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
|
/* @file dur_commitjob.cpp */
/**
* Copyright (C) 2009 10gen 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/>.
*/
#include "pch.h"
#include "dur_commitjob.h"
#include "dur_stats.h"
#include "taskqueue.h"
#include "client.h"
namespace mongo {
namespace dur {
BOOST_STATIC_ASSERT( UncommittedBytesLimit > BSONObjMaxInternalSize * 3 );
BOOST_STATIC_ASSERT( sizeof(void*)==4 || UncommittedBytesLimit > BSONObjMaxInternalSize * 6 );
void WriteIntent::absorb(const WriteIntent& other) {
dassert(overlaps(other));
void* newStart = min(start(), other.start());
p = max(p, other.p);
len = (char*)p - (char*)newStart;
dassert(contains(other));
}
void Writes::clear() {
d.dbMutex.assertAtLeastReadLocked();
commitJob.groupCommitMutex.dassertLocked();
_alreadyNoted.clear();
_intents.clear();
_durOps.clear();
#if defined(DEBUG_WRITE_INTENT)
cout << "_debug clear\n";
_debug.clear();
#endif
}
#if defined(DEBUG_WRITE_INTENT)
void assertAlreadyDeclared(void *p, int len) {
if( commitJob.wi()._debug[p] >= len )
return;
log() << "assertAlreadyDeclared fails " << (void*)p << " len:" << len << ' ' << commitJob.wi()._debug[p] << endl;
printStackTrace();
abort();
}
#endif
/** note an operation other than a "basic write" */
void CommitJob::noteOp(shared_ptr<DurOp> p) {
dassert( cmdLine.dur );
// DurOp's are rare so it is ok to have the lock cost here
SimpleMutex::scoped_lock lk(groupCommitMutex);
cc().writeHappened();
_hasWritten = true;
_wi._durOps.push_back(p);
}
size_t privateMapBytes = 0; // used by _REMAPPRIVATEVIEW to track how much / how fast to remap
void CommitJob::beginCommit() {
DEV d.dbMutex.assertAtLeastReadLocked();
_commitNumber = _notify.now();
stats.curr->_commits++;
}
void CommitJob::reset() {
_hasWritten = false;
_wi.clear();
privateMapBytes += _bytes;
_bytes = 0;
_nSinceCommitIfNeededCall = 0;
}
CommitJob::CommitJob() : _hasWritten(false),
_bytes(0), _nSinceCommitIfNeededCall(0), groupCommitMutex("groupCommit")
{
_commitNumber = 0;
}
extern unsigned notesThisLock;
void CommitJob::note(void* p, int len) {
// from the point of view of the dur module, it would be fine (i think) to only
// be read locked here. but must be at least read locked to avoid race with
// remapprivateview
DEV notesThisLock++;
//DEV d.dbMutex.assertWriteLocked();
log() << "TODO temp finish concurrency" << endl;
dassert( cmdLine.dur );
cc().writeHappened();
if( !_wi._alreadyNoted.checkAndSet(p, len) ) {
MemoryMappedFile::makeWritable(p, len);
if( !_hasWritten ) {
// we don't bother doing a group commit when nothing is written, so we have a var to track that
_hasWritten = true;
}
/** tips for debugging:
if you have an incorrect diff between data files in different folders
(see jstests/dur/quick.js for example),
turn this on and see what is logged. if you have a copy of its output from before the
regression, a simple diff of these lines would tell you a lot likely.
*/
#if 0 && defined(_DEBUG)
{
static int n;
if( ++n < 10000 ) {
size_t ofs;
MongoMMF *mmf = privateViews._find(w.p, ofs);
if( mmf ) {
log() << "DEBUG note write intent " << w.p << ' ' << mmf->filename() << " ofs:" << hex << ofs << " len:" << w.len << endl;
}
else {
log() << "DEBUG note write intent " << w.p << ' ' << w.len << " NOT FOUND IN privateViews" << endl;
}
}
else if( n == 10000 ) {
log() << "DEBUG stopping write intent logging, too much to log" << endl;
}
}
#endif
// remember intent. we will journal it in a bit
_wi.insertWriteIntent(p, len);
{
// a bit over conservative in counting pagebytes used
static size_t lastPos; // note this doesn't reset with each commit, but that is ok we aren't being that precise
size_t x = ((size_t) p) & ~0xfff; // round off to page address (4KB)
if( x != lastPos ) {
lastPos = x;
unsigned b = (len+4095) & ~0xfff;
_bytes += b;
#if defined(_DEBUG)
_nSinceCommitIfNeededCall++;
if( _nSinceCommitIfNeededCall >= 80 ) {
if( _nSinceCommitIfNeededCall % 40 == 0 ) {
log() << "debug nsincecommitifneeded:" << _nSinceCommitIfNeededCall << " bytes:" << _bytes << endl;
if( _nSinceCommitIfNeededCall == 120 || _nSinceCommitIfNeededCall == 1200 ) {
log() << "_DEBUG printing stack given high nsinccommitifneeded number" << endl;
printStackTrace();
}
}
}
#endif
if (_bytes > UncommittedBytesLimit * 3) {
static time_t lastComplain;
static unsigned nComplains;
// throttle logging
if( ++nComplains < 100 || time(0) - lastComplain >= 60 ) {
lastComplain = time(0);
warning() << "DR102 too much data written uncommitted " << _bytes/1000000.0 << "MB" << endl;
if( nComplains < 10 || nComplains % 10 == 0 ) {
// wassert makes getLastError show an error, so we just print stack trace
printStackTrace();
}
}
}
}
}
}
}
}
}
|
