Merge pull request #87 from prometheus/beorn7/fingerprint

Unify fingerprinting aka signature calculation.

Author: beorn7

model/metric.go

@@ -14,10 +14,8 @@
 package model
 
 import (
-	"encoding/binary"
 	"encoding/json"
 	"fmt"
-	"hash/fnv"
 	"sort"
 	"strings"
 )
@@ -66,37 +64,7 @@ func (m Metric) String() string {
 
 // Fingerprint returns a Metric's Fingerprint.
 func (m Metric) Fingerprint() Fingerprint {
-	labelNames := make([]string, 0, len(m))
-	maxLength := 0
-
-	for labelName, labelValue := range m {
-		labelNames = append(labelNames, string(labelName))
-		if len(labelName) > maxLength {
-			maxLength = len(labelName)
-		}
-		if len(labelValue) > maxLength {
-			maxLength = len(labelValue)
-		}
-	}
-
-	sort.Strings(labelNames)
-
-	summer := fnv.New64a()
-	buf := make([]byte, maxLength)
-
-	for _, labelName := range labelNames {
-		labelValue := m[LabelName(labelName)]
-
-		copy(buf, labelName)
-		summer.Write(buf[:len(labelName)])
-
-		summer.Write(separator)
-
-		copy(buf, labelValue)
-		summer.Write(buf[:len(labelValue)])
-	}
-
-	return Fingerprint(binary.LittleEndian.Uint64(summer.Sum(nil)))
+	return metricToFingerprint(m)
 }
 
 // Clone returns a copy of the Metric.

model/metric_test.go

@@ -22,39 +22,35 @@ func testMetric(t testing.TB) {
 	}{
 		{
 			input:       Metric{},
-			fingerprint: 2676020557754725067,
+			fingerprint: 14695981039346656037,
 		},
 		{
 			input: Metric{
 				"first_name":   "electro",
 				"occupation":   "robot",
 				"manufacturer": "westinghouse",
 			},
-			fingerprint: 13260944541294022935,
+			fingerprint: 11310079640881077873,
 		},
 		{
 			input: Metric{
 				"x": "y",
 			},
-			fingerprint: 1470933794305433534,
+			fingerprint: 13948396922932177635,
 		},
-		// The following two demonstrate a bug in fingerprinting. They
-		// should not have the same fingerprint with a sane
-		// fingerprinting function. See
-		// https://github.com/prometheus/client_golang/issues/74 .
 		{
 			input: Metric{
 				"a": "bb",
 				"b": "c",
 			},
-			fingerprint: 3734646176939799877,
+			fingerprint: 3198632812309449502,
 		},
 		{
 			input: Metric{
 				"a":  "b",
 				"bb": "c",
 			},
-			fingerprint: 3734646176939799877,
+			fingerprint: 5774953389407657638,
 		},
 	}
 

model/signature.go

@@ -63,10 +63,10 @@ func LabelsToSignature(labels map[string]string) uint64 {
 	hb := getHashAndBuf()
 	defer putHashAndBuf(hb)
 
-	for k, v := range labels {
-		hb.b.WriteString(k)
+	for labelName, labelValue := range labels {
+		hb.b.WriteString(labelName)
 		hb.b.WriteByte(SeparatorByte)
-		hb.b.WriteString(v)
+		hb.b.WriteString(labelValue)
 		hb.h.Write(hb.b.Bytes())
 		result ^= hb.h.Sum64()
 		hb.h.Reset()
@@ -75,23 +75,79 @@ func LabelsToSignature(labels map[string]string) uint64 {
 	return result
 }
 
-// LabelValuesToSignature returns a unique signature (i.e., fingerprint) for the
-// values of a given label set.
-func LabelValuesToSignature(labels map[string]string) uint64 {
-	if len(labels) == 0 {
+// metricToFingerprint works exactly as LabelsToSignature but takes a Metric as
+// parameter (rather than a label map) and returns a Fingerprint.
+func metricToFingerprint(m Metric) Fingerprint {
+	if len(m) == 0 {
+		return Fingerprint(emptyLabelSignature)
+	}
+
+	var result uint64
+	hb := getHashAndBuf()
+	defer putHashAndBuf(hb)
+
+	for labelName, labelValue := range m {
+		hb.b.WriteString(string(labelName))
+		hb.b.WriteByte(SeparatorByte)
+		hb.b.WriteString(string(labelValue))
+		hb.h.Write(hb.b.Bytes())
+		result ^= hb.h.Sum64()
+		hb.h.Reset()
+		hb.b.Reset()
+	}
+	return Fingerprint(result)
+}
+
+// SignatureForLabels works like LabelsToSignature but takes a Metric as
+// parameter (rather than a label map) and only includes the labels with the
+// specified LabelNames into the signature calculation.
+func SignatureForLabels(m Metric, labels LabelNames) uint64 {
+	if len(m) == 0 || len(labels) == 0 {
+		return emptyLabelSignature
+	}
+
+	var result uint64
+	hb := getHashAndBuf()
+	defer putHashAndBuf(hb)
+
+	for _, label := range labels {
+		hb.b.WriteString(string(label))
+		hb.b.WriteByte(SeparatorByte)
+		hb.b.WriteString(string(m[label]))
+		hb.h.Write(hb.b.Bytes())
+		result ^= hb.h.Sum64()
+		hb.h.Reset()
+		hb.b.Reset()
+	}
+	return result
+}
+
+// SignatureWithoutLabels works like LabelsToSignature but takes a Metric as
+// parameter (rather than a label map) and excludes the labels with any of the
+// specified LabelNames from the signature calculation.
+func SignatureWithoutLabels(m Metric, labels map[LabelName]struct{}) uint64 {
+	if len(m) == 0 {
 		return emptyLabelSignature
 	}
 
 	var result uint64
 	hb := getHashAndBuf()
 	defer putHashAndBuf(hb)
 
-	for _, v := range labels {
-		hb.b.WriteString(v)
+	for labelName, labelValue := range m {
+		if _, exclude := labels[labelName]; exclude {
+			continue
+		}
+		hb.b.WriteString(string(labelName))
+		hb.b.WriteByte(SeparatorByte)
+		hb.b.WriteString(string(labelValue))
 		hb.h.Write(hb.b.Bytes())
 		result ^= hb.h.Sum64()
 		hb.h.Reset()
 		hb.b.Reset()
 	}
+	if result == 0 {
+		return emptyLabelSignature
+	}
 	return result
 }