Cherry Pick: Adds ability to specify sanitizeOptions for prometheus m…

…etrics (#3170) (#3174)

Adds ability to specify sanitizeOptions for prometheus metrics (#3170)

This PR adds the ability to set specific tally.SanitizeOptions via temporal configuration. If the configuration is not set, we rely on the default that is defined in code. Otherwise, we will generate a tally.SanitizeOptions by parsing the configuration file and converting it to a tally.SanitizeOptions.

Because configuration comes in via yaml, all relevant input fields are strings. The code converts the strings to runes and performs length check validations. Any invalid configuration will fail server startup.

.gitignore

@@ -2,6 +2,7 @@
 
 .idea/
 .vscode/settings.json
+.vscode
 *.code-workspace
 .DS_Store
 .coverage

common/metrics/config.go

@@ -25,6 +25,8 @@
 package metrics
 
 import (
+	"errors"
+	"fmt"
 	"time"
 
 	"github.com/cactus/go-statsd-client/statsd"
@@ -134,6 +136,11 @@ type (
 		// on the specified listen address or registering a metric with the
 		// Prometheus. By default the registerer will panic.
 		OnError string `yaml:"onError"`
+
+		// Deprecated. SanitizeOptions is an optional field that enables a user to
+		// specify which characters are valid and/or should be replaced before metrics
+		// are emitted.
+		SanitizeOptions *SanitizeOptions `yaml:"sanitizeOptions"`
 	}
 )
 
@@ -150,6 +157,23 @@ type SummaryObjective struct {
 	AllowedError float64 `yaml:"allowedError"`
 }
 
+type SanitizeRange struct {
+	StartRange string `yaml:"startRange"`
+	EndRange   string `yaml:"endRange"`
+}
+
+type ValidCharacters struct {
+	Ranges         []SanitizeRange `yaml:"ranges"`
+	SafeCharacters string          `yaml:"safeChars"`
+}
+
+type SanitizeOptions struct {
+	NameCharacters       *ValidCharacters `yaml:"nameChars"`
+	KeyCharacters        *ValidCharacters `yaml:"keyChars"`
+	ValueCharacters      *ValidCharacters `yaml:"valueChars"`
+	ReplacementCharacter string           `yaml:"replacementChar"`
+}
+
 // Supported framework types
 const (
 	// FrameworkTally tally framework id
@@ -175,7 +199,7 @@ const (
 var (
 	safeCharacters = []rune{'_'}
 
-	sanitizeOptions = tally.SanitizeOptions{
+	defaultTallySanitizeOptions = tally.SanitizeOptions{
 		NameCharacters: tally.ValidCharacters{
 			Ranges:     tally.AlphanumericRange,
 			Characters: safeCharacters,
@@ -265,7 +289,18 @@ func NewScope(logger log.Logger, c *Config) tally.Scope {
 		return newStatsdScope(logger, c)
 	}
 	if c.Prometheus != nil {
-		return newPrometheusScope(logger, convertPrometheusConfigToTally(c.Prometheus), &c.ClientConfig)
+		sanitizeOptions, err := convertSanitizeOptionsToTally(c.Prometheus)
+		if err != nil {
+			logger.Fatal("invalid sanitize options input on prometheus config", tag.Error(err))
+			return nil
+		}
+
+		return newPrometheusScope(
+			logger,
+			convertPrometheusConfigToTally(c.Prometheus),
+			sanitizeOptions,
+			&c.ClientConfig,
+		)
 	}
 	return tally.NoopScope
 }
@@ -289,6 +324,14 @@ func convertPrometheusConfigToTally(
 	}
 }
 
+func convertSanitizeOptionsToTally(config *PrometheusConfig) (tally.SanitizeOptions, error) {
+	if config.SanitizeOptions == nil {
+		return defaultTallySanitizeOptions, nil
+	}
+
+	return config.SanitizeOptions.toTally()
+}
+
 func setDefaultPerUnitHistogramBoundaries(clientConfig *ClientConfig) {
 	buckets := maps.Clone(defaultPerUnitHistogramBoundaries)
 
@@ -338,6 +381,7 @@ func newStatsdScope(logger log.Logger, c *Config) tally.Scope {
 func newPrometheusScope(
 	logger log.Logger,
 	config *prometheus.Configuration,
+	sanitizeOptions tally.SanitizeOptions,
 	clientConfig *ClientConfig,
 ) tally.Scope {
 	reporter, err := config.NewReporter(
@@ -396,3 +440,64 @@ func configExcludeTags(cfg ClientConfig) map[string]map[string]struct{} {
 	}
 	return tagsToFilter
 }
+
+func (s SanitizeRange) toTally() (tally.SanitizeRange, error) {
+	startRangeRunes := []rune(s.StartRange)
+	if len(startRangeRunes) != 1 {
+		return tally.SanitizeRange{}, fmt.Errorf("start range '%+v' must be a single rune", startRangeRunes)
+	}
+
+	endRangeRunes := []rune(s.EndRange)
+	if len(endRangeRunes) != 1 {
+		return tally.SanitizeRange{}, fmt.Errorf("end range '%+v' must be a single rune", endRangeRunes)
+	}
+
+	return tally.SanitizeRange([2]rune{startRangeRunes[0], endRangeRunes[0]}), nil
+}
+
+func (v ValidCharacters) toTally() (tally.ValidCharacters, error) {
+	var ranges []tally.SanitizeRange
+
+	for _, r := range v.Ranges {
+		tallyRange, err := r.toTally()
+		if err != nil {
+			return tally.ValidCharacters{}, err
+		}
+
+		ranges = append(ranges, tallyRange)
+	}
+
+	return tally.ValidCharacters{
+		Ranges:     ranges,
+		Characters: []rune(v.SafeCharacters),
+	}, nil
+}
+
+func (s SanitizeOptions) toTally() (tally.SanitizeOptions, error) {
+	tallyNameChars, err := s.NameCharacters.toTally()
+	if err != nil {
+		return tally.SanitizeOptions{}, fmt.Errorf("invalid nameChars: %v", err)
+	}
+
+	tallyKeyChars, err := s.KeyCharacters.toTally()
+	if err != nil {
+		return tally.SanitizeOptions{}, fmt.Errorf("invalid keyChars: %v", err)
+	}
+
+	tallyValueChars, err := s.ValueCharacters.toTally()
+	if err != nil {
+		return tally.SanitizeOptions{}, fmt.Errorf("invalid valueChars: %v", err)
+	}
+
+	replacementChars := []rune(s.ReplacementCharacter)
+	if len(replacementChars) != 1 {
+		return tally.SanitizeOptions{}, errors.New("can only specify a single replacement character")
+	}
+
+	return tally.SanitizeOptions{
+		NameCharacters:       tallyNameChars,
+		KeyCharacters:        tallyKeyChars,
+		ValueCharacters:      tallyValueChars,
+		ReplacementCharacter: replacementChars[0],
+	}, nil
+}

common/metrics/config_test.go

@@ -113,6 +113,42 @@ func (s *MetricsSuite) TestPrometheus() {
 	s.NotNil(scope)
 }
 
+func (s *MetricsSuite) TestPrometheusWithSanitizeOptions() {
+	validChars := &ValidCharacters{
+		Ranges: []SanitizeRange{
+			{
+				StartRange: "a",
+				EndRange:   "z",
+			},
+			{
+				StartRange: "A",
+				EndRange:   "Z",
+			},
+			{
+				StartRange: "0",
+				EndRange:   "9",
+			},
+		},
+		SafeCharacters: "-",
+	}
+
+	prom := &PrometheusConfig{
+		OnError:       "panic",
+		TimerType:     "histogram",
+		ListenAddress: "127.0.0.1:0",
+		SanitizeOptions: &SanitizeOptions{
+			NameCharacters:       validChars,
+			KeyCharacters:        validChars,
+			ValueCharacters:      validChars,
+			ReplacementCharacter: "_",
+		},
+	}
+	config := new(Config)
+	config.Prometheus = prom
+	scope := NewScope(log.NewNoopLogger(), config)
+	s.NotNil(scope)
+}
+
 func (s *MetricsSuite) TestNoop() {
 	config := &Config{}
 	scope := NewScope(log.NewNoopLogger(), config)