric.HistogramDataPoint, resource pcommon
 	}
 
 	countlabels := createLabels(countStr)
-	addSample(tsMap, count, countlabels, metric.Type().String(), false)
+	addSample(tsMap, count, countlabels, metric.Type().String())
 
 	// cumulative count for conversion to cumulative histogram
 	var cumulativeCount uint64
@@ -336,7 +326,7 @@ func addSingleHistogramDataPoint(pt pmetric.HistogramDataPoint, resource pcommon
 		}
 		boundStr := strconv.FormatFloat(bound, 'f', -1, 64)
 		labels := createLabels(bucketStr, leStr, boundStr)
-		sig := addSample(tsMap, bucket, labels, metric.Type().String(), false)
+		sig := addSample(tsMap, bucket, labels, metric.Type().String())
 
 		bucketBounds = append(bucketBounds, bucketBoundsData{sig: sig, bound: bound})
 	}
@@ -350,7 +340,7 @@ func addSingleHistogramDataPoint(pt pmetric.HistogramDataPoint, resource pcommon
 		infBucket.Value = float64(pt.Count())
 	}
 	infLabels := createLabels(bucketStr, leStr, pInfStr)
-	sig := addSample(tsMap, infBucket, infLabels, metric.Type().String(), false)
+	sig := addSample(tsMap, infBucket, infLabels, metric.Type().String())
 
 	bucketBounds = append(bucketBounds, bucketBoundsData{sig: sig, bound: math.Inf(1)})
 	addExemplars(tsMap, promExemplars, bucketBounds)
@@ -496,7 +486,7 @@ func addSingleSummaryDataPoint(pt pmetric.SummaryDataPoint, resource pcommon.Res
 		sum.Value = math.Float64frombits(value.StaleNaN)
 	}
 	sumlabels := createLabels(baseName + sumStr)
-	addSample(tsMap, sum, sumlabels, metric.Type().String(), false)
+	addSample(tsMap, sum, sumlabels, metric.Type().String())
 
 	// treat count as a sample in an individual TimeSeries
 	count := &prompb.Sample{
@@ -507,7 +497,7 @@ func addSingleSummaryDataPoint(pt pmetric.SummaryDataPoint, resource pcommon.Res
 		count.Value = math.Float64frombits(value.StaleNaN)
 	}
 	countlabels := createLabels(baseName + countStr)
-	addSample(tsMap, count, countlabels, metric.Type().String(), false)
+	addSample(tsMap, count, countlabels, metric.Type().String())
 
 	// process each percentile/quantile
 	for i := 0; i < pt.QuantileValues().Len(); i++ {
@@ -521,7 +511,7 @@ func addSingleSummaryDataPoint(pt pmetric.SummaryDataPoint, resource pcommon.Res
 		}
 		percentileStr := strconv.FormatFloat(qt.Quantile(), 'f', -1, 64)
 		qtlabels := createLabels(baseName, quantileStr, percentileStr)
-		addSample(tsMap, quantile, qtlabels, metric.Type().String(), false)
+		addSample(tsMap, quantile, qtlabels, metric.Type().String())
 	}
 
 	// add _created time series if needed
@@ -586,7 +576,16 @@ func addResourceTargetInfo(resource pcommon.Resource, settings Settings, timesta
 		// convert ns to ms
 		Timestamp: convertTimeStamp(timestamp),
 	}
-	addSample(tsMap, sample, labels, infoType, true)
+	sig := addSample(tsMap, sample, labels, infoType)
+
+	dataLabels := make([]string, 0, len(labels))
+	for _, l := range labels {
+		if l.Name != model.InstanceLabel && l.Name != model.JobLabel {
+			dataLabels = append(dataLabels, l.Name)
+		}
+	}
+	slices.Sort(dataLabels)
+	tsMap[sig].DataLabels = dataLabels
 }
 
 // convertTimeStamp converts OTLP timestamp in ns to timestamp in ms

storage/remote/otlptranslator/prometheusremotewrite/number_data_points.go

@@ -47,7 +47,7 @@ func addSingleGaugeNumberDataPoint(
 	if pt.Flags().NoRecordedValue() {
 		sample.Value = math.Float64frombits(value.StaleNaN)
 	}
-	addSample(series, sample, labels, metric.Type().String(), false)
+	addSample(series, sample, labels, metric.Type().String())
 }
 
 // addSingleSumNumberDataPoint converts the Sum metric data point to a Prometheus
@@ -80,7 +80,7 @@ func addSingleSumNumberDataPoint(
 	if pt.Flags().NoRecordedValue() {
 		sample.Value = math.Float64frombits(value.StaleNaN)
 	}
-	sig := addSample(series, sample, labels, metric.Type().String(), false)
+	sig := addSample(series, sample, labels, metric.Type().String())
 
 	if ts, ok := series[sig]; sig != "" && ok {
 		exemplars := getPromExemplars[pmetric.NumberDataPoint](pt)

storage/remote/write_handler.go

@@ -114,16 +114,14 @@ func (h *writeHandler) write(ctx context.Context, req *prompb.WriteRequest) (err
 
 	var exemplarErr error
 	for _, ts := range req.Timeseries {
-		labels := labelProtosToLabels(ts.Labels, ts.IdentifyingLabels...)
+		labels := labelProtosToLabels(ts.Labels)
 		if !labels.IsValid() {
 			level.Warn(h.logger).Log("msg", "Invalid metric names or labels", "got", labels.String())
 			samplesWithInvalidLabels++
 			continue
 		}
 		var ref storage.SeriesRef
 		for _, s := range ts.Samples {
-			// When recording a sample for an info type metric with identifying labels,
-			// make sure also to persist which of the labels are identifying
 			ref, err = app.Append(ref, labels, s.Timestamp, s.Value)
 			if err != nil {
 				unwrappedErr := errors.Unwrap(err)
@@ -135,7 +133,6 @@ func (h *writeHandler) write(ctx context.Context, req *prompb.WriteRequest) (err
 				}
 				return err
 			}
-
 		}
 
 		for _, ep := range ts.Exemplars {
@@ -170,6 +167,13 @@ func (h *writeHandler) write(ctx context.Context, req *prompb.WriteRequest) (err
 				return err
 			}
 		}
+
+		if len(ts.DataLabels) > 0 {
+			// Info type metric with metadata represented as data labels
+			if err := app.AppendDataLabels(ts.DataLabels, ts.Samples[0].Timestamp); err != nil {
+				return err
+			}
+		}
 	}
 
 	if outOfOrderExemplarErrs > 0 {