-
Notifications
You must be signed in to change notification settings - Fork 3.8k
/
index_skip_table_reader.go
280 lines (240 loc) · 7.88 KB
/
index_skip_table_reader.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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
// Copyright 2019 The Cockroach Authors.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
package distsqlrun
import (
"context"
"sync"
"github.com/cockroachdb/cockroach/pkg/roachpb"
"github.com/cockroachdb/cockroach/pkg/sql/distsqlpb"
"github.com/cockroachdb/cockroach/pkg/sql/row"
"github.com/cockroachdb/cockroach/pkg/sql/sqlbase"
"github.com/pkg/errors"
)
// indexSkipTableReader is a processor that retrieves distinct rows from
// a table using the prefix of an index to skip reading some rows in the
// table. Specifically, given a prefix of an index to distinct over,
// the indexSkipTableReader returns all distinct rows where that prefix
// of the index is distinct. It uses the index to seek to distinct values
// of the prefix instead of doing a full table scan.
type indexSkipTableReader struct {
ProcessorBase
spans roachpb.Spans
// currentSpan maintains which span we are currently scanning.
currentSpan int
// keyPrefixLen holds the length of the prefix of the index
// that we are performing a distinct over.
keyPrefixLen int
// indexLen holds the number of columns in the index that
// is being considered.
indexLen int
reverse bool
ignoreMisplannedRanges bool
misplannedRanges []roachpb.RangeInfo
fetcher row.Fetcher
alloc sqlbase.DatumAlloc
}
const indexSkipTableReaderProcName = "index skip table reader"
var istrPool = sync.Pool{
New: func() interface{} {
return &indexSkipTableReader{}
},
}
var _ Processor = &indexSkipTableReader{}
var _ RowSource = &indexSkipTableReader{}
var _ distsqlpb.MetadataSource = &indexSkipTableReader{}
func newIndexSkipTableReader(
flowCtx *FlowCtx,
processorID int32,
spec *distsqlpb.IndexSkipTableReaderSpec,
post *distsqlpb.PostProcessSpec,
output RowReceiver,
) (*indexSkipTableReader, error) {
if flowCtx.NodeID == 0 {
return nil, errors.Errorf("attempting to create a tableReader with uninitialized NodeID")
}
t := istrPool.Get().(*indexSkipTableReader)
returnMutations := spec.Visibility == distsqlpb.ScanVisibility_PUBLIC_AND_NOT_PUBLIC
types := spec.Table.ColumnTypesWithMutations(returnMutations)
t.ignoreMisplannedRanges = flowCtx.local
t.reverse = spec.Reverse
if err := t.Init(
t,
post,
types,
flowCtx,
processorID,
output,
nil, /* memMonitor */
ProcStateOpts{
InputsToDrain: nil,
TrailingMetaCallback: t.generateTrailingMeta,
},
); err != nil {
return nil, err
}
neededColumns := t.out.neededColumns()
t.keyPrefixLen = neededColumns.Len()
columnIdxMap := spec.Table.ColumnIdxMapWithMutations(returnMutations)
immutDesc := sqlbase.NewImmutableTableDescriptor(spec.Table)
index, isSecondaryIndex, err := immutDesc.FindIndexByIndexIdx(int(spec.IndexIdx))
if err != nil {
return nil, err
}
t.indexLen = len(index.ColumnIDs)
cols := immutDesc.Columns
if returnMutations {
cols = immutDesc.ReadableColumns
}
tableArgs := row.FetcherTableArgs{
Desc: immutDesc,
Index: index,
ColIdxMap: columnIdxMap,
IsSecondaryIndex: isSecondaryIndex,
Cols: cols,
ValNeededForCol: neededColumns,
}
if err := t.fetcher.Init(t.reverse, true, /* returnRangeInfo */
false /* isCheck */, &t.alloc, tableArgs); err != nil {
return nil, err
}
// Make a copy of the spans for this reader, as we will modify them.
nSpans := len(spec.Spans)
if cap(t.spans) >= nSpans {
t.spans = t.spans[:nSpans]
} else {
t.spans = make(roachpb.Spans, nSpans)
}
// If we are scanning in reverse, then copy the spans in backwards.
if t.reverse {
for i, s := range spec.Spans {
t.spans[len(spec.Spans)-i-1] = s.Span
}
} else {
for i, s := range spec.Spans {
t.spans[i] = s.Span
}
}
return t, nil
}
func (t *indexSkipTableReader) Start(ctx context.Context) context.Context {
t.StartInternal(ctx, indexSkipTableReaderProcName)
return ctx
}
func (t *indexSkipTableReader) Next() (sqlbase.EncDatumRow, *distsqlpb.ProducerMetadata) {
for t.State == StateRunning {
if t.currentSpan >= len(t.spans) {
t.MoveToDraining(nil)
return nil, t.DrainHelper()
}
// Start a scan to get the smallest value within this span.
err := t.fetcher.StartScan(
t.Ctx, t.flowCtx.txn, t.spans[t.currentSpan:t.currentSpan+1],
true, 1 /* batch size limit */, t.flowCtx.traceKV,
)
if err != nil {
t.MoveToDraining(err)
return nil, &distsqlpb.ProducerMetadata{Err: err}
}
// Range info resets once a scan begins, so we need to maintain
// the range info we get after each scan.
if !t.ignoreMisplannedRanges {
ranges := misplannedRanges(t.Ctx, t.fetcher.GetRangesInfo(), t.flowCtx.NodeID)
for _, r := range ranges {
t.misplannedRanges = roachpb.InsertRangeInfo(t.misplannedRanges, r)
}
}
var key roachpb.Key
{
// Fetch the PartialKey in a separate scope so that it cannot be reused
// outside this scope.
unsafeKey, err := t.fetcher.PartialKey(t.keyPrefixLen)
if err != nil {
t.MoveToDraining(err)
return nil, &distsqlpb.ProducerMetadata{Err: err}
}
// Modifying unsafeKey in place would corrupt our current row, so make a
// copy.
// A key needs to be allocated every time because it is illegal to mutate
// batch requests which would end up happening if we used scratch space.
// TODO(asubiotto): Another option would be to construct a key after
// returning a row (at the top of this loop).
key = make([]byte, len(unsafeKey))
copy(key, unsafeKey)
}
row, _, _, err := t.fetcher.NextRow(t.Ctx)
if err != nil {
t.MoveToDraining(err)
return nil, &distsqlpb.ProducerMetadata{Err: err}
}
if row == nil {
// No more rows in this span, so move to the next one.
t.currentSpan++
continue
}
if !t.reverse {
// We set the new key to be the largest key with the prefix that we have
// so that we skip all values with the same prefix, and "skip" to the
// next distinct value.
t.spans[t.currentSpan].Key = key.PrefixEnd()
} else {
// In the case of reverse, this is much easier. The reverse fetcher
// returns the key retrieved, in this case the first key smaller
// than EndKey in the current span. Since EndKey is exclusive, we
// just set the retrieved key as EndKey for the next scan.
t.spans[t.currentSpan].EndKey = key
}
// If the changes we made turned our current span invalid, mark that
// we should move on to the next span before returning the row.
if !t.spans[t.currentSpan].Valid() {
t.currentSpan++
}
if outRow := t.ProcessRowHelper(row); outRow != nil {
return outRow, nil
}
}
return nil, t.DrainHelper()
}
func (t *indexSkipTableReader) Release() {
t.ProcessorBase.Reset()
t.fetcher.Reset()
*t = indexSkipTableReader{
ProcessorBase: t.ProcessorBase,
fetcher: t.fetcher,
spans: t.spans[:0],
misplannedRanges: t.misplannedRanges[:0],
currentSpan: 0,
}
istrPool.Put(t)
}
func (t *indexSkipTableReader) ConsumerClosed() {
t.InternalClose()
}
func (t *indexSkipTableReader) generateTrailingMeta(
ctx context.Context,
) []distsqlpb.ProducerMetadata {
trailingMeta := t.generateMeta(ctx)
t.InternalClose()
return trailingMeta
}
func (t *indexSkipTableReader) generateMeta(ctx context.Context) []distsqlpb.ProducerMetadata {
var trailingMeta []distsqlpb.ProducerMetadata
if !t.ignoreMisplannedRanges {
if len(t.misplannedRanges) != 0 {
trailingMeta = append(trailingMeta, distsqlpb.ProducerMetadata{Ranges: t.misplannedRanges})
}
}
if meta := getTxnCoordMeta(ctx, t.flowCtx.txn); meta != nil {
trailingMeta = append(trailingMeta, distsqlpb.ProducerMetadata{TxnCoordMeta: meta})
}
return trailingMeta
}
func (t *indexSkipTableReader) DrainMeta(ctx context.Context) []distsqlpb.ProducerMetadata {
return t.generateMeta(ctx)
}