You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
prometheus/tsdb/chunkenc/histogram_meta.go

589 lines
17 KiB

// Copyright 2021 The Prometheus Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package chunkenc
import (
"math"
"github.com/prometheus/prometheus/model/histogram"
)
func writeHistogramChunkLayout(
b *bstream, schema int32, zeroThreshold float64,
positiveSpans, negativeSpans []histogram.Span, customValues []float64,
) {
putZeroThreshold(b, zeroThreshold)
putVarbitInt(b, int64(schema))
putHistogramChunkLayoutSpans(b, positiveSpans)
putHistogramChunkLayoutSpans(b, negativeSpans)
if histogram.IsCustomBucketsSchema(schema) {
putHistogramChunkLayoutCustomBounds(b, customValues)
}
}
func readHistogramChunkLayout(b *bstreamReader) (
schema int32, zeroThreshold float64,
positiveSpans, negativeSpans []histogram.Span,
customValues []float64,
err error,
) {
zeroThreshold, err = readZeroThreshold(b)
if err != nil {
return
}
v, err := readVarbitInt(b)
if err != nil {
return
}
schema = int32(v)
positiveSpans, err = readHistogramChunkLayoutSpans(b)
if err != nil {
return
}
negativeSpans, err = readHistogramChunkLayoutSpans(b)
if err != nil {
return
}
if histogram.IsCustomBucketsSchema(schema) {
customValues, err = readHistogramChunkLayoutCustomBounds(b)
if err != nil {
return
}
}
return
}
func putHistogramChunkLayoutSpans(b *bstream, spans []histogram.Span) {
putVarbitUint(b, uint64(len(spans)))
for _, s := range spans {
putVarbitUint(b, uint64(s.Length))
putVarbitInt(b, int64(s.Offset))
}
}
func readHistogramChunkLayoutSpans(b *bstreamReader) ([]histogram.Span, error) {
var spans []histogram.Span
num, err := readVarbitUint(b)
if err != nil {
return nil, err
}
for i := 0; i < int(num); i++ {
length, err := readVarbitUint(b)
if err != nil {
return nil, err
}
offset, err := readVarbitInt(b)
if err != nil {
return nil, err
}
spans = append(spans, histogram.Span{
Length: uint32(length),
Offset: int32(offset),
})
}
return spans, nil
}
func putHistogramChunkLayoutCustomBounds(b *bstream, customValues []float64) {
putVarbitUint(b, uint64(len(customValues)))
for _, bound := range customValues {
putCustomBound(b, bound)
}
}
func readHistogramChunkLayoutCustomBounds(b *bstreamReader) ([]float64, error) {
var customValues []float64
num, err := readVarbitUint(b)
if err != nil {
return nil, err
}
for i := 0; i < int(num); i++ {
bound, err := readCustomBound(b)
if err != nil {
return nil, err
}
customValues = append(customValues, bound)
}
return customValues, nil
}
// putZeroThreshold writes the zero threshold to the bstream. It stores typical
// values in just one byte, but needs 9 bytes for other values. In detail:
// - If the threshold is 0, store a single zero byte.
// - If the threshold is a power of 2 between (and including) 2^-243 and 2^10,
// take the exponent from the IEEE 754 representation of the threshold, which
// covers a range between (and including) -242 and 11. (2^-243 is 0.5*2^-242
// in IEEE 754 representation, and 2^10 is 0.5*2^11.) Add 243 to the exponent
// and store the result (which will be between 1 and 254) as a single
// byte. Note that small powers of two are preferred values for the zero
// threshold. The default value for the zero threshold is 2^-128 (or
// 0.5*2^-127 in IEEE 754 representation) and will therefore be encoded as a
// single byte (with value 116).
// - In all other cases, store 255 as a single byte, followed by the 8 bytes of
// the threshold as a float64, i.e. taking 9 bytes in total.
func putZeroThreshold(b *bstream, threshold float64) {
if threshold == 0 {
b.writeByte(0)
return
}
frac, exp := math.Frexp(threshold)
if frac != 0.5 || exp < -242 || exp > 11 {
b.writeByte(255)
b.writeBits(math.Float64bits(threshold), 64)
return
}
b.writeByte(byte(exp + 243))
}
// readZeroThreshold reads the zero threshold written with putZeroThreshold.
func readZeroThreshold(br *bstreamReader) (float64, error) {
b, err := br.ReadByte()
if err != nil {
return 0, err
}
switch b {
case 0:
return 0, nil
case 255:
v, err := br.readBits(64)
if err != nil {
return 0, err
}
return math.Float64frombits(v), nil
default:
return math.Ldexp(0.5, int(b)-243), nil
}
}
// isWholeWhenMultiplied checks to see if the number when multiplied by 1000 can
// be converted into an integer without losing precision.
func isWholeWhenMultiplied(in float64) bool {
i := uint(math.Round(in * 1000))
out := float64(i) / 1000
return in == out
}
// putCustomBound writes a custom bound to the bstream. It stores values from
// 0 to 33554.430 (inclusive) that are multiples of 0.001 in unsigned varbit
// encoding of up to 4 bytes, but needs 1 bit + 8 bytes for other values like
// negative numbers, numbers greater than 33554.430, or numbers that are not
// a multiple of 0.001, on the assumption that they are less common. In detail:
// - Multiply the bound by 1000, without rounding.
// - If the multiplied bound is >= 0, <= 33554430 and a whole number,
// add 1 and store it in unsigned varbit encoding. All these numbers are
// greater than 0, so the leading bit of the varbit is always 1!
// - Otherwise, store a 0 bit, followed by the 8 bytes of the original
// bound as a float64.
//
// When reading the values, we can first decode a value as unsigned varbit,
// if it's 0, then we read the next 8 bytes as a float64, otherwise
// we can convert the value to a float64 by subtracting 1 and dividing by 1000.
func putCustomBound(b *bstream, f float64) {
tf := f * 1000
// 33554431-1 comes from the maximum that can be stored in a varbit in 4
// bytes, other values are stored in 8 bytes anyway.
if tf < 0 || tf > 33554430 || !isWholeWhenMultiplied(f) {
b.writeBit(zero)
b.writeBits(math.Float64bits(f), 64)
return
}
putVarbitUint(b, uint64(math.Round(tf))+1)
}
// readCustomBound reads the custom bound written with putCustomBound.
func readCustomBound(br *bstreamReader) (float64, error) {
b, err := readVarbitUint(br)
if err != nil {
return 0, err
}
switch b {
case 0:
v, err := br.readBits(64)
if err != nil {
return 0, err
}
return math.Float64frombits(v), nil
default:
return float64(b-1) / 1000, nil
}
}
type bucketIterator struct {
spans []histogram.Span
span int // Span position of last yielded bucket.
bucket int // Bucket position within span of last yielded bucket.
idx int // Bucket index (globally across all spans) of last yielded bucket.
}
func newBucketIterator(spans []histogram.Span) *bucketIterator {
b := bucketIterator{
spans: spans,
span: 0,
bucket: -1,
idx: -1,
}
if len(spans) > 0 {
b.idx += int(spans[0].Offset)
}
return &b
}
func (b *bucketIterator) Next() (int, bool) {
// We're already out of bounds.
if b.span >= len(b.spans) {
return 0, false
}
if b.bucket < int(b.spans[b.span].Length)-1 { // Try to move within same span.
b.bucket++
b.idx++
return b.idx, true
}
for b.span < len(b.spans)-1 { // Try to move from one span to the next.
b.span++
b.idx += int(b.spans[b.span].Offset + 1)
b.bucket = 0
if b.spans[b.span].Length == 0 {
b.idx--
continue
}
return b.idx, true
}
// We're out of options.
return 0, false
}
// An Insert describes how many new buckets have to be inserted before
// processing the pos'th bucket from the original slice.
type Insert struct {
pos int
num int
}
// expandSpansForward returns the inserts to expand the bucket spans 'a' so that
// they match the spans in 'b'. 'b' must cover the same or more buckets than
// 'a', otherwise the function will return false.
//
// Example:
//
// Let's say the old buckets look like this:
//
// span syntax: [offset, length]
// spans : [ 0 , 2 ] [2,1] [ 3 , 2 ] [3,1] [1,1]
// bucket idx : [0] [1] 2 3 [4] 5 6 7 [8] [9] 10 11 12 [13] 14 [15]
// raw values 6 3 3 2 4 5 1
// deltas 6 -3 0 -1 2 1 -4
//
// But now we introduce a new bucket layout. (Carefully chosen example where we
// have a span appended, one unchanged[*], one prepended, and two merge - in
// that order.)
//
// [*] unchanged in terms of which bucket indices they represent. but to achieve
// that, their offset needs to change if "disrupted" by spans changing ahead of
// them
//
// \/ this one is "unchanged"
// spans : [ 0 , 3 ] [1,1] [ 1 , 4 ] [ 3 , 3 ]
// bucket idx : [0] [1] [2] 3 [4] 5 [6] [7] [8] [9] 10 11 12 [13] [14] [15]
// raw values 6 3 0 3 0 0 2 4 5 0 1
// deltas 6 -3 -3 3 -3 0 2 2 1 -5 1
// delta mods: / \ / \ / \
//
// Note for histograms with delta-encoded buckets: Whenever any new buckets are
// introduced, the subsequent "old" bucket needs to readjust its delta to the
// new base of 0. Thus, for the caller who wants to transform the set of
// original deltas to a new set of deltas to match a new span layout that adds
// buckets, we simply need to generate a list of inserts.
//
// Note: Within expandSpansForward we don't have to worry about the changes to the
// spans themselves, thanks to the iterators we get to work with the more useful
// bucket indices (which of course directly correspond to the buckets we have to
// adjust).
func expandSpansForward(a, b []histogram.Span) (forward []Insert, ok bool) {
ai := newBucketIterator(a)
bi := newBucketIterator(b)
var inserts []Insert
// When inter.num becomes > 0, this becomes a valid insert that should
// be yielded when we finish a streak of new buckets.
var inter Insert
av, aOK := ai.Next()
bv, bOK := bi.Next()
loop:
for {
switch {
case aOK && bOK:
switch {
case av == bv: // Both have an identical value. move on!
// Finish WIP insert and reset.
if inter.num > 0 {
inserts = append(inserts, inter)
}
inter.num = 0
av, aOK = ai.Next()
bv, bOK = bi.Next()
inter.pos++
case av < bv: // b misses a value that is in a.
return inserts, false
case av > bv: // a misses a value that is in b. Forward b and recompare.
inter.num++
bv, bOK = bi.Next()
}
case aOK && !bOK: // b misses a value that is in a.
return inserts, false
case !aOK && bOK: // a misses a value that is in b. Forward b and recompare.
inter.num++
bv, bOK = bi.Next()
default: // Both iterators ran out. We're done.
if inter.num > 0 {
inserts = append(inserts, inter)
}
break loop
}
}
return inserts, true
}
// expandSpansBothWays is similar to expandSpansForward, but now b may also
// cover an entirely different set of buckets. The function returns the
// “forward” inserts to expand 'a' to also cover all the buckets exclusively
// covered by 'b', and it returns the “backward” inserts to expand 'b' to also
// cover all the buckets exclusively covered by 'a'.
func expandSpansBothWays(a, b []histogram.Span) (forward, backward []Insert, mergedSpans []histogram.Span) {
ai := newBucketIterator(a)
bi := newBucketIterator(b)
var fInserts, bInserts []Insert
var lastBucket int
addBucket := func(b int) {
offset := b - lastBucket - 1
if offset == 0 && len(mergedSpans) > 0 {
mergedSpans[len(mergedSpans)-1].Length++
} else {
if len(mergedSpans) == 0 {
offset++
}
mergedSpans = append(mergedSpans, histogram.Span{
Offset: int32(offset),
Length: 1,
})
}
lastBucket = b
}
// When fInter.num (or bInter.num, respectively) becomes > 0, this
// becomes a valid insert that should be yielded when we finish a streak
// of new buckets.
var fInter, bInter Insert
av, aOK := ai.Next()
bv, bOK := bi.Next()
loop:
for {
switch {
case aOK && bOK:
switch {
case av == bv: // Both have an identical value. move on!
// Finish WIP insert and reset.
if fInter.num > 0 {
fInserts = append(fInserts, fInter)
fInter.num = 0
}
if bInter.num > 0 {
bInserts = append(bInserts, bInter)
bInter.num = 0
}
addBucket(av)
av, aOK = ai.Next()
bv, bOK = bi.Next()
fInter.pos++
bInter.pos++
case av < bv: // b misses a value that is in a.
bInter.num++
// Collect the forward inserts before advancing
// the position of 'a'.
if fInter.num > 0 {
fInserts = append(fInserts, fInter)
fInter.num = 0
}
addBucket(av)
fInter.pos++
av, aOK = ai.Next()
case av > bv: // a misses a value that is in b. Forward b and recompare.
fInter.num++
// Collect the backward inserts before advancing the
// position of 'b'.
if bInter.num > 0 {
bInserts = append(bInserts, bInter)
bInter.num = 0
}
addBucket(bv)
bInter.pos++
bv, bOK = bi.Next()
}
case aOK && !bOK: // b misses a value that is in a.
bInter.num++
addBucket(av)
av, aOK = ai.Next()
case !aOK && bOK: // a misses a value that is in b. Forward b and recompare.
fInter.num++
addBucket(bv)
bv, bOK = bi.Next()
default: // Both iterators ran out. We're done.
if fInter.num > 0 {
fInserts = append(fInserts, fInter)
}
if bInter.num > 0 {
bInserts = append(bInserts, bInter)
}
break loop
}
}
return fInserts, bInserts, mergedSpans
}
type bucketValue interface {
int64 | float64
}
// insert merges 'in' with the provided inserts and writes them into 'out',
// which must already have the appropriate length. 'out' is also returned for
// convenience.
func insert[BV bucketValue](in, out []BV, inserts []Insert, deltas bool) []BV {
var (
oi int // Position in out.
v BV // The last value seen.
ii int // The next insert to process.
)
for i, d := range in {
if ii < len(inserts) && i == inserts[ii].pos {
// We have an insert!
// Add insert.num new delta values such that their
// bucket values equate 0. When deltas==false, it means
// that it is an absolute value. So we set it to 0
// directly.
if deltas {
out[oi] = -v
} else {
out[oi] = 0
}
oi++
for x := 1; x < inserts[ii].num; x++ {
out[oi] = 0
oi++
}
ii++
// Now save the value from the input. The delta value we
// should save is the original delta value + the last
// value of the point before the insert (to undo the
// delta that was introduced by the insert). When
// deltas==false, it means that it is an absolute value,
// so we set it directly to the value in the 'in' slice.
if deltas {
out[oi] = d + v
} else {
out[oi] = d
}
oi++
v = d + v
continue
}
// If there was no insert, the original delta is still valid.
out[oi] = d
oi++
v += d
}
switch ii {
case len(inserts):
// All inserts processed. Nothing more to do.
case len(inserts) - 1:
// One more insert to process at the end.
if deltas {
out[oi] = -v
} else {
out[oi] = 0
}
oi++
for x := 1; x < inserts[ii].num; x++ {
out[oi] = 0
oi++
}
default:
panic("unprocessed inserts left")
}
return out
}
// counterResetHint returns a CounterResetHint based on the CounterResetHeader
// and on the position into the chunk.
func counterResetHint(crh CounterResetHeader, numRead uint16) histogram.CounterResetHint {
switch {
case crh == GaugeType:
// A gauge histogram chunk only contains gauge histograms.
return histogram.GaugeType
case numRead > 1:
// In a counter histogram chunk, there will not be any counter
// resets after the first histogram.
return histogram.NotCounterReset
case crh == CounterReset:
// If the chunk was started because of a counter reset, we can
// safely return that hint. This histogram always has to be
// treated as a counter reset.
return histogram.CounterReset
default:
// Sadly, we have to return "unknown" as the hint for all other
// cases, even if we know that the chunk was started without a
// counter reset. But we cannot be sure that the previous chunk
// still exists in the TSDB, so we conservatively return
// "unknown". On the bright side, this case should be relatively
// rare.
//
// TODO(beorn7): Nevertheless, if the current chunk is in the
// middle of a block (not the first chunk in the block for this
// series), it's probably safe to assume that the previous chunk
// will exist in the TSDB for as long as the current chunk
// exist, and we could safely return
// "histogram.NotCounterReset". This needs some more work and
// might not be worth the effort and/or risk. To be vetted...
return histogram.UnknownCounterReset
}
}
// Handle pathological case of empty span when advancing span idx.
// Call it with idx==-1 to find the first non empty span.
func nextNonEmptySpanSliceIdx(idx int, bucketIdx int32, spans []histogram.Span) (newIdx int, newBucketIdx int32) {
for idx++; idx < len(spans); idx++ {
if spans[idx].Length > 0 {
return idx, bucketIdx + spans[idx].Offset + 1
}
bucketIdx += spans[idx].Offset
}
return idx, 0
}