-
Notifications
You must be signed in to change notification settings - Fork 568
/
merge.go
167 lines (151 loc) · 4.55 KB
/
merge.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
package fileset
import (
"bytes"
"context"
"io"
"strings"
"github.com/pachyderm/pachyderm/v2/src/internal/errors"
"github.com/pachyderm/pachyderm/v2/src/internal/storage/chunk"
"github.com/pachyderm/pachyderm/v2/src/internal/storage/fileset/index"
"github.com/pachyderm/pachyderm/v2/src/internal/stream"
)
// MergeReader is an abstraction for reading merged file sets.
type MergeReader struct {
chunks *chunk.Storage
fileSets []FileSet
}
func newMergeReader(chunks *chunk.Storage, fileSets []FileSet) *MergeReader {
return &MergeReader{
chunks: chunks,
fileSets: fileSets,
}
}
// Iterate iterates over the files in the merge reader.
func (mr *MergeReader) Iterate(ctx context.Context, cb func(File) error, opts ...index.Option) error {
var ss []stream.Stream
for i, fs := range mr.fileSets {
// Ignore the base file set's deletive set since it does not affect the state.
if i > 0 {
ss = append(ss, &fileStream{
iterator: NewIterator(ctx, fs.IterateDeletes, opts...),
deletive: true,
})
}
ss = append(ss, &fileStream{
iterator: NewIterator(ctx, fs.Iterate, opts...),
})
}
pq := stream.NewPriorityQueue(ss, compare)
return pq.Iterate(func(ss []stream.Stream) error {
var fss []*fileStream
for _, s := range ss {
fs := s.(*fileStream)
if fs.deletive {
fss = nil
continue
}
fss = append(fss, fs)
}
if len(fss) == 0 {
return nil
}
if len(fss) == 1 {
return cb(newFileReader(mr.chunks, fss[0].file.Index()))
}
var dataRefs []*chunk.DataRef
for _, fs := range fss {
idx := fs.file.Index()
dataRefs = append(dataRefs, idx.File.DataRefs...)
}
mergeIdx := fss[0].file.Index()
mergeIdx.File.DataRefs = dataRefs
return cb(newMergeFileReader(mr.chunks, mergeIdx))
})
}
func (mr *MergeReader) IterateDeletes(ctx context.Context, cb func(File) error, opts ...index.Option) error {
var ss []stream.Stream
for _, fs := range mr.fileSets {
ss = append(ss, &fileStream{
iterator: NewIterator(ctx, fs.IterateDeletes, opts...),
})
}
pq := stream.NewPriorityQueue(ss, compare)
return pq.Iterate(func(ss []stream.Stream) error {
fs := ss[0].(*fileStream)
return cb(newFileReader(mr.chunks, fs.file.Index()))
})
}
// TODO: Look at the sizes?
// TODO: Come up with better heuristics for sharding.
func (mr *MergeReader) Shards(ctx context.Context, opts ...index.Option) ([]*index.PathRange, error) {
shards, err := mr.fileSets[0].Shards(ctx, opts...)
return shards, errors.EnsureStack(err)
}
// MergeFileReader is an abstraction for reading a merged file.
type MergeFileReader struct {
chunks *chunk.Storage
idx *index.Index
}
func newMergeFileReader(chunks *chunk.Storage, idx *index.Index) *MergeFileReader {
return &MergeFileReader{
chunks: chunks,
idx: idx,
}
}
// Index returns the index for the merged file.
// TODO: Removed clone because it had a significant performance impact for small files.
// May want to revisit.
func (mfr *MergeFileReader) Index() *index.Index {
return mfr.idx
}
// Content returns the content of the merged file.
func (mfr *MergeFileReader) Content(ctx context.Context, w io.Writer, opts ...chunk.ReaderOption) error {
r := mfr.chunks.NewReader(ctx, mfr.idx.File.DataRefs, opts...)
return r.Get(w)
}
// Hash returns the hash of the file.
// TODO: It would be good to remove this potential performance footgun, but it would require removing the append functionality.
func (mfr *MergeFileReader) Hash(ctx context.Context) ([]byte, error) {
var hashes [][]byte
size := index.SizeBytes(mfr.idx)
if size >= DefaultBatchThreshold {
uploader := mfr.chunks.NewUploader(ctx, "chunk-uploader-resolver", true, func(_ interface{}, dataRefs []*chunk.DataRef) error {
for _, dataRef := range dataRefs {
hashes = append(hashes, dataRef.Hash)
}
return nil
})
if err := uploader.Copy(nil, mfr.idx.File.DataRefs); err != nil {
return nil, err
}
if err := uploader.Close(); err != nil {
return nil, err
}
} else {
buf := &bytes.Buffer{}
r := mfr.chunks.NewReader(ctx, mfr.idx.File.DataRefs)
if err := r.Get(buf); err != nil {
return nil, err
}
hashes = [][]byte{chunk.Hash(buf.Bytes())}
}
return computeFileHash(hashes)
}
type fileStream struct {
iterator *Iterator
file File
deletive bool
}
func (fs *fileStream) Next() error {
var err error
fs.file, err = fs.iterator.Next()
return err
}
func compare(s1, s2 stream.Stream) int {
idx1 := s1.(*fileStream).file.Index()
idx2 := s2.(*fileStream).file.Index()
if idx1.Path == idx2.Path {
return strings.Compare(idx1.File.Datum, idx2.File.Datum)
}
return strings.Compare(idx1.Path, idx2.Path)
}