Implement max bucket limit and automatic resolution reduction

CHANGELOG.md

@@ -9,6 +9,7 @@
 * [FEATURE] Query Frontend: Added a query rejection mechanism to block resource-intensive queries. #6005
 * [FEATURE] OTLP: Support ingesting OTLP exponential metrics as native histograms. #6071
 * [FEATURE] Ingester: Add `ingester.instance-limits.max-inflight-query-requests` to allow limiting ingester concurrent queries. #6081
+* [FEATURE] Distributor: Add `validation.max-native-histogram-buckets` to limit max number of bucket count. Distributor will try to automatically reduce histogram resolution until it is within the bucket limit or resolution cannot be reduced anymore. #6104
 * [ENHANCEMENT] rulers: Add support to persist tokens in rulers. #5987
 * [ENHANCEMENT] Query Frontend/Querier: Added store gateway postings touched count and touched size in Querier stats and log in Query Frontend. #5892
 * [ENHANCEMENT] Query Frontend/Querier: Returns `warnings` on prometheus query responses. #5916

docs/configuration/config-file-reference.md

@@ -3159,11 +3159,13 @@ The `limits_config` configures default and per-tenant limits imposed by Cortex s
 # e.g. remote_write.write_relabel_configs.
 [metric_relabel_configs: <relabel_config...> | default = []]
 
-# Enables support for exemplars in TSDB and sets the maximum number that will be
-# stored. less than zero means disabled. If the value is set to zero, cortex
-# will fallback to blocks-storage.tsdb.max-exemplars value.
-# CLI flag: -ingester.max-exemplars
-[max_exemplars: <int> | default = 0]
+# Limit on total number of positive and negative buckets allowed in a single
+# native histogram. The resolution of a histogram with more buckets will be
+# reduced until the number of buckets is within the limit. If the limit cannot
+# be reached, the sample will be discarded. 0 means no limit. Enforced at
+# Distributor.
+# CLI flag: -validation.max-native-histogram-buckets
+[max_native_histogram_buckets: <int> | default = 0]
 
 # The maximum number of active series per user, per ingester. 0 to disable.
 # CLI flag: -ingester.max-series-per-user
@@ -3213,6 +3215,12 @@ The `limits_config` configures default and per-tenant limits imposed by Cortex s
 # CLI flag: -ingester.out-of-order-time-window
 [out_of_order_time_window: <duration> | default = 0s]
 
+# Enables support for exemplars in TSDB and sets the maximum number that will be
+# stored. less than zero means disabled. If the value is set to zero, cortex
+# will fallback to blocks-storage.tsdb.max-exemplars value.
+# CLI flag: -ingester.max-exemplars
+[max_exemplars: <int> | default = 0]
+
 # Maximum number of chunks that can be fetched in a single query from ingesters
 # and long-term storage. This limit is enforced in the querier, ruler and
 # store-gateway. 0 to disable.

pkg/cortexpb/histograms.go

@@ -15,6 +15,11 @@ package cortexpb
 
 import "github.com/prometheus/prometheus/model/histogram"
 
+const (
+	ExponentialSchemaMax int32 = 8
+	ExponentialSchemaMin int32 = -4
+)
+
 func (h Histogram) IsFloatHistogram() bool {
 	_, ok := h.GetCount().(*Histogram_CountFloat)
 	return ok

pkg/distributor/distributor.go

@@ -579,12 +579,16 @@ func (d *Distributor) validateSeries(ts cortexpb.PreallocTimeseries, userID stri
 	if len(ts.Histograms) > 0 {
 		// Only alloc when data present
 		histograms = make([]cortexpb.Histogram, 0, len(ts.Histograms))
-		for _, h := range ts.Histograms {
-			// TODO(yeya24): add other validations for native histogram.
-			// For example, Prometheus scrape has bucket limit and schema check.
+		for i, h := range ts.Histograms {
 			if err := validation.ValidateSampleTimestamp(d.validateMetrics, limits, userID, ts.Labels, h.TimestampMs); err != nil {
 				return emptyPreallocSeries, err
 			}
+			// TODO(yeya24): add max schema validation for native histogram if needed.
+			convertedHistogram, err := validation.ValidateNativeHistogram(d.validateMetrics, limits, userID, ts.Labels, h)
+			if err != nil {
+				return emptyPreallocSeries, err
+			}
+			ts.Histograms[i] = convertedHistogram
 		}
 		histograms = append(histograms, ts.Histograms...)
 	}

pkg/util/validation/errors.go

@@ -225,6 +225,24 @@ func newExemplarLabelLengthError(seriesLabels []cortexpb.LabelAdapter, exemplarL
 	}
 }
 
+// histogramBucketLimitExceededError is a ValidationError implementation for samples with native histogram
+// exceeding max bucket limit and cannot reduce resolution further to be within the max bucket limit.
+type histogramBucketLimitExceededError struct {
+	series []cortexpb.LabelAdapter
+	limit  int
+}
+
+func newHistogramBucketLimitExceededError(series []cortexpb.LabelAdapter, limit int) ValidationError {
+	return &histogramBucketLimitExceededError{
+		series: series,
+		limit:  limit,
+	}
+}
+
+func (e *histogramBucketLimitExceededError) Error() string {
+	return fmt.Sprintf("native histogram bucket count exceeded for metric (limit: %d) metric: %.200q", e.limit, formatLabelSet(e.series))
+}
+
 // formatLabelSet formats label adapters as a metric name with labels, while preserving
 // label order, and keeping duplicates. If there are multiple "__name__" labels, only
 // first one is used as metric name, other ones will be included as regular labels.

pkg/util/validation/limits.go

@@ -134,7 +134,7 @@ type Limits struct {
 	EnforceMetricName         bool                `yaml:"enforce_metric_name" json:"enforce_metric_name"`
 	IngestionTenantShardSize  int                 `yaml:"ingestion_tenant_shard_size" json:"ingestion_tenant_shard_size"`
 	MetricRelabelConfigs      []*relabel.Config   `yaml:"metric_relabel_configs,omitempty" json:"metric_relabel_configs,omitempty" doc:"nocli|description=List of metric relabel configurations. Note that in most situations, it is more effective to use metrics relabeling directly in the Prometheus server, e.g. remote_write.write_relabel_configs."`
-	MaxExemplars              int                 `yaml:"max_exemplars" json:"max_exemplars"`
+	MaxNativeHistogramBuckets int                 `yaml:"max_native_histogram_buckets" json:"max_native_histogram_buckets"`
 
 	// Ingester enforced limits.
 	// Series
@@ -151,6 +151,8 @@ type Limits struct {
 	MaxGlobalMetadataPerMetric          int `yaml:"max_global_metadata_per_metric" json:"max_global_metadata_per_metric"`
 	// Out-of-order
 	OutOfOrderTimeWindow model.Duration `yaml:"out_of_order_time_window" json:"out_of_order_time_window"`
+	// Exemplars
+	MaxExemplars int `yaml:"max_exemplars" json:"max_exemplars"`
 
 	// Querier enforced limits.
 	MaxChunksPerQuery            int            `yaml:"max_fetched_chunks_per_query" json:"max_fetched_chunks_per_query"`
@@ -232,6 +234,7 @@ func (l *Limits) RegisterFlags(f *flag.FlagSet) {
 	f.Var(&l.CreationGracePeriod, "validation.create-grace-period", "Duration which table will be created/deleted before/after it's needed; we won't accept sample from before this time.")
 	f.BoolVar(&l.EnforceMetricName, "validation.enforce-metric-name", true, "Enforce every sample has a metric name.")
 	f.BoolVar(&l.EnforceMetadataMetricName, "validation.enforce-metadata-metric-name", true, "Enforce every metadata has a metric name.")
+	f.IntVar(&l.MaxNativeHistogramBuckets, "validation.max-native-histogram-buckets", 0, "Limit on total number of positive and negative buckets allowed in a single native histogram. The resolution of a histogram with more buckets will be reduced until the number of buckets is within the limit. If the limit cannot be reached, the sample will be discarded. 0 means no limit. Enforced at Distributor.")
 
 	f.IntVar(&l.MaxLocalSeriesPerUser, "ingester.max-series-per-user", 5000000, "The maximum number of active series per user, per ingester. 0 to disable.")
 	f.IntVar(&l.MaxLocalSeriesPerMetric, "ingester.max-series-per-metric", 50000, "The maximum number of active series per metric name, per ingester. 0 to disable.")
@@ -722,6 +725,12 @@ func (o *Overrides) EnforceMetadataMetricName(userID string) bool {
 	return o.GetOverridesForUser(userID).EnforceMetadataMetricName
 }
 
+// MaxNativeHistogramBuckets returns the maximum total number of positive and negative buckets of a single native histogram
+// a user is allowed to store.
+func (o *Overrides) MaxNativeHistogramBuckets(userID string) int {
+	return o.GetOverridesForUser(userID).MaxNativeHistogramBuckets
+}
+
 // MaxLocalMetricsWithMetadataPerUser returns the maximum number of metrics with metadata a user is allowed to store in a single ingester.
 func (o *Overrides) MaxLocalMetricsWithMetadataPerUser(userID string) int {
 	return o.GetOverridesForUser(userID).MaxLocalMetricsWithMetadataPerUser

pkg/util/validation/validate.go

@@ -52,6 +52,9 @@ const (
 	exemplarLabelsTooLong    = "exemplar_labels_too_long"
 	exemplarTimestampInvalid = "exemplar_timestamp_invalid"
 
+	// Native Histogram specific validation reasons
+	nativeHistogramBucketCountLimitExceeded = "native_histogram_buckets_exceeded"
+
 	// RateLimited is one of the values for the reason to discard samples.
 	// Declared here to avoid duplication in ingester and distributor.
 	RateLimited = "rate_limited"
@@ -262,6 +265,59 @@ func ValidateMetadata(validateMetrics *ValidateMetrics, cfg *Limits, userID stri
 	return nil
 }
 
+func ValidateNativeHistogram(validateMetrics *ValidateMetrics, limits *Limits, userID string, ls []cortexpb.LabelAdapter, histogram cortexpb.Histogram) (cortexpb.Histogram, error) {
+	if limits.MaxNativeHistogramBuckets == 0 {
+		return histogram, nil
+	}
+
+	var (
+		exceedLimit bool
+	)
+	if histogram.IsFloatHistogram() {
+		// Initial check to see if the bucket limit is exceeded or not. If not, we can avoid type casting.
+		exceedLimit = len(histogram.PositiveCounts)+len(histogram.NegativeCounts) > limits.MaxNativeHistogramBuckets
+		if !exceedLimit {
+			return histogram, nil
+		}
+		// Exceed limit.
+		if histogram.Schema <= cortexpb.ExponentialSchemaMin {
+			validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID).Inc()
+			return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
+		}
+		fh := cortexpb.FloatHistogramProtoToFloatHistogram(histogram)
+		for len(fh.PositiveBuckets)+len(fh.NegativeBuckets) > limits.MaxNativeHistogramBuckets {
+			if fh.Schema <= cortexpb.ExponentialSchemaMin {
+				validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID).Inc()
+				return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
+			}
+			fh = fh.ReduceResolution(fh.Schema - 1)
+		}
+		// If resolution reduced, convert new float histogram to protobuf type again.
+		return cortexpb.FloatHistogramToHistogramProto(histogram.TimestampMs, fh), nil
+	}
+
+	// Initial check to see if bucket limit is exceeded or not. If not, we can avoid type casting.
+	exceedLimit = len(histogram.PositiveDeltas)+len(histogram.NegativeDeltas) > limits.MaxNativeHistogramBuckets
+	if !exceedLimit {
+		return histogram, nil
+	}
+	// Exceed limit.
+	if histogram.Schema <= cortexpb.ExponentialSchemaMin {
+		validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID).Inc()
+		return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
+	}
+	h := cortexpb.HistogramProtoToHistogram(histogram)
+	for len(h.PositiveBuckets)+len(h.NegativeBuckets) > limits.MaxNativeHistogramBuckets {
+		if h.Schema <= cortexpb.ExponentialSchemaMin {
+			validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID).Inc()
+			return cortexpb.Histogram{}, newHistogramBucketLimitExceededError(ls, limits.MaxNativeHistogramBuckets)
+		}
+		h = h.ReduceResolution(h.Schema - 1)
+	}
+	// If resolution reduced, convert new histogram to protobuf type again.
+	return cortexpb.HistogramToHistogramProto(histogram.TimestampMs, h), nil
+}
+
 func DeletePerUserValidationMetrics(validateMetrics *ValidateMetrics, userID string, log log.Logger) {
 	filter := map[string]string{"user": userID}
 

pkg/util/validation/validate_test.go

@@ -8,6 +8,8 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/testutil"
 	"github.com/prometheus/common/model"
+	"github.com/prometheus/prometheus/model/labels"
+	"github.com/prometheus/prometheus/tsdb/tsdbutil"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"github.com/weaveworks/common/httpgrpc"
@@ -291,3 +293,110 @@ func TestValidateLabelDuplication(t *testing.T) {
 	}, "a")
 	assert.Equal(t, expected, actual)
 }
+
+func TestValidateNativeHistogram(t *testing.T) {
+	userID := "fake"
+	lbls := cortexpb.FromLabelsToLabelAdapters(labels.FromStrings("foo", "bar"))
+
+	// Test histogram has 4 positive buckets and 4 negative buckets so 8 in total. Schema set to 1.
+	h := tsdbutil.GenerateTestHistogram(0)
+	fh := tsdbutil.GenerateTestFloatHistogram(0)
+
+	histogramWithSchemaMin := tsdbutil.GenerateTestHistogram(0)
+	histogramWithSchemaMin.Schema = cortexpb.ExponentialSchemaMin
+	floatHistogramWithSchemaMin := tsdbutil.GenerateTestFloatHistogram(0)
+	floatHistogramWithSchemaMin.Schema = cortexpb.ExponentialSchemaMin
+	for _, tc := range []struct {
+		name              string
+		bucketLimit       int
+		histogram         cortexpb.Histogram
+		expectedHistogram cortexpb.Histogram
+		expectedErr       error
+	}{
+		{
+			name:              "no limit, histogram",
+			histogram:         cortexpb.HistogramToHistogramProto(0, h.Copy()),
+			expectedHistogram: cortexpb.HistogramToHistogramProto(0, h.Copy()),
+		},
+		{
+			name:              "no limit, float histogram",
+			histogram:         cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
+			expectedHistogram: cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
+		},
+		{
+			name:              "within limit, histogram",
+			bucketLimit:       8,
+			histogram:         cortexpb.HistogramToHistogramProto(0, h.Copy()),
+			expectedHistogram: cortexpb.HistogramToHistogramProto(0, h.Copy()),
+		},
+		{
+			name:              "within limit, float histogram",
+			bucketLimit:       8,
+			histogram:         cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
+			expectedHistogram: cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
+		},
+		{
+			name:              "exceed limit and reduce resolution for 1 level, histogram",
+			bucketLimit:       6,
+			histogram:         cortexpb.HistogramToHistogramProto(0, h.Copy()),
+			expectedHistogram: cortexpb.HistogramToHistogramProto(0, h.Copy().ReduceResolution(0)),
+		},
+		{
+			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)),
+		},
+		{
+			name:              "exceed limit and reduce resolution for 2 levels, histogram",
+			bucketLimit:       4,
+			histogram:         cortexpb.HistogramToHistogramProto(0, h.Copy()),
+			expectedHistogram: cortexpb.HistogramToHistogramProto(0, h.Copy().ReduceResolution(-1)),
+		},
+		{
+			name:              "exceed limit and reduce resolution for 2 levels, float histogram",
+			bucketLimit:       4,
+			histogram:         cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
+			expectedHistogram: cortexpb.FloatHistogramToHistogramProto(0, fh.Copy().ReduceResolution(-1)),
+		},
+		{
+			name:        "exceed limit but cannot reduce resolution further, histogram",
+			bucketLimit: 1,
+			histogram:   cortexpb.HistogramToHistogramProto(0, h.Copy()),
+			expectedErr: newHistogramBucketLimitExceededError(lbls, 1),
+		},
+		{
+			name:        "exceed limit but cannot reduce resolution further, float histogram",
+			bucketLimit: 1,
+			histogram:   cortexpb.FloatHistogramToHistogramProto(0, fh.Copy()),
+			expectedErr: newHistogramBucketLimitExceededError(lbls, 1),
+		},
+		{
+			name:        "exceed limit but cannot reduce resolution further with min schema, histogram",
+			bucketLimit: 4,
+			histogram:   cortexpb.HistogramToHistogramProto(0, histogramWithSchemaMin.Copy()),
+			expectedErr: newHistogramBucketLimitExceededError(lbls, 4),
+		},
+		{
+			name:        "exceed limit but cannot reduce resolution further with min schema, float histogram",
+			bucketLimit: 4,
+			histogram:   cortexpb.FloatHistogramToHistogramProto(0, floatHistogramWithSchemaMin.Copy()),
+			expectedErr: newHistogramBucketLimitExceededError(lbls, 4),
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			reg := prometheus.NewRegistry()
+			validateMetrics := NewValidateMetrics(reg)
+			limits := new(Limits)
+			limits.MaxNativeHistogramBuckets = tc.bucketLimit
+			actualHistogram, actualErr := ValidateNativeHistogram(validateMetrics, limits, userID, lbls, tc.histogram)
+			if tc.expectedErr != nil {
+				require.Equal(t, tc.expectedErr, actualErr)
+				require.Equal(t, float64(1), testutil.ToFloat64(validateMetrics.DiscardedSamples.WithLabelValues(nativeHistogramBucketCountLimitExceeded, userID)))
+			} else {
+				require.NoError(t, actualErr)
+				require.Equal(t, tc.expectedHistogram, actualHistogram)
+			}
+		})
+	}
+}