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
|
/**
* Copyright (C) 2018-present MongoDB, Inc.
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the Server Side Public License, version 1,
* as published by MongoDB, Inc.
*
* 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
* Server Side Public License for more details.
*
* You should have received a copy of the Server Side Public License
* along with this program. If not, see
* <http://www.mongodb.com/licensing/server-side-public-license>.
*
* 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 Server Side 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.
*/
#include "mongo/platform/basic.h"
#include "mongo/db/repl/repl_client_info.h"
#include "mongo/db/client.h"
#include "mongo/db/operation_context.h"
#include "mongo/db/repl/replication_coordinator.h"
#include "mongo/logv2/log.h"
#include "mongo/util/decorable.h"
#define MONGO_LOGV2_DEFAULT_COMPONENT ::mongo::logv2::LogComponent::kReplication
namespace mongo {
namespace repl {
const Client::Decoration<ReplClientInfo> ReplClientInfo::forClient =
Client::declareDecoration<ReplClientInfo>();
namespace {
// We use a struct to wrap lastOpSetExplicitly here in order to give the boolean a default value
// when initially constructed for the associated OperationContext.
struct LastOpInfo {
bool lastOpSetExplicitly = false;
};
static const OperationContext::Decoration<LastOpInfo> lastOpInfo =
OperationContext::declareDecoration<LastOpInfo>();
} // namespace
bool ReplClientInfo::lastOpWasSetExplicitlyByClientForCurrentOperation(
OperationContext* opCtx) const {
return lastOpInfo(opCtx).lastOpSetExplicitly;
}
void ReplClientInfo::setLastOp(OperationContext* opCtx, const OpTime& ot) {
invariant(ot >= _lastOp);
_lastOp = ot;
lastOpInfo(opCtx).lastOpSetExplicitly = true;
}
void ReplClientInfo::setLastOpToSystemLastOpTime(OperationContext* opCtx) {
auto replCoord = repl::ReplicationCoordinator::get(opCtx->getServiceContext());
if (replCoord->isReplEnabled() && opCtx->writesAreReplicated()) {
// If a multi-document transaction or regular write actually performs a write op, the
// lastOp in the client is enough for write concern and there is no need to call
// setLastOpToSystemLastOpTime.
// On the other hand, if a transaction is a no-op and we are using speculative majority read
// concern, there is risk of a transaction reading data that might not have yet been
// majority commited, and later waiting for write concern on a lastApplied which is older
// than the timestamp to which the read corresponds. This is possible due to the fact
// that lastApplied is advanced in an onCommit hook AFTER commiting writes (and making them
// visible), leaving a small window of time where a concurrent transaction might
// speculatively read local data, and then wait on a lastApplied which has yet to be
// advanced.
// Multi-doc transactions perform a no-op write on commit time and advance the lastOp. So it
// is no longer necessary for transactions to update the client's lastOp (calling
// setLastOpToSystemLastOpTime).
// Non-transaction operations resulting in no-op suffer from the same issue as transactions,
// but there is no built-in mechanism to advance the lastOp and client's lastOp has to be
// updated.
// Due to this, the lastOp has to be updated directly from the oplog's top entry instead of
// the lastApplied in the ReplicationCoordinator. This works because the oplog entry is
// commited together with the write op itself.
auto latestWriteOpTimeSW = replCoord->getLatestWriteOpTime(opCtx);
auto status = latestWriteOpTimeSW.getStatus();
OpTime systemOpTime;
if (status.isOK()) {
systemOpTime = latestWriteOpTimeSW.getValue();
} else {
// Fall back to use my lastAppliedOpTime if we failed to get the latest OpTime from
// storage. In most cases, it is safe to ignore errors because if
// getLatestWriteOpTime returns an error, we cannot use the same opCtx to wait for
// writeConcern anyways. But getLastError from the same client could use a different
// opCtx to wait for the lastOp. So this is a best effort attempt to set the lastOp
// to the in-memory lastAppliedOpTime (which could be lagged). And this is a known
// bug in getLastError.
systemOpTime = replCoord->getMyLastAppliedOpTime();
if (status == ErrorCodes::OplogOperationUnsupported ||
status == ErrorCodes::NamespaceNotFound ||
status == ErrorCodes::CollectionIsEmpty || ErrorCodes::isNotPrimaryError(status)) {
// It is ok if the storage engine does not support getLatestOplogTimestamp() or
// if the oplog is empty. If the node stepped down in between, it is correct to
// use lastAppliedOpTime as last OpTime.
status = Status::OK();
}
// We will continue trying to set client's lastOp to lastAppliedOpTime as a best-effort
// alternative to getLatestWriteOpTime. And we will then throw after setting client's
// lastOp if getLatestWriteOpTime has failed with a error code other than the ones
// above.
}
// If the system timestamp has gone backwards, that must mean that there was a rollback.
// If the system optime has a higher term but a lower timestamp than the client's lastOp, it
// means that this node's wallclock time was ahead of the current primary's before it rolled
// back. This is safe, but the timestamp of the last op for a Client should never go
// backwards, so just leave the last op for this Client as it was.
if (systemOpTime.getTerm() >= _lastOp.getTerm() &&
systemOpTime.getTimestamp() >= _lastOp.getTimestamp()) {
_lastOp = systemOpTime;
} else {
LOGV2(21280,
"Not setting the last OpTime for this Client from {lastOp} to the current system "
"time of {systemOpTime} as that would be moving the OpTime backwards. This "
"should only happen if there was a rollback recently",
"Not setting the last OpTime for this Client to the current system time as that "
"would be moving the OpTime backwards. This should only happen if there was a "
"rollback recently",
"lastOp"_attr = _lastOp,
"systemOpTime"_attr = systemOpTime);
}
lastOpInfo(opCtx).lastOpSetExplicitly = true;
// Throw if getLatestWriteOpTime failed.
uassertStatusOK(status);
}
}
void ReplClientInfo::setLastOpToSystemLastOpTimeIgnoringCtxInterrupted(OperationContext* opCtx) {
try {
repl::ReplClientInfo::forClient(opCtx->getClient()).setLastOpToSystemLastOpTime(opCtx);
} catch (DBException& e) {
if (opCtx && !opCtx->checkForInterruptNoAssert().isOK()) {
// In most cases, it is safe to ignore all errors when the OperationContext is
// interrupted because we cannot use that OperationContext to wait for writeConcern
// anyways.
LOGV2_DEBUG(21281,
2,
"Ignoring set last op error: {error}",
"Ignoring set last op error",
"error"_attr = e.toStatus());
return;
}
// Context was not interrupted, throw error up to caller.
throw;
}
}
} // namespace repl
} // namespace mongo
|
