Create cortex_reduced_resolution_histogram_samples_total metric (cortexproject#6182)

* Create new metric to count how many native histrogram samples had the resolution reduced

Signed-off-by: alanprot <alanprot@gmail.com>

* changelog

Signed-off-by: alanprot <alanprot@gmail.com>

---------

Signed-off-by: alanprot <alanprot@gmail.com>

Author: alanprot

CHANGELOG.md

@@ -4,6 +4,7 @@
 
 * [ENHANCEMENT] Ruler: Add new ruler metric `cortex_ruler_rule_groups_in_store` that is the total rule groups per tenant in store, which can be used to compare with `cortex_prometheus_rule_group_rules` to count the number of rule groups that are not loaded by a ruler. #5869
 * [ENHANCEMENT] Ruler: Add query statistics metrics when --ruler.query-stats-enabled=true. #6173
+* [ENHANCEMENT] Distributor: Add new `cortex_reduced_resolution_histogram_samples_total` metric to to track the number of histogram samples which resolution was reduced. #6182
 
 ## 1.18.0 2024-09-03
 

pkg/util/validation/validate.go

@@ -74,9 +74,10 @@ const (
 )
 
 type ValidateMetrics struct {
-	DiscardedSamples   *prometheus.CounterVec
-	DiscardedExemplars *prometheus.CounterVec
-	DiscardedMetadata  *prometheus.CounterVec
+	DiscardedSamples                  *prometheus.CounterVec
+	DiscardedExemplars                *prometheus.CounterVec
+	DiscardedMetadata                 *prometheus.CounterVec
+	HistogramSamplesReducedResolution *prometheus.CounterVec
 }
 
 func registerCollector(r prometheus.Registerer, c prometheus.Collector) {
@@ -111,10 +112,19 @@ func NewValidateMetrics(r prometheus.Registerer) *ValidateMetrics {
 		[]string{discardReasonLabel, "user"},
 	)
 	registerCollector(r, discardedMetadata)
+	histogramSamplesReducedResolution := prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Name: "cortex_reduced_resolution_histogram_samples_total",
+			Help: "The total number of histogram samples that had the resolution reduced.",
+		},
+		[]string{"user"},
+	)
+	registerCollector(r, histogramSamplesReducedResolution)
 	m := &ValidateMetrics{
-		DiscardedSamples:   discardedSamples,
-		DiscardedExemplars: discardedExemplars,
-		DiscardedMetadata:  discardedMetadata,
+		DiscardedSamples:                  discardedSamples,
+		DiscardedExemplars:                discardedExemplars,
+		DiscardedMetadata:                 discardedMetadata,
+		HistogramSamplesReducedResolution: histogramSamplesReducedResolution,
 	}
 
 	return m
@@ -286,13 +296,17 @@ func ValidateNativeHistogram(validateMetrics *ValidateMetrics, limits *Limits, u
 			return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
 		}
 		fh := cortexpb.FloatHistogramProtoToFloatHistogram(histogramSample)
+		oBuckets := len(fh.PositiveBuckets) + len(fh.NegativeBuckets)
 		for len(fh.PositiveBuckets)+len(fh.NegativeBuckets) > limits.MaxNativeHistogramBuckets {
 			if fh.Schema <= histogram.ExponentialSchemaMin {
 				validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID).Inc()
 				return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
 			}
 			fh = fh.ReduceResolution(fh.Schema - 1)
 		}
+		if oBuckets != len(fh.PositiveBuckets)+len(fh.NegativeBuckets) {
+			validateMetrics.HistogramSamplesReducedResolution.WithLabelValues(userID).Inc()
+		}
 		// If resolution reduced, convert new float histogram to protobuf type again.
 		return cortexpb.FloatHistogramToHistogramProto(histogramSample.TimestampMs, fh), nil
 	}
@@ -308,13 +322,17 @@ func ValidateNativeHistogram(validateMetrics *ValidateMetrics, limits *Limits, u
 		return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
 	}
 	h := cortexpb.HistogramProtoToHistogram(histogramSample)
+	oBuckets := len(h.PositiveBuckets) + len(h.NegativeBuckets)
 	for len(h.PositiveBuckets)+len(h.NegativeBuckets) > limits.MaxNativeHistogramBuckets {
 		if h.Schema <= histogram.ExponentialSchemaMin {
 			validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID).Inc()
 			return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
 		}
 		h = h.ReduceResolution(h.Schema - 1)
 	}
+	if oBuckets != len(h.PositiveBuckets)+len(h.NegativeBuckets) {
+		validateMetrics.HistogramSamplesReducedResolution.WithLabelValues(userID).Inc()
+	}
 	// If resolution reduced, convert new histogram to protobuf type again.
 	return cortexpb.HistogramToHistogramProto(histogramSample.TimestampMs, h), nil
 }
@@ -331,4 +349,7 @@ func DeletePerUserValidationMetrics(validateMetrics *ValidateMetrics, userID str
 	if err := util.DeleteMatchingLabels(validateMetrics.DiscardedMetadata, filter); err != nil {
 		level.Warn(log).Log("msg", "failed to remove cortex_discarded_metadata_total metric for user", "user", userID, "err", err)
 	}
+	if err := util.DeleteMatchingLabels(validateMetrics.HistogramSamplesReducedResolution, filter); err != nil {
+		level.Warn(log).Log("msg", "failed to remove cortex_reduced_resolution_histogram_samples_total metric for user", "user", userID, "err", err)
+	}
 }

pkg/util/validation/validate_test.go

@@ -310,6 +310,7 @@ func TestValidateNativeHistogram(t *testing.T) {
 	for _, tc := range []struct {
 		name              string
 		bucketLimit       int
+		resolutionReduced bool
 		histogram         cortexpb.Histogram
 		expectedHistogram cortexpb.Histogram
 		expectedErr       error
@@ -341,12 +342,14 @@ func TestValidateNativeHistogram(t *testing.T) {
 			bucketLimit:       6,
 			histogram:         cortexpb.HistogramToHistogramProto(0, h.Copy()),
 			expectedHistogram: cortexpb.HistogramToHistogramProto(0, h.Copy().ReduceResolution(0)),
+			resolutionReduced: true,
 		},
 		{
 			name:              "exceed limit and reduce resolution for 1 level, float histogram",
 			bucketLimit:       6,
 			histogram:         cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
 			expectedHistogram: cortexpb.FloatHistogramToHistogramProto(0, fh.Copy().ReduceResolution(0)),
+			resolutionReduced: true,
 		},
 		{
 			name:              "exceed limit and reduce resolution for 2 levels, histogram",
@@ -394,7 +397,13 @@ func TestValidateNativeHistogram(t *testing.T) {
 			if tc.expectedErr != nil {
 				require.Equal(t, tc.expectedErr, actualErr)
 				require.Equal(t, float64(1), testutil.ToFloat64(validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID)))
+				// Should never increment if error was returned
+				require.Equal(t, float64(0), testutil.ToFloat64(validateMetrics.HistogramSamplesReducedResolution.WithLabelValues(userID)))
+
 			} else {
+				if tc.resolutionReduced {
+					require.Equal(t, float64(1), testutil.ToFloat64(validateMetrics.HistogramSamplesReducedResolution.WithLabelValues(userID)))
+				}
 				require.NoError(t, actualErr)
 				require.Equal(t, tc.expectedHistogram, actualHistogram)
 			}