Capture decoding time metric (#3209)

We're trying to track down some slowness.

pkg/decoders/base64.go

@@ -28,7 +28,7 @@ func init() {
 }
 
 func (d *Base64) FromChunk(chunk *sources.Chunk) *DecodableChunk {
-	decodableChunk := &DecodableChunk{Chunk: chunk, DecoderType: detectorspb.DecoderType_BASE64}
+	decodableChunk := &DecodableChunk{Chunk: chunk, DecoderType: d.Type()}
 	encodedSubstrings := getSubstringsOfCharacterSet(chunk.Data, 20, b64CharsetMapping, b64EndChars)
 	decodedSubstrings := make(map[string][]byte)
 
@@ -67,6 +67,10 @@ func (d *Base64) FromChunk(chunk *sources.Chunk) *DecodableChunk {
 	return nil
 }
 
+func (d *Base64) Type() detectorspb.DecoderType {
+	return detectorspb.DecoderType_BASE64
+}
+
 func isASCII(b []byte) bool {
 	for i := 0; i < len(b); i++ {
 		if b[i] > unicode.MaxASCII {

pkg/decoders/decoders.go

@@ -24,6 +24,7 @@ type DecodableChunk struct {
 
 type Decoder interface {
 	FromChunk(chunk *sources.Chunk) *DecodableChunk
+	Type() detectorspb.DecoderType
 }
 
 // Fuzz is an entrypoint for go-fuzz, which is an AFL-style fuzzing tool.

pkg/decoders/escaped_unicode.go

@@ -41,7 +41,7 @@ func (d *EscapedUnicode) FromChunk(chunk *sources.Chunk) *DecodableChunk {
 
 	if matched {
 		decodableChunk := &DecodableChunk{
-			DecoderType: detectorspb.DecoderType_ESCAPED_UNICODE,
+			DecoderType: d.Type(),
 			Chunk:       chunk,
 		}
 		return decodableChunk
@@ -94,6 +94,10 @@ func decodeCodePoint(input []byte) []byte {
 	return input
 }
 
+func (d *EscapedUnicode) Type() detectorspb.DecoderType {
+	return detectorspb.DecoderType_ESCAPED_UNICODE
+}
+
 func decodeEscaped(input []byte) []byte {
 	// Find all Unicode escape sequences in the input byte slice
 	indices := escapePat.FindAllSubmatchIndex(input, -1)

pkg/decoders/utf16.go

@@ -16,7 +16,7 @@ func (d *UTF16) FromChunk(chunk *sources.Chunk) *DecodableChunk {
 		return nil
 	}
 
-	decodableChunk := &DecodableChunk{Chunk: chunk, DecoderType: detectorspb.DecoderType_UTF16}
+	decodableChunk := &DecodableChunk{Chunk: chunk, DecoderType: d.Type()}
 	if utf16Data, err := utf16ToUTF8(chunk.Data); err == nil {
 		if len(utf16Data) == 0 {
 			return nil
@@ -28,6 +28,10 @@ func (d *UTF16) FromChunk(chunk *sources.Chunk) *DecodableChunk {
 	return nil
 }
 
+func (d *UTF16) Type() detectorspb.DecoderType {
+	return detectorspb.DecoderType_UTF16
+}
+
 // utf16ToUTF8 converts a byte slice containing UTF-16 encoded data to a UTF-8 encoded byte slice.
 func utf16ToUTF8(b []byte) ([]byte, error) {
 	var bufBE, bufLE bytes.Buffer

pkg/decoders/utf8.go

@@ -15,7 +15,7 @@ func (d *UTF8) FromChunk(chunk *sources.Chunk) *DecodableChunk {
 		return nil
 	}
 
-	decodableChunk := &DecodableChunk{Chunk: chunk, DecoderType: detectorspb.DecoderType_PLAIN}
+	decodableChunk := &DecodableChunk{Chunk: chunk, DecoderType: d.Type()}
 
 	if !utf8.Valid(chunk.Data) {
 		chunk.Data = extractSubstrings(chunk.Data)
@@ -25,6 +25,10 @@ func (d *UTF8) FromChunk(chunk *sources.Chunk) *DecodableChunk {
 	return decodableChunk
 }
 
+func (d *UTF8) Type() detectorspb.DecoderType {
+	return detectorspb.DecoderType_PLAIN
+}
+
 // extractSubstrings performs similarly to the strings binutil,
 // extacting contigous portions of printable characters that we care
 // about from some bytes

pkg/engine/engine.go

@@ -733,7 +733,11 @@ func (e *Engine) scannerWorker(ctx context.Context) {
 		startTime := time.Now()
 		sourceVerify := chunk.Verify
 		for _, decoder := range e.decoders {
+			decodeStart := time.Now()
 			decoded := decoder.FromChunk(chunk)
+			decodeTime := time.Since(decodeStart).Microseconds()
+			decodeLatency.WithLabelValues(decoder.Type().String(), chunk.SourceName).Observe(float64(decodeTime))
+
 			if decoded == nil {
 				ctx.Logger().V(4).Info("no decoder found for chunk", "chunk", chunk)
 				continue

pkg/engine/metrics.go

@@ -8,6 +8,17 @@ import (
 )
 
 var (
+	decodeLatency = promauto.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: common.MetricsNamespace,
+			Subsystem: common.MetricsSubsystem,
+			Name:      "decode_latency",
+			Help:      "Time spent decoding a chunk in microseconds",
+			Buckets:   prometheus.ExponentialBuckets(50, 2, 20),
+		},
+		[]string{"decoder_type", "source_name"},
+	)
+
 	// Detector metrics.
 	detectorExecutionCount = promauto.NewCounterVec(
 		prometheus.CounterOpts{