UTF-8 support in validation, and some parsers and formatters (#537)

UTF-8 support in validation, and some parsers and formatters

---------

Signed-off-by: Owen Williams <[email protected]>

config/http_config.go

@@ -30,7 +30,7 @@ import (
 	"sync"
 	"time"
 
-	"github.com/mwitkow/go-conntrack"
+	conntrack "github.com/mwitkow/go-conntrack"
 	"golang.org/x/net/http/httpproxy"
 	"golang.org/x/net/http2"
 	"golang.org/x/oauth2"

expfmt/decode_test.go

@@ -17,6 +17,7 @@ import (
 	"bufio"
 	"errors"
 	"io"
+	"math"
 	"net/http"
 	"reflect"
 	"sort"
@@ -104,9 +105,10 @@ func TestProtoDecoder(t *testing.T) {
 	testTime := model.Now()
 
 	scenarios := []struct {
-		in       string
-		expected model.Vector
-		fail     bool
+		in             string
+		expected       model.Vector
+		legacyNameFail bool
+		fail           bool
 	}{
 		{
 			in: "",
@@ -332,6 +334,30 @@ func TestProtoDecoder(t *testing.T) {
 				},
 			},
 		},
+		{
+			in:             "\xa8\x01\n\ngauge.name\x12\x11gauge\ndoc\nstr\"ing\x18\x01\"T\n\x1b\n\x06name.1\x12\x11val with\nnew line\n*\n\x06name*2\x12 val with \\backslash and \"quotes\"\x12\t\t\x00\x00\x00\x00\x00\x00\xf0\x7f\"/\n\x10\n\x06name.1\x12\x06Björn\n\x10\n\x06name*2\x12\x06佖佥\x12\t\t\xd1\xcfD\xb9\xd0\x05\xc2H",
+			legacyNameFail: true,
+			expected: model.Vector{
+				&model.Sample{
+					Metric: model.Metric{
+						model.MetricNameLabel: "gauge.name",
+						"name.1":              "val with\nnew line",
+						"name*2":              "val with \\backslash and \"quotes\"",
+					},
+					Value:     model.SampleValue(math.Inf(+1)),
+					Timestamp: testTime,
+				},
+				&model.Sample{
+					Metric: model.Metric{
+						model.MetricNameLabel: "gauge.name",
+						"name.1":              "Björn",
+						"name*2":              "佖佥",
+					},
+					Value:     3.14e42,
+					Timestamp: testTime,
+				},
+			},
+		},
 	}
 
 	for i, scenario := range scenarios {
@@ -344,11 +370,31 @@ func TestProtoDecoder(t *testing.T) {
 
 		var all model.Vector
 		for {
+			model.NameValidationScheme = model.LegacyValidation
 			var smpls model.Vector
 			err := dec.Decode(&smpls)
 			if err != nil && errors.Is(err, io.EOF) {
 				break
 			}
+			if scenario.legacyNameFail {
+				if err == nil {
+					t.Fatal("Expected error when decoding without UTF-8 support enabled but got none")
+				}
+				model.NameValidationScheme = model.UTF8Validation
+				dec = &SampleDecoder{
+					Dec: &protoDecoder{r: strings.NewReader(scenario.in)},
+					Opts: &DecodeOptions{
+						Timestamp: testTime,
+					},
+				}
+				err = dec.Decode(&smpls)
+				if errors.Is(err, io.EOF) {
+					break
+				}
+				if err != nil {
+					t.Fatalf("Unexpected error when decoding with UTF-8 support: %v", err)
+				}
+			}
 			if scenario.fail {
 				if err == nil {
 					t.Fatal("Expected error but got none")

expfmt/expfmt.go

@@ -17,7 +17,13 @@ package expfmt
 // Format specifies the HTTP content type of the different wire protocols.
 type Format string
 
-// Constants to assemble the Content-Type values for the different wire protocols.
+// Constants to assemble the Content-Type values for the different wire
+// protocols. The Content-Type strings here are all for the legacy exposition
+// formats, where valid characters for metric names and label names are limited.
+// Support for arbitrary UTF-8 characters in those names is already partially
+// implemented in this module (see model.ValidationScheme), but to actually use
+// it on the wire, new content-type strings will have to be agreed upon and
+// added here.
 const (
 	TextVersion              = "0.0.4"
 	ProtoType                = `application/vnd.google.protobuf`

expfmt/openmetrics_create.go

@@ -35,6 +35,18 @@ import (
 // sanity checks. If the input contains duplicate metrics or invalid metric or
 // label names, the conversion will result in invalid text format output.
 //
+// If metric names conform to the legacy validation pattern, they will be placed
+// outside the brackets in the traditional way, like `foo{}`. If the metric name
+// fails the legacy validation check, it will be placed quoted inside the
+// brackets: `{"foo"}`. As stated above, the input is assumed to be santized and
+// no error will be thrown in this case.
+//
+// Similar to metric names, if label names conform to the legacy validation
+// pattern, they will be unquoted as normal, like `foo{bar="baz"}`. If the label
+// name fails the legacy validation check, it will be quoted:
+// `foo{"bar"="baz"}`. As stated above, the input is assumed to be santized and
+// no error will be thrown in this case.
+//
 // This function fulfills the type 'expfmt.encoder'.
 //
 // Note that OpenMetrics requires a final `# EOF` line. Since this function acts
@@ -98,7 +110,7 @@ func MetricFamilyToOpenMetrics(out io.Writer, in *dto.MetricFamily) (written int
 		if err != nil {
 			return
 		}
-		n, err = w.WriteString(shortName)
+		n, err = writeName(w, shortName)
 		written += n
 		if err != nil {
 			return
@@ -124,7 +136,7 @@ func MetricFamilyToOpenMetrics(out io.Writer, in *dto.MetricFamily) (written int
 	if err != nil {
 		return
 	}
-	n, err = w.WriteString(shortName)
+	n, err = writeName(w, shortName)
 	written += n
 	if err != nil {
 		return
@@ -303,21 +315,9 @@ func writeOpenMetricsSample(
 	floatValue float64, intValue uint64, useIntValue bool,
 	exemplar *dto.Exemplar,
 ) (int, error) {
-	var written int
-	n, err := w.WriteString(name)
-	written += n
-	if err != nil {
-		return written, err
-	}
-	if suffix != "" {
-		n, err = w.WriteString(suffix)
-		written += n
-		if err != nil {
-			return written, err
-		}
-	}
-	n, err = writeOpenMetricsLabelPairs(
-		w, metric.Label, additionalLabelName, additionalLabelValue,
+	written := 0
+	n, err := writeOpenMetricsNameAndLabelPairs(
+		w, name+suffix, metric.Label, additionalLabelName, additionalLabelValue,
 	)
 	written += n
 	if err != nil {
@@ -365,27 +365,58 @@ func writeOpenMetricsSample(
 	return written, nil
 }
 
-// writeOpenMetricsLabelPairs works like writeOpenMetrics but formats the float
-// in OpenMetrics style.
-func writeOpenMetricsLabelPairs(
+// writeOpenMetricsNameAndLabelPairs works like writeOpenMetricsSample but
+// formats the float in OpenMetrics style.
+func writeOpenMetricsNameAndLabelPairs(
 	w enhancedWriter,
+	name string,
 	in []*dto.LabelPair,
 	additionalLabelName string, additionalLabelValue float64,
 ) (int, error) {
-	if len(in) == 0 && additionalLabelName == "" {
-		return 0, nil
-	}
 	var (
-		written   int
-		separator byte = '{'
+		written            int
+		separator          byte = '{'
+		metricInsideBraces      = false
 	)
+
+	if name != "" {
+		// If the name does not pass the legacy validity check, we must put the
+		// metric name inside the braces, quoted.
+		if !model.IsValidLegacyMetricName(model.LabelValue(name)) {
+			metricInsideBraces = true
+			err := w.WriteByte(separator)
+			written++
+			if err != nil {
+				return written, err
+			}
+			separator = ','
+		}
+
+		n, err := writeName(w, name)
+		written += n
+		if err != nil {
+			return written, err
+		}
+	}
+
+	if len(in) == 0 && additionalLabelName == "" {
+		if metricInsideBraces {
+			err := w.WriteByte('}')
+			written++
+			if err != nil {
+				return written, err
+			}
+		}
+		return written, nil
+	}
+
 	for _, lp := range in {
 		err := w.WriteByte(separator)
 		written++
 		if err != nil {
 			return written, err
 		}
-		n, err := w.WriteString(lp.GetName())
+		n, err := writeName(w, lp.GetName())
 		written += n
 		if err != nil {
 			return written, err
@@ -451,7 +482,7 @@ func writeExemplar(w enhancedWriter, e *dto.Exemplar) (int, error) {
 	if err != nil {
 		return written, err
 	}
-	n, err = writeOpenMetricsLabelPairs(w, e.Label, "", 0)
+	n, err = writeOpenMetricsNameAndLabelPairs(w, "", e.Label, "", 0)
 	written += n
 	if err != nil {
 		return written, err