Files
mongo/db/dur_commitjob.h
2010-12-24 12:14:53 -05:00

264 lines
9.9 KiB
C++

/* @file dur_commitjob.h used by dur.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/>.
*/
#pragma once
#include "../util/alignedbuilder.h"
#include "../util/mongoutils/hash.h"
#include "../util/concurrency/synchronization.h"
#include "cmdline.h"
#include "durop.h"
#include "dur.h"
#include "taskqueue.h"
//#define DEBUG_WRITE_INTENT 1
namespace mongo {
namespace dur {
/** declaration of an intent to write to a region of a memory mapped view
*
* We store the end rather than the start pointer to make operator< faster
* since that is heavily used in set lookup.
*/
struct WriteIntent /* copyable */ {
WriteIntent() : w_ptr(0), p(0) { }
WriteIntent(void *a, unsigned b) : w_ptr(0), p((char*)a+b), len(b) { }
void* start() const { return (char*)p - len; }
void* end() const { return p; }
int length() const { return len; }
bool operator < (const WriteIntent& rhs) const { return end() < rhs.end(); }
// can they be merged?
bool overlaps(const WriteIntent& rhs) const {
return (start() <= rhs.end() && end() >= rhs.start());
}
// is merging necessary?
bool contains(const WriteIntent& rhs) const {
return (start() <= rhs.start() && end() >= rhs.end());
}
// merge into me
void absorb(const WriteIntent& other);
friend ostream& operator << (ostream& out, const WriteIntent& wi) {
return (out << "p: " << wi.p << " end: " << wi.end() << " len: " << wi.len);
}
/*temp*/ mutable void *w_ptr; // p is mapped from private to equivalent location in the writable mmap
private:
void *p; // intent to write up to p
unsigned len; // up to this len
};
/** try to remember things we have already marked for journalling. false negatives are ok if infrequent -
we will just log them twice.
*/
template<int Prime>
class Already : boost::noncopyable {
public:
Already() { clear(); }
void clear() { memset(this, 0, sizeof(*this)); }
/* see if we have Already recorded/indicated our write intent for this region of memory.
automatically upgrades the length if the length was shorter previously.
@return true if already indicated.
*/
bool checkAndSet(void* p, int len) {
unsigned x = mongoutils::hashPointer(p);
pair<void*, int> nd = nodes[x % N];
if( nd.first == p ) {
if( nd.second < len ) {
nd.second = len;
return false; // haven't indicated this len yet
}
return true; // already indicated
}
nd.first = p;
nd.second = len;
return false; // a new set
}
private:
enum { N = Prime }; // this should be small the idea is that it fits in the cpu cache easily
pair<void*,int> nodes[N];
};
/** our record of pending/uncommitted write intents */
class Writes : boost::noncopyable {
struct D {
void *p;
unsigned len;
static void go(const D& d);
};
public:
TaskQueue<D> _deferred;
Already<127> _alreadyNoted;
set<WriteIntent> _writes;
vector< shared_ptr<DurOp> > _ops; // all the ops other than basic writes
bool _drained; // _deferred is drained? for asserting/testing
/** reset the Writes structure (empties all the above) */
void clear();
/** merges into set (ie non-deferred version) */
void _insertWriteIntent(void* p, int len);
void insertWriteIntent(void* p, int len) {
#if defined(DEBUG_WRITE_INTENT)
if( _debug[p] < len )
_debug[p] = len;
#endif
D d;
d.p = p;
d.len = len;
_deferred.defer(d);
}
#ifdef _DEBUG
WriteIntent _last;
#endif
#if defined(DEBUG_WRITE_INTENT)
map<void*,int> _debug;
#endif
};
#if defined(DEBUG_WRITE_INTENT)
void assertAlreadyDeclared(void *, int len);
#else
inline void assertAlreadyDeclared(void *, int len) { }
#endif
/** A commit job object for a group commit. Currently there is one instance of this object.
concurrency: assumption is caller is appropriately locking.
for example note() invocations are from the write lock.
other uses are in a read lock from a single thread (durThread)
*/
class CommitJob : boost::noncopyable {
public:
AlignedBuilder _ab; // for direct i/o writes to journal
CommitJob() : _ab(4 * 1024 * 1024) , _hasWritten(false), _bytes(0) { }
/** record/note an intent to write */
void note(void* p, int len);
/** note an operation other than a "basic write" */
void noteOp(shared_ptr<DurOp> p);
set<WriteIntent>& writes() {
if( !_wi._drained ) {
// generally, you don't want to use the set until it is prepared (after deferred ops are applied)
// thus this assert here.
assert(false);
}
return _wi._writes;
}
vector< shared_ptr<DurOp> >& ops() { return _wi._ops; }
/** this method is safe to call outside of locks. when haswritten is false we don't do any group commit and avoid even
trying to acquire a lock, which might be helpful at times.
*/
bool hasWritten() const { return _hasWritten; }
/** we use the commitjob object over and over, calling reset() rather than reconstructing */
void reset();
/** the commit code calls this when data reaches the journal (on disk) */
void notifyCommitted() { _notify.notifyAll(); }
/** Wait until the next group commit occurs. That is, wait until someone calls notifyCommitted. */
void awaitNextCommit() {
if( hasWritten() )
_notify.wait();
}
/** we check how much written and if it is getting to be a lot, we commit sooner. */
size_t bytes() const { return _bytes; }
#if defined(_DEBUG)
const WriteIntent& lastWrite() const { return _wi._last; }
#endif
Writes& wi() { return _wi; }
private:
bool _hasWritten;
Writes _wi; // todo: fix name
size_t _bytes;
NotifyAll _notify; // for getlasterror fsync:true acknowledgements
};
extern CommitJob commitJob;
// inlines
inline 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 dbMutex.assertWriteLocked();
dassert( cmdLine.dur );
if( !_wi._alreadyNoted.checkAndSet(p, len) ) {
if( !_hasWritten ) {
// you can't be writing if one of these is pending, so this is a verification.
assert( !dbMutex._remapPrivateViewRequested );
// 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);
_bytes += len;
wassert( _wi._writes.size() < 2000000 );
assert( _wi._writes.size() < 20000000 );
}
}
}
}