-
Notifications
You must be signed in to change notification settings - Fork 512
/
partitioner.go
182 lines (158 loc) · 6.06 KB
/
partitioner.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
// SPDX-License-Identifier: AGPL-3.0-only
// Provenance-includes-location: https://github.com/cortexproject/cortex/blob/master/pkg/storegateway/partitioner.go
// Provenance-includes-license: Apache-2.0
// Provenance-includes-copyright: The Cortex Authors.
// Provenance-includes-location: https://github.com/thanos-io/thanos/blob/main/pkg/store/bucket.go
// Provenance-includes-license: Apache-2.0
// Provenance-includes-copyright: The Thanos Authors.
package storegateway
import (
"encoding/binary"
"io"
"github.com/efficientgo/core/errors"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/promauto"
)
type blockPartitioners struct {
chunks, series, postings Partitioner
}
func newGapBasedPartitioners(maxGapBytes uint64, reg prometheus.Registerer) blockPartitioners {
return blockPartitioners{
chunks: newGapBasedPartitioner(maxGapBytes, prometheus.WrapRegistererWith(map[string]string{"data_type": "chunks"}, reg)),
series: newGapBasedPartitioner(maxGapBytes, prometheus.WrapRegistererWith(map[string]string{"data_type": "series"}, reg)),
postings: newGapBasedPartitioner(maxGapBytes, prometheus.WrapRegistererWith(map[string]string{"data_type": "postings"}, reg)),
}
}
type gapBasedPartitioner struct {
maxGapBytes uint64
// Metrics.
requestedBytes prometheus.Counter
requestedRanges prometheus.Counter
expandedBytes prometheus.Counter
expandedRanges prometheus.Counter
extendedRanges prometheus.Counter
}
func newGapBasedPartitioner(maxGapBytes uint64, reg prometheus.Registerer) *gapBasedPartitioner {
return &gapBasedPartitioner{
maxGapBytes: maxGapBytes,
requestedBytes: promauto.With(reg).NewCounter(prometheus.CounterOpts{
Name: "cortex_bucket_store_partitioner_requested_bytes_total",
Help: "Total size of byte ranges required to fetch from the storage before they are extended by the partitioner.",
}),
expandedBytes: promauto.With(reg).NewCounter(prometheus.CounterOpts{
Name: "cortex_bucket_store_partitioner_expanded_bytes_total",
Help: "Total size of byte ranges required to fetch from the storage after they are extended by the partitioner.",
}),
requestedRanges: promauto.With(reg).NewCounter(prometheus.CounterOpts{
Name: "cortex_bucket_store_partitioner_requested_ranges_total",
Help: "Total number of byte ranges required to fetch from the storage before they are passed to the partitioner.",
}),
expandedRanges: promauto.With(reg).NewCounter(prometheus.CounterOpts{
Name: "cortex_bucket_store_partitioner_expanded_ranges_total",
Help: "Total number of byte ranges returned by the partitioner after they've been combined together to reduce the number of bucket API calls.",
}),
extendedRanges: promauto.With(reg).NewCounter(prometheus.CounterOpts{
Name: "cortex_bucket_store_partitioner_extended_ranges_total",
Help: "Total number of byte ranges that were not adjacent or overlapping but were joined because they were closer than the configured maximum gap.",
}),
}
}
// Partition partitions length entries into n <= length ranges that cover all
// input ranges by combining entries that are separated by reasonably small gaps.
// It is used to combine multiple small ranges from object storage into bigger, more efficient/cheaper ones.
func (g *gapBasedPartitioner) Partition(length int, rng func(int) (uint64, uint64)) []Part {
// Run the upstream partitioner to compute the actual ranges that will be fetched.
parts, stats := g.partition(length, rng)
// Calculate the size of ranges that will be fetched.
expandedBytes := uint64(0)
for _, p := range parts {
expandedBytes += p.End - p.Start
}
g.requestedBytes.Add(float64(stats.requestedBytesTotal))
g.expandedBytes.Add(float64(expandedBytes))
g.requestedRanges.Add(float64(length))
g.expandedRanges.Add(float64(len(parts)))
g.extendedRanges.Add(float64(stats.extendedNonOverlappingRanges))
return parts
}
type partitionStats struct {
extendedNonOverlappingRanges int
requestedBytesTotal uint64
}
func (g *gapBasedPartitioner) partition(length int, rng func(int) (uint64, uint64)) (parts []Part, stats partitionStats) {
j := 0
k := 0
for k < length {
j = k
k++
p := Part{}
p.Start, p.End = rng(j)
stats.requestedBytesTotal += p.End - p.Start
// Keep growing the range until the end or we encounter a large gap.
for ; k < length; k++ {
s, e := rng(k)
if p.End >= s {
// The start of the next range overlaps with the current range's end, so we can merge them.
if p.End < e {
// If the next range extends after the current one,
// then we count the extra bytes between the current range's end and the next one's end.
stats.requestedBytesTotal += e - p.End
}
} else if p.End+g.maxGapBytes >= s {
// We can afford to fill a gap between the current range's end and the next range's start.
// We do so, but we also keep track of how much of it we do.
stats.extendedNonOverlappingRanges++
// We count the whole range as a request range since it doesn't overlap with the previous range.
stats.requestedBytesTotal += e - s
} else {
break
}
if p.End <= e {
p.End = e
}
}
p.ElemRng = [2]int{j, k}
parts = append(parts, p)
}
return
}
type uvarintSequenceReader struct {
offset uint64
r io.Reader
b [1]byte
}
var discardBuf [4096]byte
func (r *uvarintSequenceReader) ReadNext(at uint64) (uint64, error) {
if at < r.offset {
panic("cannot reverse reader")
}
for at > r.offset {
b := discardBuf[:]
if uint64(len(b)) > at-r.offset {
b = discardBuf[:at-r.offset]
}
n, err := r.r.Read(b)
r.offset += uint64(n)
if errors.Is(err, io.EOF) {
break
}
if err != nil {
return 0, errors.Wrap(err, "discarding sequence reader")
}
}
return binary.ReadUvarint(r)
}
func (r *uvarintSequenceReader) ReadByte() (byte, error) {
n, err := r.r.Read(r.b[:])
r.offset += uint64(n)
if n == 1 {
// If there are read bytes we must ignore any error returned, even EOF.
return r.b[0], nil
}
return 0, err
}
func (r *uvarintSequenceReader) Read(p []byte) (int, error) {
n, err := r.r.Read(p)
r.offset += uint64(n)
return n, err
}