Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

histogram: Enable detection of a native histogram without observations #1314

Merged
merged 2 commits into from Jul 26, 2023
Merged

histogram: Enable detection of a native histogram without observations #1314

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f9db382
histograms: Small code comment and code formatting improvements
beorn7 Apr 8, 2023
3d82c94
histogram: Add a no-op span for an otherwise empty histogram
beorn7 Jul 20, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
66 changes: 42 additions & 24 deletions prometheus/histogram.go
View file
Open in desktop
Expand Up @@ -428,12 +428,12 @@ type HistogramOpts struct {
// a major version bump.
NativeHistogramBucketFactor float64
// All observations with an absolute value of less or equal
// NativeHistogramZeroThreshold are accumulated into a “zero”
// bucket. For best results, this should be close to a bucket
// boundary. This is usually the case if picking a power of two. If
// NativeHistogramZeroThreshold are accumulated into a “zero” bucket.
// For best results, this should be close to a bucket boundary. This is
// usually the case if picking a power of two. If
// NativeHistogramZeroThreshold is left at zero,
// DefNativeHistogramZeroThreshold is used as the threshold. To configure
// a zero bucket with an actual threshold of zero (i.e. only
// DefNativeHistogramZeroThreshold is used as the threshold. To
// configure a zero bucket with an actual threshold of zero (i.e. only
// observations of precisely zero will go into the zero bucket), set
// NativeHistogramZeroThreshold to the NativeHistogramZeroThresholdZero
// constant (or any negative float value).
Expand All @@ -446,23 +446,28 @@ type HistogramOpts struct {
// Histogram are sufficiently wide-spread. In particular, this could be
// used as a DoS attack vector. Where the observed values depend on
// external inputs, it is highly recommended to set a
// NativeHistogramMaxBucketNumber.) Once the set
// NativeHistogramMaxBucketNumber.) Once the set
// NativeHistogramMaxBucketNumber is exceeded, the following strategy is
// enacted: First, if the last reset (or the creation) of the histogram
// is at least NativeHistogramMinResetDuration ago, then the whole
// histogram is reset to its initial state (including regular
// buckets). If less time has passed, or if
// NativeHistogramMinResetDuration is zero, no reset is
// performed. Instead, the zero threshold is increased sufficiently to
// reduce the number of buckets to or below
// NativeHistogramMaxBucketNumber, but not to more than
// NativeHistogramMaxZeroThreshold. Thus, if
// NativeHistogramMaxZeroThreshold is already at or below the current
// zero threshold, nothing happens at this step. After that, if the
// number of buckets still exceeds NativeHistogramMaxBucketNumber, the
// resolution of the histogram is reduced by doubling the width of the
// sparse buckets (up to a growth factor between one bucket to the next
// of 2^(2^4) = 65536, see above).
// enacted:
// - First, if the last reset (or the creation) of the histogram is at
// least NativeHistogramMinResetDuration ago, then the whole
// histogram is reset to its initial state (including regular
// buckets).
// - If less time has passed, or if NativeHistogramMinResetDuration is
// zero, no reset is performed. Instead, the zero threshold is
// increased sufficiently to reduce the number of buckets to or below
// NativeHistogramMaxBucketNumber, but not to more than
// NativeHistogramMaxZeroThreshold. Thus, if
// NativeHistogramMaxZeroThreshold is already at or below the current
// zero threshold, nothing happens at this step.
// - After that, if the number of buckets still exceeds
// NativeHistogramMaxBucketNumber, the resolution of the histogram is
// reduced by doubling the width of the sparse buckets (up to a
// growth factor between one bucket to the next of 2^(2^4) = 65536,
// see above).
// - Any increased zero threshold or reduced resolution is reset back
// to their original values once NativeHistogramMinResetDuration has
// passed (since the last reset or the creation of the histogram).
NativeHistogramMaxBucketNumber uint32
NativeHistogramMinResetDuration time.Duration
NativeHistogramMaxZeroThreshold float64
Expand Down Expand Up @@ -782,6 +787,16 @@ func (h *histogram) Write(out *dto.Metric) error {
his.ZeroCount = proto.Uint64(zeroBucket)
his.NegativeSpan, his.NegativeDelta = makeBuckets(&coldCounts.nativeHistogramBucketsNegative)
his.PositiveSpan, his.PositiveDelta = makeBuckets(&coldCounts.nativeHistogramBucketsPositive)

// Add a no-op span to a histogram without observations and with
// a zero threshold of zero. Otherwise, a native histogram would
// look like a classic histogram to scrapers.
if *his.ZeroThreshold == 0 && *his.ZeroCount == 0 && len(his.PositiveSpan) == 0 && len(his.NegativeSpan) == 0 {
his.PositiveSpan = []*dto.BucketSpan{{
Offset: proto.Int32(0),
Length: proto.Uint32(0),
}}
}
}
addAndResetCounts(hotCounts, coldCounts)
return nil
Expand Down Expand Up @@ -854,13 +869,16 @@ func (h *histogram) limitBuckets(counts *histogramCounts, value float64, bucket
h.doubleBucketWidth(hotCounts, coldCounts)
}

// maybeReset resests the whole histogram if at least h.nativeHistogramMinResetDuration
// maybeReset resets the whole histogram if at least h.nativeHistogramMinResetDuration
// has been passed. It returns true if the histogram has been reset. The caller
// must have locked h.mtx.
func (h *histogram) maybeReset(hot, cold *histogramCounts, coldIdx uint64, value float64, bucket int) bool {
func (h *histogram) maybeReset(
hot, cold *histogramCounts, coldIdx uint64, value float64, bucket int,
) bool {
// We are using the possibly mocked h.now() rather than
// time.Since(h.lastResetTime) to enable testing.
if h.nativeHistogramMinResetDuration == 0 || h.now().Sub(h.lastResetTime) < h.nativeHistogramMinResetDuration {
if h.nativeHistogramMinResetDuration == 0 ||
h.now().Sub(h.lastResetTime) < h.nativeHistogramMinResetDuration {
return false
}
// Completely reset coldCounts.
Expand Down
11 changes: 11 additions & 0 deletions prometheus/histogram_test.go
View file
Open in desktop
Expand Up @@ -485,6 +485,17 @@ func TestNativeHistogram(t *testing.T) {
factor: 1,
want: `sample_count:3 sample_sum:6 bucket:<cumulative_count:0 upper_bound:0.005 > bucket:<cumulative_count:0 upper_bound:0.01 > bucket:<cumulative_count:0 upper_bound:0.025 > bucket:<cumulative_count:0 upper_bound:0.05 > bucket:<cumulative_count:0 upper_bound:0.1 > bucket:<cumulative_count:0 upper_bound:0.25 > bucket:<cumulative_count:0 upper_bound:0.5 > bucket:<cumulative_count:1 upper_bound:1 > bucket:<cumulative_count:2 upper_bound:2.5 > bucket:<cumulative_count:3 upper_bound:5 > bucket:<cumulative_count:3 upper_bound:10 > `, // Has conventional buckets because there are no sparse buckets.
},
{
name: "no observations",
factor: 1.1,
want: `sample_count:0 sample_sum:0 schema:3 zero_threshold:2.938735877055719e-39 zero_count:0 `,
},
{
name: "no observations and zero threshold of zero resulting in no-op span",
factor: 1.1,
zeroThreshold: NativeHistogramZeroThresholdZero,
want: `sample_count:0 sample_sum:0 schema:3 zero_threshold:0 zero_count:0 positive_span:<offset:0 length:0 > `,
},
{
name: "factor 1.1 results in schema 3",
observations: []float64{0, 1, 2, 3},
Expand Down