forked from firedancer-io/radiance
-
Notifications
You must be signed in to change notification settings - Fork 0
/
query.go
186 lines (164 loc) · 4.71 KB
/
query.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
183
184
185
186
package compactindex
import (
"errors"
"fmt"
"io"
)
// DB is a compactindex handle.
type DB struct {
Header
Stream io.ReaderAt
}
// Open returns a handle to access a compactindex.
//
// The provided stream must start with the Magic byte sequence.
// Tip: Use io.NewSectionReader to create aligned substreams when dealing with a file that contains multiple indexes.
func Open(stream io.ReaderAt) (*DB, error) {
// Read the static 32-byte header.
// Ignore errors if the read fails after filling the buffer (e.g. EOF).
var fileHeader [headerSize]byte
n, readErr := stream.ReadAt(fileHeader[:], 0)
if n < len(fileHeader) {
// ReadAt must return non-nil error here.
return nil, readErr
}
db := new(DB)
if err := db.Header.Load(&fileHeader); err != nil {
return nil, err
}
db.Stream = stream
return db, nil
}
// Lookup queries for a key in the index and returns the value (offset), if any.
//
// Returns ErrNotFound if the key is unknown.
func (db *DB) Lookup(key []byte) (uint64, error) {
bucket, err := db.LookupBucket(key)
if err != nil {
return 0, err
}
return bucket.Lookup(key)
}
// LookupBucket returns a handle to the bucket that might contain the given key.
func (db *DB) LookupBucket(key []byte) (*Bucket, error) {
return db.GetBucket(db.Header.BucketHash(key))
}
// GetBucket returns a handle to the bucket at the given index.
func (db *DB) GetBucket(i uint) (*Bucket, error) {
if i >= uint(db.Header.NumBuckets) {
return nil, fmt.Errorf("out of bounds bucket index: %d >= %d", i, db.Header.NumBuckets)
}
// Fill bucket handle.
bucket := &Bucket{
BucketDescriptor: BucketDescriptor{
Stride: db.entryStride(),
OffsetWidth: intWidth(db.FileSize),
},
}
// Read bucket header.
readErr := bucket.BucketHeader.readFrom(db.Stream, i)
if readErr != nil {
return nil, readErr
}
bucket.Entries = io.NewSectionReader(db.Stream, int64(bucket.FileOffset), int64(bucket.NumEntries)*int64(bucket.Stride))
return bucket, nil
}
func (db *DB) entryStride() uint8 {
hashSize := 3 // TODO remove hardcoded constant
offsetSize := intWidth(db.FileSize)
return uint8(hashSize) + offsetSize
}
func bucketOffset(i uint) int64 {
return headerSize + int64(i)*bucketHdrLen
}
func (b *BucketHeader) readFrom(rd io.ReaderAt, i uint) error {
var buf [bucketHdrLen]byte
n, err := rd.ReadAt(buf[:], bucketOffset(i))
if n < len(buf) {
return err
}
b.Load(&buf)
return nil
}
func (b *BucketHeader) writeTo(wr io.WriterAt, i uint) error {
var buf [bucketHdrLen]byte
b.Store(&buf)
_, err := wr.WriteAt(buf[:], bucketOffset(i))
return err
}
// Bucket is a database handle pointing to a subset of the index.
type Bucket struct {
BucketDescriptor
Entries *io.SectionReader
}
// maxEntriesPerBucket is the hardcoded maximum permitted number of entries per bucket.
const maxEntriesPerBucket = 1 << 24 // (16 * stride) MiB
// targetEntriesPerBucket is the average number of records in each hashtable bucket we aim for.
const targetEntriesPerBucket = 10000
// Load retrieves all entries in the hashtable.
func (b *Bucket) Load(batchSize int) ([]Entry, error) {
if batchSize <= 0 {
batchSize = 512 // default to reasonable batch size
}
// TODO bounds check
if b.NumEntries > maxEntriesPerBucket {
return nil, fmt.Errorf("refusing to load bucket with %d entries", b.NumEntries)
}
entries := make([]Entry, 0, b.NumEntries)
stride := int(b.Stride)
buf := make([]byte, batchSize*stride)
off := int64(0)
for {
// Read another chunk.
n, err := b.Entries.ReadAt(buf, off)
// Decode all entries in it.
sub := buf[:n]
for len(sub) >= stride {
entries = append(entries, b.unmarshalEntry(sub))
sub = sub[stride:]
off += int64(stride)
}
// Handle error.
if errors.Is(err, io.EOF) || errors.Is(err, io.ErrUnexpectedEOF) {
break
} else if err != nil {
return nil, err
}
}
return entries, nil
}
// TODO: This binary search algo is not optimized for high-latency remotes yet.
// Lookup queries for a key using binary search.
func (b *Bucket) Lookup(key []byte) (uint64, error) {
return b.binarySearch(b.Hash(key))
}
func (b *Bucket) binarySearch(target uint64) (uint64, error) {
low := 0
high := int(b.NumEntries)
for low <= high {
median := (low + high) / 2
entry, err := b.loadEntry(median)
if err != nil {
return 0, err
}
if entry.Hash == target {
return entry.Value, nil
} else if entry.Hash < target {
low = median + 1
} else {
high = median - 1
}
}
return 0, ErrNotFound
}
func (b *Bucket) loadEntry(i int) (Entry, error) {
off := int64(i) * int64(b.Stride)
buf := make([]byte, b.Stride)
n, err := b.Entries.ReadAt(buf, off)
if n != len(buf) {
return Entry{}, err
}
return b.unmarshalEntry(buf), nil
}
// ErrNotFound marks a missing entry.
var ErrNotFound = errors.New("not found")