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
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
|
/**
* Copyright (C) 2017 MongoDB Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License, version 3,
* as published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*
* As a special exception, the copyright holders give permission to link the
* code of portions of this program with the OpenSSL library under certain
* conditions as described in each individual source file and distribute
* linked combinations including the program with the OpenSSL library. You
* must comply with the GNU Affero General Public License in all respects for
* all of the code used other than as permitted herein. If you modify file(s)
* with this exception, you may extend this exception to your version of the
* file(s), but you are not obligated to do so. If you do not wish to do so,
* delete this exception statement from your version. If you delete this
* exception statement from all source files in the program, then also delete
* it in the license file.
*/
#define MONGO_LOG_DEFAULT_COMPONENT ::mongo::logger::LogComponent::kIndex
#include "mongo/platform/basic.h"
#include "mongo/db/catalog/index_create_impl.h"
#include "mongo/base/error_codes.h"
#include "mongo/base/init.h"
#include "mongo/client/dbclientinterface.h"
#include "mongo/db/audit.h"
#include "mongo/db/background.h"
#include "mongo/db/catalog/collection.h"
#include "mongo/db/client.h"
#include "mongo/db/clientcursor.h"
#include "mongo/db/concurrency/write_conflict_exception.h"
#include "mongo/db/curop.h"
#include "mongo/db/exec/working_set_common.h"
#include "mongo/db/index/multikey_paths.h"
#include "mongo/db/multi_key_path_tracker.h"
#include "mongo/db/op_observer.h"
#include "mongo/db/operation_context.h"
#include "mongo/db/query/internal_plans.h"
#include "mongo/db/repl/replication_coordinator.h"
#include "mongo/db/server_parameters.h"
#include "mongo/stdx/memory.h"
#include "mongo/stdx/mutex.h"
#include "mongo/util/fail_point.h"
#include "mongo/util/fail_point_service.h"
#include "mongo/util/log.h"
#include "mongo/util/processinfo.h"
#include "mongo/util/progress_meter.h"
#include "mongo/util/quick_exit.h"
#include "mongo/util/scopeguard.h"
namespace mongo {
using std::unique_ptr;
using std::string;
using std::endl;
MONGO_FAIL_POINT_DEFINE(crashAfterStartingIndexBuild);
MONGO_FAIL_POINT_DEFINE(hangAfterStartingIndexBuild);
MONGO_FAIL_POINT_DEFINE(hangAfterStartingIndexBuildUnlocked);
MONGO_FAIL_POINT_DEFINE(slowBackgroundIndexBuild);
AtomicInt32 maxIndexBuildMemoryUsageMegabytes(500);
class ExportedMaxIndexBuildMemoryUsageParameter
: public ExportedServerParameter<std::int32_t, ServerParameterType::kStartupAndRuntime> {
public:
ExportedMaxIndexBuildMemoryUsageParameter()
: ExportedServerParameter<std::int32_t, ServerParameterType::kStartupAndRuntime>(
ServerParameterSet::getGlobal(),
"maxIndexBuildMemoryUsageMegabytes",
&maxIndexBuildMemoryUsageMegabytes) {}
virtual Status validate(const std::int32_t& potentialNewValue) {
if (potentialNewValue < 100) {
return Status(
ErrorCodes::BadValue,
"maxIndexBuildMemoryUsageMegabytes must be greater than or equal to 100 MB");
}
return Status::OK();
}
} exportedMaxIndexBuildMemoryUsageParameter;
/**
* On rollback sets MultiIndexBlockImpl::_needToCleanup to true.
*/
class MultiIndexBlockImpl::SetNeedToCleanupOnRollback : public RecoveryUnit::Change {
public:
explicit SetNeedToCleanupOnRollback(MultiIndexBlockImpl* indexer) : _indexer(indexer) {}
virtual void commit(boost::optional<Timestamp>) {}
virtual void rollback() {
_indexer->_needToCleanup = true;
}
private:
MultiIndexBlockImpl* const _indexer;
};
/**
* On rollback in init(), cleans up _indexes so that ~MultiIndexBlock doesn't try to clean
* up _indexes manually (since the changes were already rolled back).
* Due to this, it is thus legal to call init() again after it fails.
*/
class MultiIndexBlockImpl::CleanupIndexesVectorOnRollback : public RecoveryUnit::Change {
public:
explicit CleanupIndexesVectorOnRollback(MultiIndexBlockImpl* indexer) : _indexer(indexer) {}
virtual void commit(boost::optional<Timestamp>) {}
virtual void rollback() {
_indexer->_indexes.clear();
}
private:
MultiIndexBlockImpl* const _indexer;
};
MultiIndexBlockImpl::MultiIndexBlockImpl(OperationContext* opCtx, Collection* collection)
: _collection(collection),
_opCtx(opCtx),
_buildInBackground(false),
_allowInterruption(false),
_ignoreUnique(false),
_needToCleanup(true) {}
MultiIndexBlockImpl::~MultiIndexBlockImpl() {
if (!_needToCleanup && !_indexes.empty()) {
_collection->infoCache()->clearQueryCache();
}
if (!_needToCleanup || _indexes.empty())
return;
// Make lock acquisition uninterruptible because onOpMessage() can take locks.
UninterruptibleLockGuard noInterrupt(_opCtx->lockState());
while (true) {
try {
WriteUnitOfWork wunit(_opCtx);
// This cleans up all index builds. Because that may need to write, it is done inside of
// a WUOW. Nothing inside this block can fail, and it is made fatal if it does.
for (size_t i = 0; i < _indexes.size(); i++) {
_indexes[i].block->fail();
}
auto replCoord = repl::ReplicationCoordinator::get(_opCtx);
// Nodes building an index on behalf of a user (e.g: `createIndexes`, `applyOps`) may
// fail, removing the existence of the index from the catalog. This update must be
// timestamped. A failure from `createIndexes` should not have a commit timestamp and
// instead write a noop entry. A foreground `applyOps` index build may have a commit
// timestamp already set.
if (_opCtx->recoveryUnit()->getCommitTimestamp().isNull() &&
replCoord->canAcceptWritesForDatabase(_opCtx, "admin")) {
_opCtx->getServiceContext()->getOpObserver()->onOpMessage(
_opCtx,
BSON("msg" << std::string(str::stream() << "Failing index builds. Coll: "
<< _collection->ns().ns())));
}
wunit.commit();
return;
} catch (const WriteConflictException&) {
continue;
} catch (const DBException& e) {
if (e.toStatus() == ErrorCodes::ExceededMemoryLimit)
continue;
error() << "Caught exception while cleaning up partially built indexes: " << redact(e);
} catch (const std::exception& e) {
error() << "Caught exception while cleaning up partially built indexes: " << e.what();
} catch (...) {
error() << "Caught unknown exception while cleaning up partially built indexes.";
}
fassertFailed(18644);
}
}
void MultiIndexBlockImpl::removeExistingIndexes(std::vector<BSONObj>* specs) const {
for (size_t i = 0; i < specs->size(); i++) {
Status status =
_collection->getIndexCatalog()->prepareSpecForCreate(_opCtx, (*specs)[i]).getStatus();
if (status.code() == ErrorCodes::IndexAlreadyExists) {
specs->erase(specs->begin() + i);
i--;
}
// intentionally ignoring other error codes
}
}
StatusWith<std::vector<BSONObj>> MultiIndexBlockImpl::init(const BSONObj& spec) {
const auto indexes = std::vector<BSONObj>(1, spec);
return init(indexes);
}
StatusWith<std::vector<BSONObj>> MultiIndexBlockImpl::init(const std::vector<BSONObj>& indexSpecs) {
WriteUnitOfWork wunit(_opCtx);
invariant(_indexes.empty());
_opCtx->recoveryUnit()->registerChange(new CleanupIndexesVectorOnRollback(this));
const string& ns = _collection->ns().ns();
const auto idxCat = _collection->getIndexCatalog();
invariant(idxCat);
invariant(idxCat->ok());
Status status = idxCat->checkUnfinished();
if (!status.isOK())
return status;
for (size_t i = 0; i < indexSpecs.size(); i++) {
BSONObj info = indexSpecs[i];
string pluginName = IndexNames::findPluginName(info["key"].Obj());
if (pluginName.size()) {
Status s = _collection->getIndexCatalog()->_upgradeDatabaseMinorVersionIfNeeded(
_opCtx, pluginName);
if (!s.isOK())
return s;
}
// Any foreground indexes make all indexes be built in the foreground.
_buildInBackground = (_buildInBackground && initBackgroundIndexFromSpec(info));
}
std::vector<BSONObj> indexInfoObjs;
indexInfoObjs.reserve(indexSpecs.size());
std::size_t eachIndexBuildMaxMemoryUsageBytes = 0;
if (!indexSpecs.empty()) {
eachIndexBuildMaxMemoryUsageBytes =
static_cast<std::size_t>(maxIndexBuildMemoryUsageMegabytes.load()) * 1024 * 1024 /
indexSpecs.size();
}
for (size_t i = 0; i < indexSpecs.size(); i++) {
BSONObj info = indexSpecs[i];
StatusWith<BSONObj> statusWithInfo =
_collection->getIndexCatalog()->prepareSpecForCreate(_opCtx, info);
Status status = statusWithInfo.getStatus();
if (!status.isOK())
return status;
info = statusWithInfo.getValue();
indexInfoObjs.push_back(info);
IndexToBuild index;
index.block.reset(new IndexCatalogImpl::IndexBuildBlock(_opCtx, _collection, info));
status = index.block->init();
if (!status.isOK())
return status;
index.real = index.block->getEntry()->accessMethod();
status = index.real->initializeAsEmpty(_opCtx);
if (!status.isOK())
return status;
if (!_buildInBackground) {
// Bulk build process requires foreground building as it assumes nothing is changing
// under it.
index.bulk = index.real->initiateBulk(eachIndexBuildMaxMemoryUsageBytes);
}
const IndexDescriptor* descriptor = index.block->getEntry()->descriptor();
IndexCatalog::prepareInsertDeleteOptions(_opCtx, descriptor, &index.options);
index.options.dupsAllowed = index.options.dupsAllowed || _ignoreUnique;
if (_ignoreUnique) {
index.options.getKeysMode = IndexAccessMethod::GetKeysMode::kRelaxConstraints;
}
log() << "build index on: " << ns << " properties: " << descriptor->toString();
if (index.bulk)
log() << "\t building index using bulk method; build may temporarily use up to "
<< eachIndexBuildMaxMemoryUsageBytes / 1024 / 1024 << " megabytes of RAM";
index.filterExpression = index.block->getEntry()->getFilterExpression();
// TODO SERVER-14888 Suppress this in cases we don't want to audit.
audit::logCreateIndex(_opCtx->getClient(), &info, descriptor->indexName(), ns);
_indexes.push_back(std::move(index));
}
if (_buildInBackground)
_backgroundOperation.reset(new BackgroundOperation(ns));
auto replCoord = repl::ReplicationCoordinator::get(_opCtx);
if (_opCtx->recoveryUnit()->getCommitTimestamp().isNull() &&
replCoord->canAcceptWritesForDatabase(_opCtx, "admin")) {
// Only primaries must timestamp this write. Secondaries run this from within a
// `TimestampBlock`. Primaries performing an index build via `applyOps` may have a
// wrapping commit timestamp that will be used instead.
_opCtx->getServiceContext()->getOpObserver()->onOpMessage(
_opCtx, BSON("msg" << std::string(str::stream() << "Creating indexes. Coll: " << ns)));
}
wunit.commit();
if (MONGO_FAIL_POINT(crashAfterStartingIndexBuild)) {
log() << "Index build interrupted due to 'crashAfterStartingIndexBuild' failpoint. Exiting "
"after waiting for changes to become durable.";
Locker::LockSnapshot lockInfo;
_opCtx->lockState()->saveLockStateAndUnlock(&lockInfo);
if (_opCtx->recoveryUnit()->waitUntilDurable()) {
quickExit(EXIT_TEST);
}
}
return indexInfoObjs;
}
Status MultiIndexBlockImpl::insertAllDocumentsInCollection(std::set<RecordId>* dupsOut) {
invariant(!_opCtx->lockState()->inAWriteUnitOfWork());
// Refrain from persisting any multikey updates as a result from building the index. Instead,
// accumulate them in the `MultikeyPathTracker` and do the write as part of the update that
// commits the index.
auto stopTracker =
MakeGuard([this] { MultikeyPathTracker::get(_opCtx).stopTrackingMultikeyPathInfo(); });
if (MultikeyPathTracker::get(_opCtx).isTrackingMultikeyPathInfo()) {
stopTracker.Dismiss();
}
MultikeyPathTracker::get(_opCtx).startTrackingMultikeyPathInfo();
const char* curopMessage = _buildInBackground ? "Index Build (background)" : "Index Build";
const auto numRecords = _collection->numRecords(_opCtx);
stdx::unique_lock<Client> lk(*_opCtx->getClient());
ProgressMeterHolder progress(
CurOp::get(_opCtx)->setMessage_inlock(curopMessage, curopMessage, numRecords));
lk.unlock();
Timer t;
unsigned long long n = 0;
PlanExecutor::YieldPolicy yieldPolicy;
if (_buildInBackground) {
invariant(_allowInterruption);
yieldPolicy = PlanExecutor::YIELD_AUTO;
} else {
yieldPolicy = PlanExecutor::WRITE_CONFLICT_RETRY_ONLY;
}
auto exec =
InternalPlanner::collectionScan(_opCtx, _collection->ns().ns(), _collection, yieldPolicy);
Snapshotted<BSONObj> objToIndex;
RecordId loc;
PlanExecutor::ExecState state;
int retries = 0; // non-zero when retrying our last document.
while (retries ||
(PlanExecutor::ADVANCED == (state = exec->getNextSnapshotted(&objToIndex, &loc))) ||
MONGO_FAIL_POINT(hangAfterStartingIndexBuild)) {
try {
if (_allowInterruption)
_opCtx->checkForInterrupt();
if (!(retries || PlanExecutor::ADVANCED == state) ||
MONGO_FAIL_POINT(slowBackgroundIndexBuild)) {
log() << "Hanging index build due to failpoint";
invariant(_allowInterruption);
sleepmillis(1000);
continue;
}
// Make sure we are working with the latest version of the document.
if (objToIndex.snapshotId() != _opCtx->recoveryUnit()->getSnapshotId() &&
!_collection->findDoc(_opCtx, loc, &objToIndex)) {
// doc was deleted so don't index it.
retries = 0;
continue;
}
// Done before insert so we can retry document if it WCEs.
progress->setTotalWhileRunning(_collection->numRecords(_opCtx));
WriteUnitOfWork wunit(_opCtx);
Status ret = insert(objToIndex.value(), loc);
if (_buildInBackground)
exec->saveState();
if (ret.isOK()) {
wunit.commit();
} else if (dupsOut && ret.code() == ErrorCodes::DuplicateKey) {
// If dupsOut is non-null, we should only fail the specific insert that
// led to a DuplicateKey rather than the whole index build.
dupsOut->insert(loc);
} else {
// Fail the index build hard.
return ret;
}
if (_buildInBackground) {
auto restoreStatus = exec->restoreState(); // Handles any WCEs internally.
if (!restoreStatus.isOK()) {
return restoreStatus;
}
}
// Go to the next document
progress->hit();
n++;
retries = 0;
} catch (const WriteConflictException&) {
CurOp::get(_opCtx)->debug().writeConflicts++;
retries++; // logAndBackoff expects this to be 1 on first call.
WriteConflictException::logAndBackoff(
retries, "index creation", _collection->ns().ns());
// Can't use writeConflictRetry since we need to save/restore exec around call to
// abandonSnapshot.
exec->saveState();
_opCtx->recoveryUnit()->abandonSnapshot();
auto restoreStatus = exec->restoreState(); // Handles any WCEs internally.
if (!restoreStatus.isOK()) {
return restoreStatus;
}
}
}
if (state != PlanExecutor::IS_EOF) {
return WorkingSetCommon::getMemberObjectStatus(objToIndex.value());
}
if (MONGO_FAIL_POINT(hangAfterStartingIndexBuildUnlocked)) {
// Unlock before hanging so replication recognizes we've completed.
Locker::LockSnapshot lockInfo;
_opCtx->lockState()->saveLockStateAndUnlock(&lockInfo);
while (MONGO_FAIL_POINT(hangAfterStartingIndexBuildUnlocked)) {
log() << "Hanging index build with no locks due to "
"'hangAfterStartingIndexBuildUnlocked' failpoint";
sleepmillis(1000);
}
if (_buildInBackground) {
_opCtx->lockState()->restoreLockState(_opCtx, lockInfo);
_opCtx->recoveryUnit()->abandonSnapshot();
return Status(ErrorCodes::OperationFailed,
"background index build aborted due to failpoint");
} else {
invariant(
!"the hangAfterStartingIndexBuildUnlocked failpoint can't be turned off for foreground index builds");
}
}
progress->finished();
Status ret = doneInserting(dupsOut);
if (!ret.isOK())
return ret;
log() << "build index done. scanned " << n << " total records. " << t.seconds() << " secs";
return Status::OK();
}
Status MultiIndexBlockImpl::insert(const BSONObj& doc, const RecordId& loc) {
for (size_t i = 0; i < _indexes.size(); i++) {
if (_indexes[i].filterExpression && !_indexes[i].filterExpression->matchesBSON(doc)) {
continue;
}
int64_t unused;
Status idxStatus(ErrorCodes::InternalError, "");
if (_indexes[i].bulk) {
idxStatus = _indexes[i].bulk->insert(_opCtx, doc, loc, _indexes[i].options, &unused);
} else {
idxStatus = _indexes[i].real->insert(_opCtx, doc, loc, _indexes[i].options, &unused);
}
if (!idxStatus.isOK())
return idxStatus;
}
return Status::OK();
}
Status MultiIndexBlockImpl::doneInserting(std::set<RecordId>* dupsOut) {
invariant(!_opCtx->lockState()->inAWriteUnitOfWork());
for (size_t i = 0; i < _indexes.size(); i++) {
if (_indexes[i].bulk == NULL)
continue;
LOG(1) << "\t bulk commit starting for index: "
<< _indexes[i].block->getEntry()->descriptor()->indexName();
Status status = _indexes[i].real->commitBulk(_opCtx,
_indexes[i].bulk.get(),
_allowInterruption,
_indexes[i].options.dupsAllowed,
dupsOut);
if (!status.isOK()) {
return status;
}
}
return Status::OK();
}
void MultiIndexBlockImpl::abortWithoutCleanup() {
_indexes.clear();
_needToCleanup = false;
}
void MultiIndexBlockImpl::commit(stdx::function<void(const BSONObj& spec)> onCreateFn) {
// Do not interfere with writing multikey information when committing index builds.
auto restartTracker =
MakeGuard([this] { MultikeyPathTracker::get(_opCtx).startTrackingMultikeyPathInfo(); });
if (!MultikeyPathTracker::get(_opCtx).isTrackingMultikeyPathInfo()) {
restartTracker.Dismiss();
}
MultikeyPathTracker::get(_opCtx).stopTrackingMultikeyPathInfo();
for (size_t i = 0; i < _indexes.size(); i++) {
if (onCreateFn) {
onCreateFn(_indexes[i].block->getSpec());
}
_indexes[i].block->success();
// The bulk builder will track multikey information itself. Non-bulk builders re-use the
// code path that a typical insert/update uses. State is altered on the non-bulk build
// path to accumulate the multikey information on the `MultikeyPathTracker`.
if (_indexes[i].bulk) {
const auto& bulkBuilder = _indexes[i].bulk;
if (bulkBuilder->isMultikey()) {
_indexes[i].block->getEntry()->setMultikey(_opCtx, bulkBuilder->getMultikeyPaths());
}
} else {
auto multikeyPaths =
boost::optional<MultikeyPaths>(MultikeyPathTracker::get(_opCtx).getMultikeyPathInfo(
_collection->ns(), _indexes[i].block->getIndexName()));
if (multikeyPaths) {
_indexes[i].block->getEntry()->setMultikey(_opCtx, *multikeyPaths);
}
}
}
_opCtx->recoveryUnit()->registerChange(new SetNeedToCleanupOnRollback(this));
_needToCleanup = false;
}
} // namespace mongo
|