|
| 1 | +// Unless explicitly stated otherwise all files in this repository are licensed |
| 2 | +// under the Apache License Version 2.0. |
| 3 | +// This product includes software developed at Datadog (https://www.datadoghq.com/). |
| 4 | +// Copyright 2022-present Datadog, Inc. |
| 5 | + |
| 6 | +package metrics |
| 7 | + |
| 8 | +import ( |
| 9 | + "fmt" |
| 10 | + "math" |
| 11 | + "time" |
| 12 | + |
| 13 | + "go.opentelemetry.io/collector/pdata/pcommon" |
| 14 | + "go.opentelemetry.io/collector/pdata/pmetric" |
| 15 | + |
| 16 | + "github.com/DataDog/sketches-go/ddsketch" |
| 17 | + "github.com/DataDog/sketches-go/ddsketch/mapping" |
| 18 | + "github.com/DataDog/sketches-go/ddsketch/store" |
| 19 | +) |
| 20 | + |
| 21 | +// These methods will be imported in dd-go to convert OTLP to DD trace metrics - see https://github.com/DataDog/dd-go/pull/198508 |
| 22 | + |
| 23 | +// getTimeUnitScaleToNanos returns the scaling factor to convert the given unit to nanoseconds |
| 24 | +func getTimeUnitScaleToNanos(unit string) float64 { |
| 25 | + switch unit { |
| 26 | + case "ns": |
| 27 | + return float64(time.Nanosecond) |
| 28 | + case "us", "μs": |
| 29 | + return float64(time.Microsecond) |
| 30 | + case "ms": |
| 31 | + return float64(time.Millisecond) |
| 32 | + case "s": |
| 33 | + return float64(time.Second) |
| 34 | + case "min": |
| 35 | + return float64(time.Minute) |
| 36 | + case "h": |
| 37 | + return float64(time.Hour) |
| 38 | + default: |
| 39 | + // If unit is unknown, assume seconds (common for duration metrics) |
| 40 | + return float64(time.Second) |
| 41 | + } |
| 42 | +} |
| 43 | + |
| 44 | +// getBounds returns the lower and upper bounds for a histogram bucket |
| 45 | +func getBounds(explicitBounds pcommon.Float64Slice, idx int) (lowerBound float64, upperBound float64) { |
| 46 | + // See https://github.com/open-telemetry/opentelemetry-proto/blob/v0.10.0/opentelemetry/proto/metrics/v1/metrics.proto#L427-L439 |
| 47 | + lowerBound = math.Inf(-1) |
| 48 | + upperBound = math.Inf(1) |
| 49 | + if idx > 0 { |
| 50 | + lowerBound = explicitBounds.At(idx - 1) |
| 51 | + } |
| 52 | + if idx < explicitBounds.Len() { |
| 53 | + upperBound = explicitBounds.At(idx) |
| 54 | + } |
| 55 | + return |
| 56 | +} |
| 57 | + |
| 58 | +// CreateDDSketchFromHistogramOfDuration creates a DDSketch from regular histogram data point |
| 59 | +func CreateDDSketchFromHistogramOfDuration(dp pmetric.HistogramDataPoint, unit string) (*ddsketch.DDSketch, error) { |
| 60 | + relativeAccuracy := 0.01 // 1% relative accuracy |
| 61 | + maxNumBins := 2048 |
| 62 | + newSketch, err := ddsketch.LogCollapsingLowestDenseDDSketch(relativeAccuracy, maxNumBins) |
| 63 | + if err != nil { |
| 64 | + return nil, err |
| 65 | + } |
| 66 | + |
| 67 | + bucketCounts := dp.BucketCounts() |
| 68 | + explicitBounds := dp.ExplicitBounds() |
| 69 | + |
| 70 | + // Get scaling factor to convert unit to nanoseconds |
| 71 | + scaleToNanos := getTimeUnitScaleToNanos(unit) |
| 72 | + |
| 73 | + // Find first and last bucket indices with count > 0 |
| 74 | + lowestBucketIndex := -1 |
| 75 | + highestBucketIndex := -1 |
| 76 | + for j := 0; j < bucketCounts.Len(); j++ { |
| 77 | + count := bucketCounts.At(j) |
| 78 | + if count > 0 { |
| 79 | + if lowestBucketIndex == -1 { |
| 80 | + lowestBucketIndex = j |
| 81 | + } |
| 82 | + highestBucketIndex = j |
| 83 | + } |
| 84 | + } |
| 85 | + |
| 86 | + hasMin := dp.HasMin() |
| 87 | + hasMax := dp.HasMax() |
| 88 | + minNanoseconds := dp.Min() * scaleToNanos |
| 89 | + maxNanoseconds := dp.Max() * scaleToNanos |
| 90 | + |
| 91 | + for j := 0; j < bucketCounts.Len(); j++ { |
| 92 | + lowerBound, upperBound := getBounds(explicitBounds, j) |
| 93 | + |
| 94 | + if math.IsInf(upperBound, 1) { |
| 95 | + upperBound = lowerBound |
| 96 | + } else if math.IsInf(lowerBound, -1) { |
| 97 | + lowerBound = upperBound |
| 98 | + } |
| 99 | + |
| 100 | + count := bucketCounts.At(j) |
| 101 | + |
| 102 | + if count > 0 { |
| 103 | + insertionPoint := 0.0 |
| 104 | + adjustedCount := float64(count) |
| 105 | + midpoint := (lowerBound + upperBound) / 2 * scaleToNanos |
| 106 | + // Determine insertion point based on bucket position |
| 107 | + if j == lowestBucketIndex && j == highestBucketIndex { |
| 108 | + // Special case: min and max are in the same bucket |
| 109 | + if hasMin && hasMax { |
| 110 | + insertionPoint = (minNanoseconds + maxNanoseconds) / 2 |
| 111 | + } |
| 112 | + } else if j == lowestBucketIndex { |
| 113 | + // Bottom bucket: insert at min value |
| 114 | + if hasMin { |
| 115 | + insertionPoint = minNanoseconds |
| 116 | + } |
| 117 | + } else if j == highestBucketIndex { |
| 118 | + // Top bucket: insert at max value |
| 119 | + if hasMax { |
| 120 | + insertionPoint = maxNanoseconds |
| 121 | + } |
| 122 | + } |
| 123 | + |
| 124 | + if insertionPoint == 0.0 { |
| 125 | + insertionPoint = midpoint |
| 126 | + } |
| 127 | + |
| 128 | + err := newSketch.AddWithCount(insertionPoint, adjustedCount) |
| 129 | + if err != nil { |
| 130 | + return nil, fmt.Errorf("failed to add value to DDSketch: %w", err) |
| 131 | + } |
| 132 | + } |
| 133 | + } |
| 134 | + |
| 135 | + return newSketch, nil |
| 136 | +} |
| 137 | + |
| 138 | +func toStoreFromExponentialBucketsWithUnitScale(b pmetric.ExponentialHistogramDataPointBuckets, mapping *mapping.LogarithmicMapping, base float64, scaleToNanos float64) store.Store { |
| 139 | + offset := b.Offset() |
| 140 | + bucketCounts := b.BucketCounts() |
| 141 | + |
| 142 | + store := store.NewDenseStore() |
| 143 | + for j := 0; j < bucketCounts.Len(); j++ { |
| 144 | + bucketIndex := j + int(offset) |
| 145 | + count := bucketCounts.At(j) |
| 146 | + |
| 147 | + if count > 0 { |
| 148 | + // Calculate the actual bucket boundary value |
| 149 | + bucketValue := math.Pow(base, float64(bucketIndex)) |
| 150 | + |
| 151 | + // Scale the bucket value to nanoseconds |
| 152 | + scaledValue := bucketValue * scaleToNanos |
| 153 | + |
| 154 | + // Convert back to the index in the nanosecond space |
| 155 | + // Using the same gamma since we're keeping the same precision |
| 156 | + scaledIndex := mapping.Index(scaledValue) |
| 157 | + store.AddWithCount(scaledIndex, float64(count)) |
| 158 | + } |
| 159 | + } |
| 160 | + return store |
| 161 | +} |
| 162 | + |
| 163 | +// CreateDDSketchFromExponentialHistogramOfDuration creates a DDSketch from exponential histogram data point |
| 164 | +func CreateDDSketchFromExponentialHistogramOfDuration(p pmetric.ExponentialHistogramDataPoint, unit string) (*ddsketch.DDSketch, error) { |
| 165 | + // Create the DDSketch stores |
| 166 | + scaleToNanos := getTimeUnitScaleToNanos(unit) |
| 167 | + |
| 168 | + // Create the DDSketch mapping that corresponds to the ExponentialHistogram settings |
| 169 | + gammaWithOnePercentAccuracy := 1.01 / 0.99 |
| 170 | + gamma := math.Pow(2, math.Pow(2, float64(-p.Scale()))) |
| 171 | + gamma = math.Min(gamma, gammaWithOnePercentAccuracy) |
| 172 | + indexOffset := math.Log(scaleToNanos) |
| 173 | + mapping, err := mapping.NewLogarithmicMappingWithGamma(gamma, indexOffset) |
| 174 | + if err != nil { |
| 175 | + return nil, fmt.Errorf("couldn't create LogarithmicMapping for DDSketch: %w", err) |
| 176 | + } |
| 177 | + |
| 178 | + // Calculate the base for the exponential histogram |
| 179 | + base := math.Pow(2, math.Pow(2, float64(-p.Scale()))) |
| 180 | + positiveStore := toStoreFromExponentialBucketsWithUnitScale(p.Positive(), mapping, base, scaleToNanos) |
| 181 | + negativeStore := toStoreFromExponentialBucketsWithUnitScale(p.Negative(), mapping, base, scaleToNanos) |
| 182 | + |
| 183 | + // Create DDSketch with the above mapping and stores |
| 184 | + sketch := ddsketch.NewDDSketch(mapping, positiveStore, negativeStore) |
| 185 | + err = sketch.AddWithCount(0, float64(p.ZeroCount())) |
| 186 | + if err != nil { |
| 187 | + return nil, fmt.Errorf("failed to add ZeroCount to DDSketch: %w", err) |
| 188 | + } |
| 189 | + |
| 190 | + return sketch, nil |
| 191 | +} |
0 commit comments