forked from cockroachdb/cockroach
-
Notifications
You must be signed in to change notification settings - Fork 0
/
split_at.go
505 lines (450 loc) · 14.2 KB
/
split_at.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
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
// Copyright 2016 The Cockroach 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.
//
// Author: Matt Jibson
package sql
import (
"github.com/pkg/errors"
"golang.org/x/net/context"
"github.com/cockroachdb/cockroach/pkg/gossip"
"github.com/cockroachdb/cockroach/pkg/internal/client"
"github.com/cockroachdb/cockroach/pkg/keys"
"github.com/cockroachdb/cockroach/pkg/roachpb"
"github.com/cockroachdb/cockroach/pkg/sql/parser"
"github.com/cockroachdb/cockroach/pkg/sql/privilege"
"github.com/cockroachdb/cockroach/pkg/sql/sqlbase"
"github.com/cockroachdb/cockroach/pkg/storage"
"github.com/cockroachdb/cockroach/pkg/util/log"
)
// getRowKey generates a key that corresponds to a row (or prefix of a row) in a table or index.
// Both tableDesc and index are required (index can be the primary index).
func getRowKey(
tableDesc *sqlbase.TableDescriptor, index *sqlbase.IndexDescriptor, values []parser.Datum,
) ([]byte, error) {
colMap := make(map[sqlbase.ColumnID]int)
for i := range values {
colMap[index.ColumnIDs[i]] = i
}
prefix := sqlbase.MakeIndexKeyPrefix(tableDesc, index.ID)
key, _, err := sqlbase.EncodePartialIndexKey(
tableDesc, index, len(values), colMap, values, prefix,
)
if err != nil {
return nil, err
}
return keys.MakeRowSentinelKey(key), nil
}
// Split executes a KV split.
// Privileges: INSERT on table.
func (p *planner) Split(ctx context.Context, n *parser.Split) (planNode, error) {
tableDesc, index, err := p.getTableAndIndex(ctx, n.Table, n.Index, privilege.INSERT)
if err != nil {
return nil, err
}
// Calculate the desired types for the select statement. It is OK if the
// select statement returns fewer columns (the relevant prefix is used).
desiredTypes := make([]parser.Type, len(index.ColumnIDs))
for i, colID := range index.ColumnIDs {
c, err := tableDesc.FindColumnByID(colID)
if err != nil {
return nil, err
}
desiredTypes[i] = c.Type.ToDatumType()
}
// Create the plan for the split rows source.
rows, err := p.newPlan(ctx, n.Rows, desiredTypes)
if err != nil {
return nil, err
}
cols := rows.Columns()
if len(cols) == 0 {
return nil, errors.Errorf("no columns in SPLIT AT data")
}
if len(cols) > len(index.ColumnIDs) {
return nil, errors.Errorf("too many columns in SPLIT AT data")
}
for i := range cols {
if !cols[i].Typ.Equivalent(desiredTypes[i]) {
return nil, errors.Errorf(
"SPLIT AT data column %d (%s) must be of type %s, not type %s",
i+1, index.ColumnNames[i], desiredTypes[i], cols[i].Typ,
)
}
}
return &splitNode{
p: p,
tableDesc: tableDesc,
index: index,
rows: rows,
}, nil
}
type splitNode struct {
p *planner
tableDesc *sqlbase.TableDescriptor
index *sqlbase.IndexDescriptor
rows planNode
lastSplitKey []byte
}
func (n *splitNode) Start(ctx context.Context) error {
return n.rows.Start(ctx)
}
func (n *splitNode) Next(ctx context.Context) (bool, error) {
// TODO(radu): instead of performing the splits sequentially, accumulate all
// the split keys and then perform the splits in parallel (e.g. split at the
// middle key and recursively to the left and right).
if ok, err := n.rows.Next(ctx); err != nil || !ok {
return ok, err
}
rowKey, err := getRowKey(n.tableDesc, n.index, n.rows.Values())
if err != nil {
return false, err
}
if err := n.p.session.execCfg.DB.AdminSplit(ctx, rowKey); err != nil {
return false, err
}
n.lastSplitKey = rowKey
return true, nil
}
func (n *splitNode) Values() parser.Datums {
return parser.Datums{
parser.NewDBytes(parser.DBytes(n.lastSplitKey)),
parser.NewDString(keys.PrettyPrint(n.lastSplitKey)),
}
}
func (n *splitNode) Close(ctx context.Context) {
n.rows.Close(ctx)
}
func (*splitNode) Columns() sqlbase.ResultColumns {
return sqlbase.ResultColumns{
{
Name: "key",
Typ: parser.TypeBytes,
},
{
Name: "pretty",
Typ: parser.TypeString,
},
}
}
func (*splitNode) Ordering() orderingInfo { return orderingInfo{} }
func (*splitNode) MarkDebug(_ explainMode) { panic("unimplemented") }
func (*splitNode) DebugValues() debugValues { panic("unimplemented") }
func (*splitNode) Spans(context.Context) (_, _ roachpb.Spans, _ error) {
panic("unimplemented")
}
// Relocate moves ranges to specific stores
// (`ALTER TABLE/INDEX ... TESTING_RELOCATE ...` statement)
// Privileges: INSERT on table.
func (p *planner) Relocate(ctx context.Context, n *parser.Relocate) (planNode, error) {
tableDesc, index, err := p.getTableAndIndex(ctx, n.Table, n.Index, privilege.INSERT)
if err != nil {
return nil, err
}
// Calculate the desired types for the select statement:
// - int array (list of stores)
// - column values; it is OK if the select statement returns fewer columns
// (the relevant prefix is used).
desiredTypes := make([]parser.Type, len(index.ColumnIDs)+1)
desiredTypes[0] = parser.TypeIntArray
for i, colID := range index.ColumnIDs {
c, err := tableDesc.FindColumnByID(colID)
if err != nil {
return nil, err
}
desiredTypes[i+1] = c.Type.ToDatumType()
}
// Create the plan for the split rows source.
rows, err := p.newPlan(ctx, n.Rows, desiredTypes)
if err != nil {
return nil, err
}
cols := rows.Columns()
if len(cols) < 2 {
return nil, errors.Errorf("less than two columns in TESTING_RELOCATE data")
}
if len(cols) > len(index.ColumnIDs)+1 {
return nil, errors.Errorf("too many columns in TESTING_RELOCATE data")
}
for i := range cols {
if !cols[i].Typ.Equivalent(desiredTypes[i]) {
colName := "relocation array"
if i > 0 {
colName = index.ColumnNames[i-1]
}
return nil, errors.Errorf(
"TESTING_RELOCATE data column %d (%s) must be of type %s, not type %s",
i+1, colName, desiredTypes[i], cols[i].Typ,
)
}
}
return &relocateNode{
p: p,
tableDesc: tableDesc,
index: index,
rows: rows,
storeMap: make(map[roachpb.StoreID]roachpb.NodeID),
}, nil
}
type relocateNode struct {
p *planner
tableDesc *sqlbase.TableDescriptor
index *sqlbase.IndexDescriptor
rows planNode
lastRangeStartKey []byte
// storeMap caches information about stores seen in relocation strings (to
// avoid looking them up for every row).
storeMap map[roachpb.StoreID]roachpb.NodeID
}
func (n *relocateNode) Start(ctx context.Context) error {
return n.rows.Start(ctx)
}
func lookupRangeDescriptor(
ctx context.Context, db *client.DB, rowKey []byte,
) (roachpb.RangeDescriptor, error) {
startKey := keys.RangeMetaKey(keys.MustAddr(rowKey))
endKey := keys.Meta2Prefix.PrefixEnd()
kvs, err := db.Scan(ctx, startKey, endKey, 1)
if err != nil {
return roachpb.RangeDescriptor{}, err
}
if len(kvs) != 1 {
log.Fatalf(ctx, "expected 1 KV, got %v", kvs)
}
var desc roachpb.RangeDescriptor
if err := kvs[0].ValueProto(&desc); err != nil {
return roachpb.RangeDescriptor{}, err
}
if desc.EndKey.Equal(rowKey) {
log.Fatalf(ctx, "row key should not be valid range split point: %s", keys.PrettyPrint(rowKey))
}
return desc, nil
}
func (n *relocateNode) Next(ctx context.Context) (bool, error) {
// Each Next call relocates one range (corresponding to one row from n.rows).
// TODO(radu): perform multiple relocations in parallel.
if ok, err := n.rows.Next(ctx); err != nil || !ok {
return ok, err
}
// First column is the relocation string; the rest of the columns indicate the
// table/index row.
data := n.rows.Values()
if !data[0].ResolvedType().Equivalent(parser.TypeIntArray) {
return false, errors.Errorf(
"expected int array in the first TESTING_RELOCATE data column; got %s",
data[0].ResolvedType(),
)
}
relocation := data[0].(*parser.DArray)
if len(relocation.Array) == 0 {
return false, errors.Errorf("empty relocation array for TESTING_RELOCATE")
}
// Create an array of the desired replication targets.
targets := make([]roachpb.ReplicationTarget, len(relocation.Array))
for i, d := range relocation.Array {
storeID := roachpb.StoreID(*d.(*parser.DInt))
nodeID, ok := n.storeMap[storeID]
if !ok {
// Lookup the store in gossip.
var storeDesc roachpb.StoreDescriptor
gossipStoreKey := gossip.MakeStoreKey(storeID)
if err := n.p.session.execCfg.Gossip.GetInfoProto(gossipStoreKey, &storeDesc); err != nil {
return false, errors.Wrapf(err, "error looking up store %d", storeID)
}
nodeID = storeDesc.Node.NodeID
n.storeMap[storeID] = nodeID
}
targets[i] = roachpb.ReplicationTarget{NodeID: nodeID, StoreID: storeID}
}
// Find the current list of replicas. This is inherently racy, so the
// implementation is best effort; in tests, the replication queues should be
// stopped to make this reliable.
rowKey, err := getRowKey(n.tableDesc, n.index, data[1:])
if err != nil {
return false, err
}
rangeDesc, err := lookupRangeDescriptor(ctx, n.p.session.execCfg.DB, rowKey)
if err != nil {
return false, errors.Wrap(err, "error looking up range descriptor")
}
n.lastRangeStartKey = rangeDesc.StartKey.AsRawKey()
if err := storage.RelocateRange(ctx, n.p.ExecCfg().DB, rangeDesc, targets); err != nil {
return false, err
}
return true, nil
}
func (n *relocateNode) Values() parser.Datums {
return parser.Datums{
parser.NewDBytes(parser.DBytes(n.lastRangeStartKey)),
parser.NewDString(keys.PrettyPrint(n.lastRangeStartKey)),
}
}
func (n *relocateNode) Close(ctx context.Context) {
n.rows.Close(ctx)
}
func (*relocateNode) Columns() sqlbase.ResultColumns {
return sqlbase.ResultColumns{
{
Name: "key",
Typ: parser.TypeBytes,
},
{
Name: "pretty",
Typ: parser.TypeString,
},
}
}
func (*relocateNode) Ordering() orderingInfo { return orderingInfo{} }
func (*relocateNode) MarkDebug(_ explainMode) { panic("unimplemented") }
func (*relocateNode) DebugValues() debugValues { panic("unimplemented") }
func (*relocateNode) Spans(context.Context) (_, _ roachpb.Spans, _ error) {
panic("unimplemented")
}
// Scatter moves ranges to random stores
// (`ALTER TABLE/INDEX ... SCATTER ...` statement)
// Privileges: INSERT on table.
func (p *planner) Scatter(ctx context.Context, n *parser.Scatter) (planNode, error) {
tableDesc, index, err := p.getTableAndIndex(ctx, n.Table, n.Index, privilege.INSERT)
if err != nil {
return nil, err
}
var span roachpb.Span
if n.From == nil {
// No FROM/TO specified; the span is the entire table/index.
span = tableDesc.IndexSpan(index.ID)
} else {
switch {
case len(n.From) == 0:
return nil, errors.Errorf("no columns in SCATTER FROM expression")
case len(n.From) > len(index.ColumnIDs):
return nil, errors.Errorf("too many columns in SCATTER FROM expression")
case len(n.To) == 0:
return nil, errors.Errorf("no columns in SCATTER TO expression")
case len(n.To) > len(index.ColumnIDs):
return nil, errors.Errorf("too many columns in SCATTER TO expression")
}
// Calculate the desired types for the select statement:
// - column values; it is OK if the select statement returns fewer columns
// (the relevant prefix is used).
desiredTypes := make([]parser.Type, len(index.ColumnIDs))
for i, colID := range index.ColumnIDs {
c, err := tableDesc.FindColumnByID(colID)
if err != nil {
return nil, err
}
desiredTypes[i] = c.Type.ToDatumType()
}
fromVals := make([]parser.Datum, len(n.From))
for i, expr := range n.From {
typedExpr, err := p.analyzeExpr(
ctx, expr, nil, parser.IndexedVarHelper{}, desiredTypes[i], true, "SCATTER",
)
if err != nil {
return nil, err
}
fromVals[i], err = typedExpr.Eval(&p.evalCtx)
if err != nil {
return nil, err
}
}
toVals := make([]parser.Datum, len(n.From))
for i, expr := range n.To {
typedExpr, err := p.analyzeExpr(
ctx, expr, nil, parser.IndexedVarHelper{}, desiredTypes[i], true, "SCATTER",
)
if err != nil {
return nil, err
}
toVals[i], err = typedExpr.Eval(&p.evalCtx)
if err != nil {
return nil, err
}
}
span.Key, err = getRowKey(tableDesc, index, fromVals)
if err != nil {
return nil, err
}
span.EndKey, err = getRowKey(tableDesc, index, toVals)
if err != nil {
return nil, err
}
// Tolerate reversing FROM and TO; this can be useful for descending
// indexes.
if span.Key.Compare(span.EndKey) > 0 {
span.Key, span.EndKey = span.EndKey, span.Key
}
}
return &scatterNode{
p: p,
span: span,
}, nil
}
type scatterNode struct {
p *planner
span roachpb.Span
rangeIdx int
ranges []roachpb.AdminScatterResponse_Range
}
func (n *scatterNode) Start(ctx context.Context) error {
db := n.p.ExecCfg().DB
req := &roachpb.AdminScatterRequest{
Span: roachpb.Span{Key: n.span.Key, EndKey: n.span.EndKey},
}
res, pErr := client.SendWrapped(ctx, db.GetSender(), req)
if pErr != nil {
return pErr.GoError()
}
n.rangeIdx = -1
n.ranges = res.(*roachpb.AdminScatterResponse).Ranges
return nil
}
func (n *scatterNode) Next(ctx context.Context) (bool, error) {
n.rangeIdx++
hasNext := n.rangeIdx < len(n.ranges)
return hasNext, nil
}
func (*scatterNode) Columns() sqlbase.ResultColumns {
return sqlbase.ResultColumns{
{
Name: "key",
Typ: parser.TypeBytes,
},
{
Name: "pretty",
Typ: parser.TypeString,
},
{
Name: "error",
Typ: parser.TypeString,
},
}
}
func (n *scatterNode) Values() parser.Datums {
r := n.ranges[n.rangeIdx]
dErr := parser.DNull
if r.Error != nil {
dErr = parser.NewDString(r.Error.String())
}
return parser.Datums{
parser.NewDBytes(parser.DBytes(r.Span.Key)),
parser.NewDString(keys.PrettyPrint(r.Span.Key)),
dErr,
}
}
func (*scatterNode) Close(ctx context.Context) {}
func (*scatterNode) Ordering() orderingInfo { return orderingInfo{} }
func (*scatterNode) MarkDebug(_ explainMode) { panic("unimplemented") }
func (*scatterNode) DebugValues() debugValues { panic("unimplemented") }
func (*scatterNode) Spans(context.Context) (_, _ roachpb.Spans, _ error) {
panic("unimplemented")
}