summaryrefslogtreecommitdiff
path: root/src/mongo/db/operation_context.h
blob: 2c8be360a77ff06af3a96f6acfe99e785307fd4a (plain)
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
/**
 *    Copyright (C) 2014 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.
 */

#pragma once

#include <memory>

#include "mongo/base/disallow_copying.h"
#include "mongo/base/status.h"
#include "mongo/db/client.h"
#include "mongo/db/concurrency/d_concurrency.h"
#include "mongo/db/storage/recovery_unit.h"
#include "mongo/db/storage/storage_options.h"
#include "mongo/db/write_concern_options.h"
#include "mongo/platform/atomic_word.h"
#include "mongo/stdx/condition_variable.h"
#include "mongo/stdx/mutex.h"
#include "mongo/util/decorable.h"
#include "mongo/util/time_support.h"
#include "mongo/util/timer.h"

namespace mongo {

class Client;
class CurOp;
class Locker;
class ProgressMeter;
class ServiceContext;
class StringData;
class WriteUnitOfWork;

/**
 * This class encompasses the state required by an operation and lives from the time a network
 * operation is dispatched until its execution is finished. Note that each "getmore" on a cursor
 * is a separate operation. On construction, an OperationContext associates itself with the
 * current client, and only on destruction it deassociates itself. At any time a client can be
 * associated with at most one OperationContext. Each OperationContext has a RecoveryUnit
 * associated with it, though the lifetime is not necesarily the same, see releaseRecoveryUnit
 * and setRecoveryUnit. The operation context also keeps track of some transaction state
 * (RecoveryUnitState) to reduce complexity and duplication in the storage-engine specific
 * RecoveryUnit and to allow better invariant checking.
 */
class OperationContext : public Decorable<OperationContext> {
    MONGO_DISALLOW_COPYING(OperationContext);

public:
    /**
     * The RecoveryUnitState is used by WriteUnitOfWork to ensure valid state transitions.
     */
    enum RecoveryUnitState {
        kNotInUnitOfWork,   // not in a unit of work, no writes allowed
        kActiveUnitOfWork,  // in a unit of work that still may either commit or abort
        kFailedUnitOfWork   // in a unit of work that has failed and must be aborted
    };

    virtual ~OperationContext() = default;

    /**
     * Interface for durability.  Caller DOES NOT own pointer.
     */
    RecoveryUnit* recoveryUnit() const {
        return _recoveryUnit.get();
    }

    /**
     * Returns the RecoveryUnit (same return value as recoveryUnit()) but the caller takes
     * ownership of the returned RecoveryUnit, and the OperationContext instance relinquishes
     * ownership.  Sets the RecoveryUnit to NULL.
     *
     * Used to transfer ownership of storage engine state from OperationContext
     * to ClientCursor for getMore-able queries.
     *
     * Note that we don't allow the top-level locks to be stored across getMore.
     * We rely on active cursors being killed when collections or databases are dropped,
     * or when collection metadata changes.
     */
    RecoveryUnit* releaseRecoveryUnit();

    /**
     * Associates the OperatingContext with a different RecoveryUnit for getMore or
     * subtransactions, see RecoveryUnitSwap. The new state is passed and the old state is
     * returned separately even though the state logically belongs to the RecoveryUnit,
     * as it is managed by the OperationContext.
     */
    RecoveryUnitState setRecoveryUnit(RecoveryUnit* unit, RecoveryUnitState state);

    /**
     * Interface for locking.  Caller DOES NOT own pointer.
     */
    Locker* lockState() const {
        return _locker.get();
    }

    /**
     * Sets the locker for use by this OperationContext. Call during OperationContext
     * initialization, only.
     */
    void setLockState(std::unique_ptr<Locker> locker);

    /**
     * Releases the locker to the caller. Call during OperationContext cleanup or initialization,
     * only.
     */
    std::unique_ptr<Locker> releaseLockState();

    /**
     * Raises a UserException if this operation is in a killed state.
     */
    void checkForInterrupt();

    /**
     * Returns Status::OK() unless this operation is in a killed state.
     */
    Status checkForInterruptNoAssert();

    /**
     * Waits for either the condition "cv" to be signaled, this operation to be interrupted, or the
     * deadline on this operation to expire.  In the event of interruption or operation deadline
     * expiration, raises a UserException with an error code indicating the interruption type.
     */
    void waitForConditionOrInterrupt(stdx::condition_variable& cv,
                                     stdx::unique_lock<stdx::mutex>& m);

    /**
     * Waits on condition "cv" for "pred" until "pred" returns true, or this operation
     * is interrupted or its deadline expires. Throws a DBException for interruption and
     * deadline expiration.
     */
    template <typename Pred>
    void waitForConditionOrInterrupt(stdx::condition_variable& cv,
                                     stdx::unique_lock<stdx::mutex>& m,
                                     Pred pred) {
        while (!pred()) {
            waitForConditionOrInterrupt(cv, m);
        }
    }

    /**
     * Same as waitForConditionOrInterrupt, except returns a Status instead of throwing
     * a DBException to report interruption.
     */
    Status waitForConditionOrInterruptNoAssert(stdx::condition_variable& cv,
                                               stdx::unique_lock<stdx::mutex>& m) noexcept;

    /**
     * Waits for condition "cv" to be signaled, or for the given "deadline" to expire, or
     * for the operation to be interrupted, or for the operation's own deadline to expire.
     *
     * If the operation deadline expires or the operation is interrupted, throws a DBException.  If
     * the given "deadline" expires, returns cv_status::timeout. Otherwise, returns
     * cv_status::no_timeout.
     */
    stdx::cv_status waitForConditionOrInterruptUntil(stdx::condition_variable& cv,
                                                     stdx::unique_lock<stdx::mutex>& m,
                                                     Date_t deadline);

    /**
     * Waits on condition "cv" for "pred" until "pred" returns true, or the given "deadline"
     * expires, or this operation is interrupted, or this operation's own deadline expires.
     *
     *
     * If the operation deadline expires or the operation is interrupted, throws a DBException.  If
     * the given "deadline" expires, returns cv_status::timeout. Otherwise, returns
     * cv_status::no_timeout indicating that "pred" finally returned true.
     */
    template <typename Pred>
    stdx::cv_status waitForConditionOrInterruptUntil(stdx::condition_variable& cv,
                                                     stdx::unique_lock<stdx::mutex>& m,
                                                     Date_t deadline,
                                                     Pred pred) {
        while (!pred()) {
            if (stdx::cv_status::timeout == waitForConditionOrInterruptUntil(cv, m, deadline)) {
                return stdx::cv_status::timeout;
            }
        }
        return stdx::cv_status::no_timeout;
    }

    /**
     * Same as waitForConditionOrInterruptUntil, except returns StatusWith<stdx::cv_status> and
     * non-ok status indicates the error instead of a DBException.
     */
    StatusWith<stdx::cv_status> waitForConditionOrInterruptNoAssertUntil(
        stdx::condition_variable& cv, stdx::unique_lock<stdx::mutex>& m, Date_t deadline) noexcept;

    /**
     * Delegates to CurOp, but is included here to break dependencies.
     * Caller does not own the pointer.
     *
     * Caller must have locked the "Client" associated with this context.
     */
    virtual ProgressMeter* setMessage_inlock(const char* msg,
                                             const std::string& name = "Progress",
                                             unsigned long long progressMeterTotal = 0,
                                             int secondsBetween = 3) = 0;

    /**
     * Returns the service context under which this operation context runs.
     */
    ServiceContext* getServiceContext() const {
        return _client->getServiceContext();
    }

    /**
     * Returns the client under which this context runs.
     */
    Client* getClient() const {
        return _client;
    }

    /**
     * Returns the operation ID associated with this operation.
     */
    unsigned int getOpID() const {
        return _opId;
    }

    /**
     * Returns WriteConcernOptions of the current operation
     */
    const WriteConcernOptions& getWriteConcern() const {
        return _writeConcern;
    }

    void setWriteConcern(const WriteConcernOptions& writeConcern) {
        _writeConcern = writeConcern;
    }

    /**
     * Set whether or not operations should generate oplog entries.
     * TODO SERVER-26965: Make this private.
     */
    void setReplicatedWrites(bool writesAreReplicated = true) {
        _writesAreReplicated = writesAreReplicated;
    }

    /**
     * Returns true if operations should generate oplog entries.
     */
    bool writesAreReplicated() const {
        return _writesAreReplicated;
    }

    /**
     * Marks this operation as killed so that subsequent calls to checkForInterrupt and
     * checkForInterruptNoAssert by the thread executing the operation will start returning the
     * specified error code.
     *
     * If multiple threads kill the same operation with different codes, only the first code
     * will be preserved.
     *
     * May be called by any thread that has locked the Client owning this operation context, or
     * by the thread executing this on behalf of this OperationContext.
     */
    void markKilled(ErrorCodes::Error killCode = ErrorCodes::Interrupted);

    /**
     * Returns the code passed to markKilled if this operation context has been killed previously
     * or ErrorCodes::OK otherwise.
     *
     * May be called by any thread that has locked the Client owning this operation context, or
     * without lock by the thread executing on behalf of this operation context.
     */
    ErrorCodes::Error getKillStatus() const {
        return _killCode.loadRelaxed();
    }

    /**
     * Shortcut method, which checks whether getKillStatus returns a non-OK value. Has the same
     * concurrency rules as getKillStatus.
     */
    bool isKillPending() const {
        return getKillStatus() != ErrorCodes::OK;
    }

    /**
     * Returns the amount of time since the operation was constructed. Uses the system's most
     * precise tick source, and may not be cheap to call in a tight loop.
     */
    Microseconds getElapsedTime() const {
        return _elapsedTime.elapsed();
    }

    /**
     * Sets the deadline for this operation to the given point in time.
     *
     * To remove a deadline, pass in Date_t::max().
     */
    void setDeadlineByDate(Date_t when);

    /**
     * Sets the deadline for this operation to the maxTime plus the current time reported
     * by the ServiceContext's fast clock source.
     */
    void setDeadlineAfterNowBy(Microseconds maxTime);
    template <typename D>
    void setDeadlineAfterNowBy(D maxTime) {
        if (maxTime <= D::zero()) {
            maxTime = D::zero();
        }
        if (maxTime <= Microseconds::max()) {
            setDeadlineAfterNowBy(duration_cast<Microseconds>(maxTime));
        } else {
            setDeadlineByDate(Date_t::max());
        }
    }

    /**
     * Returns true if this operation has a deadline.
     */
    bool hasDeadline() const {
        return getDeadline() < Date_t::max();
    }

    /**
     * Returns the deadline for this operation, or Date_t::max() if there is no deadline.
     */
    Date_t getDeadline() const {
        return _deadline;
    }

    /**
     * Returns the number of milliseconds remaining for this operation's time limit or
     * Milliseconds::max() if the operation has no time limit.
     */
    Milliseconds getRemainingMaxTimeMillis() const;

    /**
     * NOTE: This is a legacy "max time" method for controlling operation deadlines and it should
     * not be used in new code. Use getRemainingMaxTimeMillis instead.
     *
     * Returns the number of microseconds remaining for this operation's time limit, or the special
     * value Microseconds::max() if the operation has no time limit.
     */
    Microseconds getRemainingMaxTimeMicros() const;

protected:
    OperationContext(Client* client, unsigned int opId);

private:
    /**
     * Returns true if this operation has a deadline and it has passed according to the fast clock
     * on ServiceContext.
     */
    bool hasDeadlineExpired() const;

    /**
     * Sets the deadline and maxTime as described. It is up to the caller to ensure that
     * these correctly correspond.
     */
    void setDeadlineAndMaxTime(Date_t when, Microseconds maxTime);

    friend class WriteUnitOfWork;
    Client* const _client;
    const unsigned int _opId;

    std::unique_ptr<Locker> _locker;

    std::unique_ptr<RecoveryUnit> _recoveryUnit;
    RecoveryUnitState _ruState = kNotInUnitOfWork;

    // Follows the values of ErrorCodes::Error. The default value is 0 (OK), which means the
    // operation is not killed. If killed, it will contain a specific code. This value changes only
    // once from OK to some kill code.
    AtomicWord<ErrorCodes::Error> _killCode{ErrorCodes::OK};


    // If non-null, _waitMutex and _waitCV are the (mutex, condition variable) pair that the
    // operation is currently waiting on inside a call to waitForConditionOrInterrupt...().
    // All access guarded by the Client's lock.
    stdx::mutex* _waitMutex = nullptr;
    stdx::condition_variable* _waitCV = nullptr;

    // If _waitMutex and _waitCV are non-null, this is the number of threads in a call to markKilled
    // actively attempting to kill the operation. If this value is non-zero, the operation is inside
    // waitForConditionOrInterrupt...() and must stay there until _numKillers reaches 0.
    //
    // All access guarded by the Client's lock.
    int _numKillers = 0;

    WriteConcernOptions _writeConcern;

    Date_t _deadline =
        Date_t::max();  // The timepoint at which this operation exceeds its time limit.

    // Max operation time requested by the user or by the cursor in the case of a getMore with no
    // user-specified maxTime. This is tracked with microsecond granularity for the purpose of
    // assigning unused execution time back to a cursor at the end of an operation, only. The
    // _deadline and the service context's fast clock are the only values consulted for determining
    // if the operation's timelimit has been exceeded.
    Microseconds _maxTime = Microseconds::max();

    // Timer counting the elapsed time since the construction of this OperationContext.
    Timer _elapsedTime;

    bool _writesAreReplicated = true;
};

class WriteUnitOfWork {
    MONGO_DISALLOW_COPYING(WriteUnitOfWork);

public:
    WriteUnitOfWork(OperationContext* txn)
        : _txn(txn),
          _committed(false),
          _toplevel(txn->_ruState == OperationContext::kNotInUnitOfWork) {
        uassert(ErrorCodes::IllegalOperation,
                "Cannot execute a write operation in read-only mode",
                !storageGlobalParams.readOnly);
        _txn->lockState()->beginWriteUnitOfWork();
        if (_toplevel) {
            _txn->recoveryUnit()->beginUnitOfWork(_txn);
            _txn->_ruState = OperationContext::kActiveUnitOfWork;
        }
    }

    ~WriteUnitOfWork() {
        dassert(!storageGlobalParams.readOnly);
        if (!_committed) {
            invariant(_txn->_ruState != OperationContext::kNotInUnitOfWork);
            if (_toplevel) {
                _txn->recoveryUnit()->abortUnitOfWork();
                _txn->_ruState = OperationContext::kNotInUnitOfWork;
            } else {
                _txn->_ruState = OperationContext::kFailedUnitOfWork;
            }
            _txn->lockState()->endWriteUnitOfWork();
        }
    }

    void commit() {
        invariant(!_committed);
        invariant(_txn->_ruState == OperationContext::kActiveUnitOfWork);
        if (_toplevel) {
            _txn->recoveryUnit()->commitUnitOfWork();
            _txn->_ruState = OperationContext::kNotInUnitOfWork;
        }
        _txn->lockState()->endWriteUnitOfWork();
        _committed = true;
    }

private:
    OperationContext* const _txn;

    bool _committed;
    bool _toplevel;
};


/**
 * RAII-style class to mark the scope of a transaction. ScopedTransactions may be nested.
 * An outermost ScopedTransaction calls abandonSnapshot() on destruction, so that the storage
 * engine can release resources, such as snapshots or locks, that it may have acquired during
 * the transaction. Note that any writes are committed in nested WriteUnitOfWork scopes,
 * so write conflicts cannot happen on completing a ScopedTransaction.
 *
 * TODO: The ScopedTransaction should hold the global lock
 */
class ScopedTransaction {
    MONGO_DISALLOW_COPYING(ScopedTransaction);

public:
    /**
     * The mode for the transaction indicates whether the transaction will write (MODE_IX) or
     * only read (MODE_IS), or needs to run without other writers (MODE_S) or any other
     * operations (MODE_X) on the server.
     */
    ScopedTransaction(OperationContext* txn, LockMode mode) : _txn(txn) {}

    ~ScopedTransaction() {
        if (!_txn->lockState()->isLocked()) {
            _txn->recoveryUnit()->abandonSnapshot();
        }
    }

private:
    OperationContext* _txn;
};

namespace repl {
/**
 * RAII-style class to turn off replicated writes. Writes do not create oplog entries while the
 * object is in scope.
 */
class UnreplicatedWritesBlock {
    MONGO_DISALLOW_COPYING(UnreplicatedWritesBlock);

public:
    UnreplicatedWritesBlock(OperationContext* txn)
        : _txn(txn), _shouldReplicateWrites(txn->writesAreReplicated()) {
        txn->setReplicatedWrites(false);
    }

    ~UnreplicatedWritesBlock() {
        _txn->setReplicatedWrites(_shouldReplicateWrites);
    }

private:
    OperationContext* _txn;
    const bool _shouldReplicateWrites;
};
}  // namespace repl

}  // namespace mongo