lib/promrelabel: add SanitizeName() function for sanitizing Prometheus metric names and label names

Optimize this function by using results cache for input strings.
Use this function all over the code.

This is a follow-up for fcffdba9dc

Updates https://github.com/VictoriaMetrics/VictoriaMetrics/issues/3113

app/vmagent/remotewrite/relabel.go

@@ -3,7 +3,6 @@ package remotewrite
 import (
 	"flag"
 	"fmt"
-	"regexp"
 	"strings"
 	"sync"
 
@@ -118,9 +117,9 @@ func (rctx *relabelCtx) applyRelabeling(tss []prompbmarshal.TimeSeries, extraLab
 			for j := range tmpLabels {
 				label := &tmpLabels[j]
 				if label.Name == "__name__" {
-					label.Value = unsupportedPromChars.ReplaceAllString(label.Value, "_")
+					label.Value = promrelabel.SanitizeName(label.Value)
 				} else {
-					label.Name = unsupportedPromChars.ReplaceAllString(label.Name, "_")
+					label.Name = promrelabel.SanitizeName(label.Name)
 				}
 			}
 		}
@@ -138,9 +137,6 @@ func (rctx *relabelCtx) applyRelabeling(tss []prompbmarshal.TimeSeries, extraLab
 	return tssDst
 }
 
-// See https://prometheus.io/docs/concepts/data_model/#metric-names-and-labels
-var unsupportedPromChars = regexp.MustCompile(`[^a-zA-Z0-9_:]`)
-
 type relabelCtx struct {
 	// pool for labels, which are used during the relabeling.
 	labels []prompbmarshal.Label

app/vmctl/opentsdb/parser.go

@@ -6,13 +6,12 @@ import (
 	"strconv"
 	"strings"
 	"time"
+
+	"github.com/VictoriaMetrics/VictoriaMetrics/lib/promrelabel"
 )
 
 var (
-	allowedNames     = regexp.MustCompile("^[a-zA-Z][a-zA-Z0-9_:]*$")
 	allowedFirstChar = regexp.MustCompile("^[a-zA-Z]")
-	replaceChars     = regexp.MustCompile("[^a-zA-Z0-9_:]")
-	allowedTagKeys   = regexp.MustCompile("^[a-zA-Z][a-zA-Z0-9_]*$")
 )
 
 func convertDuration(duration string) (time.Duration, error) {
@@ -180,13 +179,8 @@ func modifyData(msg Metric, normalize bool) (Metric, error) {
 	}
 	/*
 		replace bad characters in metric name with _ per the data model
-		only replace if needed to reduce string processing time
 	*/
-	if !allowedNames.MatchString(name) {
-		finalMsg.Metric = replaceChars.ReplaceAllString(name, "_")
-	} else {
-		finalMsg.Metric = name
-	}
+	finalMsg.Metric = promrelabel.SanitizeName(name)
 	// replace bad characters in tag keys with _ per the data model
 	for key, value := range msg.Tags {
 		// if normalization requested, lowercase the key and value
@@ -196,11 +190,8 @@ func modifyData(msg Metric, normalize bool) (Metric, error) {
 		}
 		/*
 			replace all explicitly bad characters with _
-			only replace if needed to reduce string processing time
 		*/
-		if !allowedTagKeys.MatchString(key) {
-			key = replaceChars.ReplaceAllString(key, "_")
-		}
+		key = promrelabel.SanitizeName(key)
 		// tags that start with __ are considered custom stats for internal prometheus stuff, we should drop them
 		if !strings.HasPrefix(key, "__") {
 			finalMsg.Tags[key] = value

app/vminsert/relabel/relabel.go

@@ -3,7 +3,6 @@ package relabel
 import (
 	"flag"
 	"fmt"
-	"regexp"
 	"sync/atomic"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
@@ -115,9 +114,9 @@ func (ctx *Ctx) ApplyRelabeling(labels []prompb.Label) []prompb.Label {
 		for i := range tmpLabels {
 			label := &tmpLabels[i]
 			if label.Name == "__name__" {
-				label.Value = unsupportedPromChars.ReplaceAllString(label.Value, "_")
+				label.Value = promrelabel.SanitizeName(label.Value)
 			} else {
-				label.Name = unsupportedPromChars.ReplaceAllString(label.Name, "_")
+				label.Name = promrelabel.SanitizeName(label.Name)
 			}
 		}
 	}
@@ -149,6 +148,3 @@ func (ctx *Ctx) ApplyRelabeling(labels []prompb.Label) []prompb.Label {
 }
 
 var metricsDropped = metrics.NewCounter(`vm_relabel_metrics_dropped_total`)
-
-// See https://prometheus.io/docs/concepts/data_model/#metric-names-and-labels
-var unsupportedPromChars = regexp.MustCompile(`[^a-zA-Z0-9_:]`)

lib/promrelabel/relabel.go

@@ -5,6 +5,8 @@ import (
 	"regexp"
 	"strconv"
 	"strings"
+	"sync"
+	"sync/atomic"
 
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/bytesutil"
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/logger"
@@ -558,3 +560,45 @@ func fillLabelReferences(dst []byte, replacement string, labels []prompbmarshal.
 	}
 	return dst
 }
+
+// SanitizeName replaces unsupported by Prometheus chars in metric names and label names with _.
+//
+// See https://prometheus.io/docs/concepts/data_model/#metric-names-and-labels
+func SanitizeName(name string) string {
+	m := sanitizedNames.Load().(*sync.Map)
+	v, ok := m.Load(name)
+	if ok {
+		// Fast path - the sanitized name is found in the cache.
+		sp := v.(*string)
+		return *sp
+	}
+	// Slow path - sanitize name and store it in the cache.
+	sanitizedName := unsupportedPromChars.ReplaceAllString(name, "_")
+	// Make a copy of name in order to limit memory usage to the name length,
+	// since the name may point to bigger string.
+	s := string(append([]byte{}, name...))
+	if sanitizedName == name {
+		// point sanitizedName to just allocated s, since it may point to name,
+		// which, in turn, can point to bigger string.
+		sanitizedName = s
+	}
+	sp := &sanitizedName
+	m.Store(s, sp)
+	n := atomic.AddUint64(&sanitizedNamesLen, 1)
+	if n > 100e3 {
+		atomic.StoreUint64(&sanitizedNamesLen, 0)
+		sanitizedNames.Store(&sync.Map{})
+	}
+	return sanitizedName
+}
+
+var (
+	sanitizedNames    atomic.Value
+	sanitizedNamesLen uint64
+
+	unsupportedPromChars = regexp.MustCompile(`[^a-zA-Z0-9_:]`)
+)
+
+func init() {
+	sanitizedNames.Store(&sync.Map{})
+}

lib/promrelabel/relabel_test.go

@@ -7,6 +7,22 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
 )
 
+func TestSanitizeName(t *testing.T) {
+	f := func(s, resultExpected string) {
+		t.Helper()
+		for i := 0; i < 5; i++ {
+			result := SanitizeName(s)
+			if result != resultExpected {
+				t.Fatalf("unexpected result for SanitizeName(%q) at iteration %d; got %q; want %q", s, i, result, resultExpected)
+			}
+		}
+	}
+	f("", "")
+	f("a", "a")
+	f("foo.bar/baz:a", "foo_bar_baz:a")
+	f("foo...bar", "foo___bar")
+}
+
 func TestLabelsToString(t *testing.T) {
 	f := func(labels []prompbmarshal.Label, sExpected string) {
 		t.Helper()

lib/promrelabel/relabel_timing_test.go

@@ -8,6 +8,27 @@ import (
 	"github.com/VictoriaMetrics/VictoriaMetrics/lib/prompbmarshal"
 )
 
+func BenchmarkSanitizeName(b *testing.B) {
+	for _, name := range []string{"", "foo", "foo-bar-baz", "http_requests_total"} {
+		b.Run(name, func(b *testing.B) {
+			benchmarkSanitizeName(b, name)
+		})
+	}
+}
+
+func benchmarkSanitizeName(b *testing.B, name string) {
+	b.ReportAllocs()
+	b.SetBytes(1)
+	b.RunParallel(func(pb *testing.PB) {
+		for pb.Next() {
+			sanitizedName := SanitizeName(name)
+			GlobalSink += len(sanitizedName)
+		}
+	})
+}
+
+var GlobalSink int
+
 func BenchmarkMatchRegexPrefixDotPlusMatchOptimized(b *testing.B) {
 	const pattern = "^foo.+$"
 	const s = "foobar"