Add TODOs, drop based on temporality, and sanitize labels

exporters/prometheus/exporter.go

@@ -19,9 +19,14 @@ package prometheus // import "go.opentelemetry.io/otel/exporters/prometheus"
 
 import (
 	"context"
+	"sort"
+	"strings"
+	"unicode"
 
 	"github.com/prometheus/client_golang/prometheus"
 
+	"go.opentelemetry.io/otel/internal/global"
+
 	"go.opentelemetry.io/otel"
 	"go.opentelemetry.io/otel/attribute"
 	"go.opentelemetry.io/otel/sdk/metric"
@@ -67,6 +72,8 @@ func (e *exporter) Collect(ch chan<- prometheus.Metric) {
 		otel.Handle(err)
 	}
 
+	// TODO(damemi): convert otel resource to target_info
+	// see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/metrics/data-model.md#resource-attributes-1
 	for _, metricData := range getMetricData(metrics) {
 		if metricData.valueType == prometheus.UntypedValue {
 			m, err := prometheus.NewConstHistogram(metricData.description, metricData.histogramCount, metricData.histogramSum, metricData.histogramBuckets, metricData.attributeValues...)
@@ -86,6 +93,8 @@ func (e *exporter) Collect(ch chan<- prometheus.Metric) {
 
 // metricData holds the metadata as well as values for individual data points.
 type metricData struct {
+	// name should include the unit as a suffix (before _total on counters)
+	// see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/metrics/data-model.md#metric-metadata-1
 	name             string
 	description      *prometheus.Desc
 	attributeValues  []string
@@ -119,6 +128,13 @@ func getMetricData(metrics metricdata.ResourceMetrics) []*metricData {
 }
 
 func getHistogramMetricData(histogram metricdata.Histogram, m metricdata.Metrics) []*metricData {
+	// Drop histograms with delta aggregation temporality
+	if histogram.Temporality == metricdata.DeltaTemporality {
+		global.Info("dropping histogram with delta temporality", "name", m.Name)
+		return []*metricData{}
+	}
+
+	// TODO(https://github.com/open-telemetry/opentelemetry-go/issues/3163): support exemplars
 	dataPoints := make([]*metricData, 0)
 	for _, dp := range histogram.DataPoints {
 		keys, values := getAttrs(dp.Attributes)
@@ -142,6 +158,13 @@ func getHistogramMetricData(histogram metricdata.Histogram, m metricdata.Metrics
 }
 
 func getSumMetricData[N int64 | float64](sum metricdata.Sum[N], m metricdata.Metrics) []*metricData {
+	// TODO(damemi): convert delta aggregation temporality to cumulative
+	// see https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/metrics/data-model.md#sums-1
+	if sum.Temporality == metricdata.DeltaTemporality {
+		global.Info("dropping sum with delta temporality", "name", m.Name)
+		return []*metricData{}
+	}
+
 	dataPoints := make([]*metricData, 0)
 	for _, dp := range sum.DataPoints {
 		keys, values := getAttrs(dp.Attributes)
@@ -175,12 +198,36 @@ func getGaugeMetricData[N int64 | float64](gauge metricdata.Gauge[N], m metricda
 	return dataPoints
 }
 
+// getAttrs parses the attribute.Set to two lists of matching Prometheus-style
+// keys and values. It sanitizes invalid characters and handles duplicate keys
+// (due to sanitization) by sorting and concatenating the values following the spec.
 func getAttrs(attrs attribute.Set) ([]string, []string) {
+	keysMap := make(map[string][]string)
+	for _, kv := range attrs.ToSlice() {
+		key := strings.Map(sanitizeRune, string(kv.Key))
+		if _, ok := keysMap[key]; !ok {
+			keysMap[key] = []string{kv.Value.AsString()}
+		} else {
+			// if the sanitized key is a duplicate, append to the list of keys
+			keysMap[key] = append(keysMap[key], kv.Value.AsString())
+		}
+	}
+
 	keys := make([]string, 0, attrs.Len())
 	values := make([]string, 0, attrs.Len())
-	for _, kv := range attrs.ToSlice() {
-		keys = append(keys, string(kv.Key))
-		values = append(values, kv.Value.AsString())
+	for key, vals := range keysMap {
+		keys = append(keys, key)
+		sort.Slice(vals, func(i, j int) bool {
+			return i < j
+		})
+		values = append(values, strings.Join(vals, ";"))
 	}
 	return keys, values
 }
+
+func sanitizeRune(r rune) rune {
+	if unicode.IsLetter(r) || unicode.IsDigit(r) || string(r) == ":" || string(r) == "_" {
+		return r
+	}
+	return '_'
+}

exporters/prometheus/exporter_test.go

@@ -84,6 +84,26 @@ func TestPrometheusExporter(t *testing.T) {
 				histogram.Record(ctx, 105, labels...)
 			},
 		},
+		{
+			name:         "sanitized attributes to labels",
+			expectedFile: "testdata/sanitized_labels.txt",
+			recordMetrics: func(ctx context.Context, meter otelmetric.Meter) {
+				labels := []attribute.KeyValue{
+					// exact match, value should be overwritten
+					attribute.Key("A.B").String("X"),
+					attribute.Key("A.B").String("Q"),
+
+					// unintended match due to sanitization, values should be concatenated
+					attribute.Key("C.D").String("Y"),
+					attribute.Key("C/D").String("Z"),
+				}
+				counter, err := meter.SyncFloat64().Counter("foo", instrument.WithDescription("a sanitary counter"))
+				require.NoError(t, err)
+				counter.Add(ctx, 5, labels...)
+				counter.Add(ctx, 10.3, labels...)
+				counter.Add(ctx, 9, labels...)
+			},
+		},
 	}
 
 	for _, tc := range testCases {

exporters/prometheus/testdata/sanitized_labels.txt

@@ -0,0 +1,3 @@
+# HELP foo a sanitary counter
+# TYPE foo counter
+foo{A_B="Q",C_D="Y;Z"} 24.3