Feature request for histogram-like type that keeps raw durations by bucket size

[arnikola]

Recently came across a use case which is difficult to represent with native Tally types. I had a function that dealt with many different cardinalities of data, and wanted to capture metrics by bucket size, similar to a histogram.

While this worked for raw count, I was looking for some way to capture the duration by the same buckets, which is currently not possible.

Below is the workaround I used to achieve this in case my explanation is hard to follow

type bucketMetric struct {
	threshold int
	timing    tally.Timer
}

func newBucketMetric(threshold int, value string, scope tally.Scope) *bucketMetric {
	return &bucketMetric{
		threshold: threshold,
		timing:    scope.Timer("timing"),
	}
}

type bucketMetrics struct {
	countHistogram  tally.Histogram
	durationMetrics []*bucketMetric
}

func newBucketMetrics(metricType string, scope tally.Scope) bucketMetrics {
	buckets := tally.ValueBuckets([]float64{50, 200, 1000, 5000, 10000, math.MaxInt32})

	return bucketMetrics{
		countHistogram: scope.Histogram("count", buckets),
		durationMetrics: []*bucketMetric{
			newBucketMetric(0, "0", scope),
			newBucketMetric(50, "50", scope),
			newBucketMetric(200, "200", scope),
			newBucketMetric(1000, "1000", scope),
			newBucketMetric(5000, "5000", scope),
			newBucketMetric(10000, "10000", scope),
			newBucketMetric(int(math.MaxInt32), "max", scope),
		},
	}
}

func (bm bucketMetrics) record(cardinality int, duration time.Duration) {
	// Record count on the general bucketMetrics histogram
	bm.countHistogram.RecordValue(float64(cardinality))

	for _, m := range bm.durationMetrics {
		if cardinality <= m.threshold {
			m.timing.Record(duration)
			return
		}
	}
}