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
|
/**
* 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 <climits> // For UINT_MAX
#include <vector>
#include "mongo/db/concurrency/lock_manager.h"
#include "mongo/db/concurrency/lock_stats.h"
#include "mongo/stdx/thread.h"
namespace mongo {
/**
* Interface for acquiring locks. One of those objects will have to be instantiated for each
* request (transaction).
*
* Lock/unlock methods must always be called from a single thread.
*/
class Locker {
MONGO_DISALLOW_COPYING(Locker);
public:
virtual ~Locker() {}
/**
* Returns true if this is an instance of LockerNoop. Because LockerNoop doesn't implement many
* methods, some users may need to check this first to find out what is safe to call. LockerNoop
* is only used in unittests and for a brief period at startup, so you can assume you hold the
* equivalent of a MODE_X lock when using it.
*
* TODO get rid of this once we kill LockerNoop.
*/
virtual bool isNoop() const {
return false;
}
/**
* Require global lock attempts to obtain tickets from 'reading' (for MODE_S and MODE_IS),
* and from 'writing' (for MODE_IX), which must have static lifetimes. There is no throttling
* for MODE_X, as there can only ever be a single locker using this mode. The throttling is
* intended to defend against arge drops in throughput under high load due to too much
* concurrency.
*/
static void setGlobalThrottling(class TicketHolder* reading, class TicketHolder* writing);
/**
* State for reporting the number of active and queued reader and writer clients.
*/
enum ClientState { kInactive, kActiveReader, kActiveWriter, kQueuedReader, kQueuedWriter };
/**
* Return whether client is holding any locks (active), or is queued on any locks or waiting
* for a ticket (throttled).
*/
virtual ClientState getClientState() const = 0;
virtual LockerId getId() const = 0;
/**
* Get a platform-specific thread identifier of the thread which owns the this locker for
* tracing purposes.
*/
virtual stdx::thread::id getThreadId() const = 0;
/**
* This should be the first method invoked for a particular Locker object. It acquires the
* Global lock in the specified mode and effectively indicates the mode of the operation.
* This is what the lock modes on the global lock mean:
*
* IX - Regular write operation
* IS - Regular read operation
* S - Stops all *write* activity. Used for administrative operations (repl, etc).
* X - Stops all activity. Used for administrative operations (repl state changes,
* shutdown, etc).
*
* This method can be called recursively, but each call to lockGlobal must be accompanied
* by a call to unlockGlobal.
*
* @param mode Mode in which the global lock should be acquired. Also indicates the intent
* of the operation.
*
* @return LOCK_OK, if the global lock (and the flush lock, for the MMAP V1 engine) were
* acquired within the specified time bound. Otherwise, the respective failure
* code and neither lock will be acquired.
*/
virtual LockResult lockGlobal(LockMode mode) = 0;
/**
* Requests the global lock to be acquired in the specified mode.
*
* See the comments for lockBegin/Complete for more information on the semantics.
* The timeout indicates how long to wait for the lock to be acquired. The lockGlobalBegin
* method has a timeout for use with the TicketHolder, if there is one.
*/
virtual LockResult lockGlobalBegin(LockMode mode, Milliseconds timeout) = 0;
virtual LockResult lockGlobalComplete(Milliseconds timeout) = 0;
/**
* This method is used only in the MMAP V1 storage engine, otherwise it is a no-op. See the
* comments in the implementation for more details on how MMAP V1 journaling works.
*/
virtual void lockMMAPV1Flush() = 0;
/**
* Decrements the reference count on the global lock. If the reference count on the
* global lock hits zero, the transaction is over, and unlockGlobal unlocks all other locks
* except for RESOURCE_MUTEX locks.
*
* @return true if this is the last endTransaction call (i.e., the global lock was
* released); false if there are still references on the global lock. This value
* should not be relied on and is only used for assertion purposes.
*
* @return false if the global lock is still held.
*/
virtual bool unlockGlobal() = 0;
/**
* This is only necessary for the MMAP V1 engine and in particular, the fsyncLock command
* which needs to first acquire the global lock in X-mode for truncating the journal and
* then downgrade to S before it blocks.
*
* The downgrade is necessary in order to be nice and not block readers while under
* fsyncLock.
*/
virtual void downgradeGlobalXtoSForMMAPV1() = 0;
/**
* beginWriteUnitOfWork/endWriteUnitOfWork must only be called by WriteUnitOfWork. See
* comments there for the semantics of units of work.
*/
virtual void beginWriteUnitOfWork() = 0;
virtual void endWriteUnitOfWork() = 0;
virtual bool inAWriteUnitOfWork() const = 0;
/**
* Acquires lock on the specified resource in the specified mode and returns the outcome
* of the operation. See the details for LockResult for more information on what the
* different results mean.
*
* Each successful acquisition of a lock on a given resource increments the reference count
* of the lock. Therefore, each call, which returns LOCK_OK must be matched with a
* corresponding call to unlock.
*
* @param resId Id of the resource to be locked.
* @param mode Mode in which the resource should be locked. Lock upgrades are allowed.
* @param timeout How long to wait for the lock to be granted, before
* returning LOCK_TIMEOUT. This parameter defaults to an infinite timeout.
* If Milliseconds(0) is passed, the request will return immediately, if
* the request could not be granted right away.
* @param checkDeadlock Whether to enable deadlock detection for this acquisition. This
* parameter is put in place until we can handle deadlocks at all places,
* which acquire locks.
*
* @return All LockResults except for LOCK_WAITING, because it blocks.
*/
virtual LockResult lock(ResourceId resId,
LockMode mode,
Milliseconds timeout = Milliseconds::max(),
bool checkDeadlock = false) = 0;
/**
* Downgrades the specified resource's lock mode without changing the reference count.
*/
virtual void downgrade(ResourceId resId, LockMode newMode) = 0;
/**
* Releases a lock previously acquired through a lock call. It is an error to try to
* release lock which has not been previously acquired (invariant violation).
*
* @return true if the lock was actually released; false if only the reference count was
* decremented, but the lock is still held.
*/
virtual bool unlock(ResourceId resId) = 0;
/**
* Retrieves the mode in which a lock is held or checks whether the lock held for a
* particular resource covers the specified mode.
*
* For example isLockHeldForMode will return true for MODE_S, if MODE_X is already held,
* because MODE_X covers MODE_S.
*/
virtual LockMode getLockMode(ResourceId resId) const = 0;
virtual bool isLockHeldForMode(ResourceId resId, LockMode mode) const = 0;
// These are shortcut methods for the above calls. They however check that the entire
// hierarchy is properly locked and because of this they are very expensive to call.
// Do not use them in performance critical code paths.
virtual bool isDbLockedForMode(StringData dbName, LockMode mode) const = 0;
virtual bool isCollectionLockedForMode(StringData ns, LockMode mode) const = 0;
/**
* Returns the resource that this locker is waiting/blocked on (if any). If the locker is
* not waiting for a resource the returned value will be invalid (isValid() == false).
*/
virtual ResourceId getWaitingResource() const = 0;
/**
* Describes a single lock acquisition for reporting/serialization purposes.
*/
struct OneLock {
// What lock resource is held?
ResourceId resourceId;
// In what mode is it held?
LockMode mode;
// Reporting/serialization order is by resourceId, which is the canonical locking order
bool operator<(const OneLock& rhs) const {
return resourceId < rhs.resourceId;
}
};
/**
* Returns information and locking statistics for this instance of the locker. Used to
* support the db.currentOp view. This structure is not thread-safe and ideally should
* be used only for obtaining the necessary information and then discarded instead of
* reused.
*/
struct LockerInfo {
// List of high-level locks held by this locker, sorted by ResourceId
std::vector<OneLock> locks;
// If isValid(), then what lock this particular locker is sleeping on
ResourceId waitingResource;
// Lock timing statistics
SingleThreadedLockStats stats;
};
virtual void getLockerInfo(LockerInfo* lockerInfo) const = 0;
/**
* LockSnapshot captures the state of all resources that are locked, what modes they're
* locked in, and how many times they've been locked in that mode.
*/
struct LockSnapshot {
// The global lock is handled differently from all other locks.
LockMode globalMode;
// The non-global non-flush locks held, sorted by granularity. That is, locks[i] is
// coarser or as coarse as locks[i + 1].
std::vector<OneLock> locks;
};
/**
* Retrieves all locks held by this transaction, other than RESOURCE_MUTEX locks, and what mode
* they're held in.
* Stores these locks in 'stateOut', destroying any previous state. Unlocks all locks
* held by this transaction. This functionality is used for yielding, which is
* voluntary/cooperative lock release and reacquisition in order to allow for interleaving
* of otherwise conflicting long-running operations.
*
* This functionality is also used for releasing locks on databases and collections
* when cursors are dormant and waiting for a getMore request.
*
* Returns true if locks are released. It is expected that restoreLockerImpl will be called
* in the future.
*
* Returns false if locks are not released. restoreLockState(...) does not need to be
* called in this case.
*/
virtual bool saveLockStateAndUnlock(LockSnapshot* stateOut) = 0;
/**
* Re-locks all locks whose state was stored in 'stateToRestore'.
*/
virtual void restoreLockState(const LockSnapshot& stateToRestore) = 0;
//
// These methods are legacy from LockerImpl and will eventually go away or be converted to
// calls into the Locker methods
//
virtual void dump() const = 0;
virtual bool isW() const = 0;
virtual bool isR() const = 0;
virtual bool isLocked() const = 0;
virtual bool isWriteLocked() const = 0;
virtual bool isReadLocked() const = 0;
/**
* Asserts that the Locker is effectively not in use and resets the locking statistics.
* This means, there should be no locks on it, no WUOW, etc, so it would be safe to call
* the destructor or reuse the Locker.
*/
virtual void assertEmptyAndReset() = 0;
/**
* Pending means we are currently trying to get a lock (could be the parallel batch writer
* lock).
*/
virtual bool hasLockPending() const = 0;
/**
* If set to false, this opts out of conflicting with replication's use of the
* ParallelBatchWriterMode lock. Code that opts-out must be ok with seeing an inconsistent view
* of data because within a batch, secondaries apply operations in a different order than on the
* primary. User operations should *never* opt out.
*/
void setShouldConflictWithSecondaryBatchApplication(bool newValue) {
_shouldConflictWithSecondaryBatchApplication = newValue;
}
bool shouldConflictWithSecondaryBatchApplication() const {
return _shouldConflictWithSecondaryBatchApplication;
}
protected:
Locker() {}
private:
bool _shouldConflictWithSecondaryBatchApplication = true;
};
} // namespace mongo
|