promql: Optimize HistogramStatsIterator by disallowing integer histograms

The `HistogramStatsIterator` is only meant to be used within PromQL.
PromQL only ever uses float histograms. If `HistogramStatsIterator` is
capable of handling integer histograms, it will still be used, for
example by the `BufferedSeriesIterator`, which buffers samples and
will use an integer `Histogram` for it, if the underlying chunk is an
integer histogram chunk (which is common).

However, we can simply intercept the `Next` and `Seek` calls and
pretend to only ever be able te return float histograms. This has the
welcome side effect that we do not have to handle a mix of float and
integer histograms in the `HistogramStatsIterator` anymore.

With this commit, the `AtHistogram` call has been changed to panic so
that we ensure it is never called.

Benchmark differences between this and the previous commit:

name                                                                       old time/op    new time/op    delta
NativeHistograms/histogram_count_with_short_rate_interval-16                  837ms ± 3%     616ms ± 2%  -26.36%  (p=0.008 n=5+5)
NativeHistograms/histogram_count_with_long_rate_interval-16                   1.11s ± 1%     0.91s ± 3%  -17.75%  (p=0.008 n=5+5)
NativeHistogramsCustomBuckets/histogram_count_with_short_rate_interval-16     751ms ± 6%     581ms ± 1%  -22.63%  (p=0.008 n=5+5)
NativeHistogramsCustomBuckets/histogram_count_with_long_rate_interval-16      1.13s ±11%     0.85s ± 2%  -24.59%  (p=0.008 n=5+5)

name                                                                       old alloc/op   new alloc/op   delta
NativeHistograms/histogram_count_with_short_rate_interval-16                  531MB ± 0%     148MB ± 0%  -72.08%  (p=0.008 n=5+5)
NativeHistograms/histogram_count_with_long_rate_interval-16                   528MB ± 0%     145MB ± 0%  -72.60%  (p=0.016 n=5+4)
NativeHistogramsCustomBuckets/histogram_count_with_short_rate_interval-16     452MB ± 0%     145MB ± 0%  -67.97%  (p=0.016 n=5+4)
NativeHistogramsCustomBuckets/histogram_count_with_long_rate_interval-16      452MB ± 0%     141MB ± 0%  -68.70%  (p=0.016 n=5+4)

name                                                                       old allocs/op  new allocs/op  delta
NativeHistograms/histogram_count_with_short_rate_interval-16                  8.95M ± 0%     1.60M ± 0%  -82.15%  (p=0.008 n=5+5)
NativeHistograms/histogram_count_with_long_rate_interval-16                   8.84M ± 0%     1.49M ± 0%  -83.16%  (p=0.008 n=5+5)
NativeHistogramsCustomBuckets/histogram_count_with_short_rate_interval-16     5.96M ± 0%     1.57M ± 0%  -73.68%  (p=0.008 n=5+5)
NativeHistogramsCustomBuckets/histogram_count_with_long_rate_interval-16      5.86M ± 0%     1.46M ± 0%  -75.05%  (p=0.016 n=5+4)

Signed-off-by: beorn7 <beorn@grafana.com>
This commit is contained in:
beorn7 2025-09-02 23:35:49 +02:00
parent 87d7c12563
commit 5010bd4bb1
2 changed files with 72 additions and 151 deletions

View file

@ -19,27 +19,24 @@ import (
"github.com/prometheus/prometheus/tsdb/chunkenc"
)
// HistogramStatsIterator is an iterator that returns histogram objects
// which have only their sum and count values populated. The iterator handles
// counter reset detection internally and sets the counter reset hint accordingly
// in each returned histogram object.
// HistogramStatsIterator is an iterator that returns histogram objects that
// have only their sum and count values populated. The iterator handles counter
// reset detection internally and sets the counter reset hint accordingly in
// each returned histogram object. The Next and Seek methods of the iterator
// will never return ValHistogram, but ValFloatHistogram instead. Effectively,
// the iterator enforces conversion of (integer) Histogram to FloatHistogram.
// The AtHistogram method must not be called (and will panic).
type HistogramStatsIterator struct {
chunkenc.Iterator
currentH *histogram.Histogram
lastH *histogram.Histogram
currentFH *histogram.FloatHistogram
lastFH *histogram.FloatHistogram
currentSeriesRead bool
}
// NewHistogramStatsIterator creates a new HistogramStatsIterator.
func NewHistogramStatsIterator(it chunkenc.Iterator) *HistogramStatsIterator {
return &HistogramStatsIterator{
Iterator: it,
currentH: &histogram.Histogram{},
currentFH: &histogram.FloatHistogram{},
}
}
@ -48,44 +45,38 @@ func NewHistogramStatsIterator(it chunkenc.Iterator) *HistogramStatsIterator {
// objects already allocated where possible.
func (hsi *HistogramStatsIterator) Reset(it chunkenc.Iterator) {
hsi.Iterator = it
hsi.currentSeriesRead = false
hsi.lastFH = nil
}
// AtHistogram returns the next timestamp/histogram pair. The counter reset
// detection is guaranteed to be correct only when the caller does not switch
// between AtHistogram and AtFloatHistogram calls.
func (hsi *HistogramStatsIterator) AtHistogram(h *histogram.Histogram) (int64, *histogram.Histogram) {
var t int64
t, hsi.currentH = hsi.Iterator.AtHistogram(hsi.currentH)
if value.IsStaleNaN(hsi.currentH.Sum) {
h = &histogram.Histogram{Sum: hsi.currentH.Sum}
return t, h
// Next mostly relays to the underlying iterator, but changes a ValHistogram
// return into a ValFloatHistogram return.
func (hsi *HistogramStatsIterator) Next() chunkenc.ValueType {
vt := hsi.Iterator.Next()
if vt == chunkenc.ValHistogram {
return chunkenc.ValFloatHistogram
}
if h == nil {
h = &histogram.Histogram{
CounterResetHint: hsi.getResetHint(hsi.currentH),
Count: hsi.currentH.Count,
Sum: hsi.currentH.Sum,
}
hsi.setLastH(hsi.currentH)
return t, h
}
returnValue := histogram.Histogram{
CounterResetHint: hsi.getResetHint(hsi.currentH),
Count: hsi.currentH.Count,
Sum: hsi.currentH.Sum,
}
returnValue.CopyTo(h)
hsi.setLastH(hsi.currentH)
return t, h
return vt
}
// AtFloatHistogram returns the next timestamp/float histogram pair. The counter
// reset detection is guaranteed to be correct only when the caller does not
// switch between AtHistogram and AtFloatHistogram calls.
// Seek mostly relays to the underlying iterator, but changes a ValHistogram
// return into a ValFloatHistogram return.
func (hsi *HistogramStatsIterator) Seek(t int64) chunkenc.ValueType {
vt := hsi.Iterator.Seek(t)
if vt == chunkenc.ValHistogram {
return chunkenc.ValFloatHistogram
}
return vt
}
// AtHistogram must never be called.
func (*HistogramStatsIterator) AtHistogram(*histogram.Histogram) (int64, *histogram.Histogram) {
panic("HistogramStatsIterator.AtHistogram must never be called")
}
// AtFloatHistogram returns the next timestamp/float histogram pair. The method
// performs a counter reset detection on the fly. It will return an explicit
// hint (not UnknownCounterReset) if the previous sample has been accessed with
// the same iterator.
func (hsi *HistogramStatsIterator) AtFloatHistogram(fh *histogram.FloatHistogram) (int64, *histogram.FloatHistogram) {
var t int64
t, hsi.currentFH = hsi.Iterator.AtFloatHistogram(hsi.currentFH)
@ -114,26 +105,12 @@ func (hsi *HistogramStatsIterator) AtFloatHistogram(fh *histogram.FloatHistogram
return t, fh
}
func (hsi *HistogramStatsIterator) setLastH(h *histogram.Histogram) {
hsi.lastFH = nil
if hsi.lastH == nil {
hsi.lastH = h.Copy()
} else {
h.CopyTo(hsi.lastH)
}
hsi.currentSeriesRead = true
}
func (hsi *HistogramStatsIterator) setLastFH(fh *histogram.FloatHistogram) {
hsi.lastH = nil
if hsi.lastFH == nil {
hsi.lastFH = fh.Copy()
} else {
fh.CopyTo(hsi.lastFH)
}
hsi.currentSeriesRead = true
}
func (hsi *HistogramStatsIterator) getFloatResetHint(hint histogram.CounterResetHint) histogram.CounterResetHint {
@ -141,44 +118,19 @@ func (hsi *HistogramStatsIterator) getFloatResetHint(hint histogram.CounterReset
return hint
}
prevFH := hsi.lastFH
if prevFH == nil || !hsi.currentSeriesRead {
if hsi.lastH == nil || !hsi.currentSeriesRead {
// We don't know if there's a counter reset.
return histogram.UnknownCounterReset
}
prevFH = hsi.lastH.ToFloat(nil)
if prevFH == nil {
// We don't know if there's a counter reset. Note that this
// generally will trigger an explicit counter reset detection by
// the PromQL engine, which in turn isn't as reliable in this
// case because the PromQL engine will not see the buckets.
// However, we can assume that in cases where the counter reset
// detection is relevant, an iteration through the series has
// happened, and therefore we do not end up here in the first
// place.
return histogram.UnknownCounterReset
}
if hsi.currentFH.DetectReset(prevFH) {
return histogram.CounterReset
}
return histogram.NotCounterReset
}
func (hsi *HistogramStatsIterator) getResetHint(h *histogram.Histogram) histogram.CounterResetHint {
if h.CounterResetHint != histogram.UnknownCounterReset {
return h.CounterResetHint
}
var prevFH *histogram.FloatHistogram
if hsi.lastH == nil || !hsi.currentSeriesRead {
if hsi.lastFH == nil || !hsi.currentSeriesRead {
// We don't know if there's a counter reset. Note that
// this generally will trigger an explicit counter reset
// detection by the PromQL engine, which in turn isn't
// as reliable in this case because the PromQL engine
// will not see the buckets. However, we can assume that
// in cases where the counter reset detection is
// relevant, an iteration through the series has
// happened, and therefore we do not end up here in the
// first place.
return histogram.UnknownCounterReset
}
prevFH = hsi.lastFH
} else {
prevFH = hsi.lastH.ToFloat(nil)
}
fh := h.ToFloat(nil)
if fh.DetectReset(prevFH) {
return histogram.CounterReset
}
return histogram.NotCounterReset
}

View file

@ -114,64 +114,32 @@ func TestHistogramStatsDecoding(t *testing.T) {
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
t.Run("histogram_stats", func(t *testing.T) {
check := func(statsIterator *HistogramStatsIterator) {
decodedStats := make([]*histogram.Histogram, 0)
for statsIterator.Next() != chunkenc.ValNone {
_, h := statsIterator.AtHistogram(nil)
decodedStats = append(decodedStats, h)
}
for i := 0; i < len(tc.histograms); i++ {
require.Equalf(t, tc.expectedHints[i], decodedStats[i].CounterResetHint, "mismatch in counter reset hint for histogram %d", i)
h := tc.histograms[i]
if value.IsStaleNaN(h.Sum) {
require.True(t, value.IsStaleNaN(decodedStats[i].Sum))
require.Equal(t, uint64(0), decodedStats[i].Count)
} else {
require.Equal(t, tc.histograms[i].Count, decodedStats[i].Count)
require.Equal(t, tc.histograms[i].Sum, decodedStats[i].Sum)
}
check := func(statsIterator *HistogramStatsIterator) {
decodedStats := make([]*histogram.FloatHistogram, 0)
for statsIterator.Next() != chunkenc.ValNone {
_, h := statsIterator.AtFloatHistogram(nil)
decodedStats = append(decodedStats, h)
}
for i := 0; i < len(tc.histograms); i++ {
require.Equal(t, tc.expectedHints[i], decodedStats[i].CounterResetHint)
fh := tc.histograms[i].ToFloat(nil)
if value.IsStaleNaN(fh.Sum) {
require.True(t, value.IsStaleNaN(decodedStats[i].Sum))
require.Equal(t, float64(0), decodedStats[i].Count)
} else {
require.Equal(t, fh.Count, decodedStats[i].Count)
require.Equal(t, fh.Sum, decodedStats[i].Sum)
}
}
}
// Check that we get the expected results with a fresh iterator.
statsIterator := NewHistogramStatsIterator(newHistogramSeries(tc.histograms).Iterator(nil))
check(statsIterator)
// Check that we get the expected results with a fresh iterator.
statsIterator := NewHistogramStatsIterator(newHistogramSeries(tc.histograms).Iterator(nil))
check(statsIterator)
// Check that we get the same results if we reset and reuse that iterator.
statsIterator.Reset(newHistogramSeries(tc.histograms).Iterator(nil))
check(statsIterator)
})
t.Run("float_histogram_stats", func(t *testing.T) {
check := func(statsIterator *HistogramStatsIterator) {
decodedStats := make([]*histogram.FloatHistogram, 0)
for statsIterator.Next() != chunkenc.ValNone {
_, h := statsIterator.AtFloatHistogram(nil)
decodedStats = append(decodedStats, h)
}
for i := 0; i < len(tc.histograms); i++ {
require.Equal(t, tc.expectedHints[i], decodedStats[i].CounterResetHint)
fh := tc.histograms[i].ToFloat(nil)
if value.IsStaleNaN(fh.Sum) {
require.True(t, value.IsStaleNaN(decodedStats[i].Sum))
require.Equal(t, float64(0), decodedStats[i].Count)
} else {
require.Equal(t, fh.Count, decodedStats[i].Count)
require.Equal(t, fh.Sum, decodedStats[i].Sum)
}
}
}
// Check that we get the expected results with a fresh iterator.
statsIterator := NewHistogramStatsIterator(newHistogramSeries(tc.histograms).Iterator(nil))
check(statsIterator)
// Check that we get the same results if we reset and reuse that iterator.
statsIterator.Reset(newHistogramSeries(tc.histograms).Iterator(nil))
check(statsIterator)
})
// Check that we get the same results if we reset and reuse that iterator.
statsIterator.Reset(newHistogramSeries(tc.histograms).Iterator(nil))
check(statsIterator)
})
}
}
@ -193,17 +161,18 @@ func TestHistogramStatsMixedUse(t *testing.T) {
histogram.NotCounterReset,
histogram.CounterReset,
}
// Note that statsIterator always returns float histograms.
actualHints := make([]histogram.CounterResetHint, 3)
typ := statsIterator.Next()
require.Equal(t, chunkenc.ValHistogram, typ)
_, h := statsIterator.AtHistogram(nil)
require.Equal(t, chunkenc.ValFloatHistogram, typ)
_, h := statsIterator.AtFloatHistogram(nil)
actualHints[0] = h.CounterResetHint
typ = statsIterator.Next()
require.Equal(t, chunkenc.ValHistogram, typ)
_, h = statsIterator.AtHistogram(nil)
require.Equal(t, chunkenc.ValFloatHistogram, typ)
_, h = statsIterator.AtFloatHistogram(nil)
actualHints[1] = h.CounterResetHint
typ = statsIterator.Next()
require.Equal(t, chunkenc.ValHistogram, typ)
require.Equal(t, chunkenc.ValFloatHistogram, typ)
_, fh := statsIterator.AtFloatHistogram(nil)
actualHints[2] = fh.CounterResetHint