void RecordStoreV1Base::deleteRecord( TransactionExperiment* txn, const DiskLoc& dl ) {
Record* todelete = recordFor( dl );
/* remove ourself from the record next/prev chain */
{
if ( todelete->prevOfs() != DiskLoc::NullOfs ) {
DiskLoc prev = getPrevRecordInExtent( dl );
Record* prevRecord = recordFor( prev );
txn->writingInt( prevRecord->nextOfs() ) = todelete->nextOfs();
}
if ( todelete->nextOfs() != DiskLoc::NullOfs ) {
DiskLoc next = getNextRecord( dl );
Record* nextRecord = recordFor( next );
txn->writingInt( nextRecord->prevOfs() ) = todelete->prevOfs();
}
}
/* remove ourself from extent pointers */
{
Extent *e = txn->writing( _getExtent( _getExtentLocForRecord( dl ) ) );
if ( e->firstRecord == dl ) {
if ( todelete->nextOfs() == DiskLoc::NullOfs )
e->firstRecord.Null();
else
e->firstRecord.set(dl.a(), todelete->nextOfs() );
}
if ( e->lastRecord == dl ) {
if ( todelete->prevOfs() == DiskLoc::NullOfs )
e->lastRecord.Null();
else
e->lastRecord.set(dl.a(), todelete->prevOfs() );
}
}
/* add to the free list */
{
_details->incrementStats( txn, -1 * todelete->netLength(), -1 );
if ( _isSystemIndexes ) {
/* temp: if in system.indexes, don't reuse, and zero out: we want to be
careful until validated more, as IndexDetails has pointers
to this disk location. so an incorrectly done remove would cause
a lot of problems.
*/
memset( txn->writingPtr(todelete, todelete->lengthWithHeaders() ),
0, todelete->lengthWithHeaders() );
}
else {
DEV {
unsigned long long *p = reinterpret_cast<unsigned long long *>( todelete->data() );
*txn->writing(p) = 0;
}
addDeletedRec(txn, dl);
}
}
}
StatusWith<DiskLoc> HeapRecordStore::updateRecord(OperationContext* txn,
const DiskLoc& loc,
const char* data,
int len,
bool enforceQuota,
UpdateMoveNotifier* notifier ) {
HeapRecord* oldRecord = recordFor( loc );
int oldLen = oldRecord->size;
if (_isCapped && len > oldLen) {
return StatusWith<DiskLoc>( ErrorCodes::InternalError,
"failing update: objects in a capped ns cannot grow",
10003 );
}
HeapRecord newRecord(len);
memcpy(newRecord.data.get(), data, len);
txn->recoveryUnit()->registerChange(new RemoveChange(_data, loc, *oldRecord));
_data->dataSize += len - oldLen;
*oldRecord = newRecord;
cappedDeleteAsNeeded(txn);
return StatusWith<DiskLoc>(loc);
}
StatusWith<DiskLoc> RecordStoreV1Base::_insertRecord( OperationContext* txn,
const char* data,
int len,
bool enforceQuota ) {
int lenWHdr = getRecordAllocationSize( len + Record::HeaderSize );
fassert( 17208, lenWHdr >= ( len + Record::HeaderSize ) );
StatusWith<DiskLoc> loc = allocRecord( txn, lenWHdr, enforceQuota );
if ( !loc.isOK() )
return loc;
Record *r = recordFor( loc.getValue() );
fassert( 17210, r->lengthWithHeaders() >= lenWHdr );
// copy the data
r = reinterpret_cast<Record*>( txn->recoveryUnit()->writingPtr(r, lenWHdr) );
memcpy( r->data(), data, len );
_addRecordToRecListInExtent(txn, r, loc.getValue());
_details->incrementStats( txn, r->netLength(), 1 );
return loc;
}
StatusWith<DiskLoc> RecordStoreV1Base::insertRecord( OperationContext* txn,
const DocWriter* doc,
bool enforceQuota ) {
int docSize = doc->documentSize();
if ( docSize < 4 ) {
return StatusWith<DiskLoc>( ErrorCodes::InvalidLength,
"record has to be >= 4 bytes" );
}
int lenWHdr = docSize + Record::HeaderSize;
if ( doc->addPadding() )
lenWHdr = getRecordAllocationSize( lenWHdr );
StatusWith<DiskLoc> loc = allocRecord( txn, lenWHdr, enforceQuota );
if ( !loc.isOK() )
return loc;
Record *r = recordFor( loc.getValue() );
fassert( 17319, r->lengthWithHeaders() >= lenWHdr );
r = reinterpret_cast<Record*>( txn->recoveryUnit()->writingPtr(r, lenWHdr) );
doc->writeDocument( r->data() );
_addRecordToRecListInExtent(txn, r, loc.getValue());
_details->incrementStats( txn, r->netLength(), 1 );
_paddingFits( txn );
return loc;
}
Status HeapRecordStore::updateWithDamages( OperationContext* txn,
const DiskLoc& loc,
const RecordData& oldRec,
const char* damageSource,
const mutablebson::DamageVector& damages ) {
HeapRecord* oldRecord = recordFor( loc );
const int len = oldRecord->size;
HeapRecord newRecord(len);
memcpy(newRecord.data.get(), oldRecord->data.get(), len);
txn->recoveryUnit()->registerChange(new RemoveChange(_data, loc, *oldRecord));
*oldRecord = newRecord;
cappedDeleteAsNeeded(txn);
char* root = newRecord.data.get();
mutablebson::DamageVector::const_iterator where = damages.begin();
const mutablebson::DamageVector::const_iterator end = damages.end();
for( ; where != end; ++where ) {
const char* sourcePtr = damageSource + where->sourceOffset;
char* targetPtr = root + where->targetOffset;
std::memcpy(targetPtr, sourcePtr, where->size);
}
*oldRecord = newRecord;
return Status::OK();
}
StatusWith<RecordId> RecordStoreV1Base::_insertRecord(OperationContext* txn,
const char* data,
int len,
bool enforceQuota) {
const int lenWHdr = len + MmapV1RecordHeader::HeaderSize;
const int lenToAlloc = shouldPadInserts() ? quantizeAllocationSpace(lenWHdr) : lenWHdr;
fassert(17208, lenToAlloc >= lenWHdr);
StatusWith<DiskLoc> loc = allocRecord(txn, lenToAlloc, enforceQuota);
if (!loc.isOK())
return StatusWith<RecordId>(loc.getStatus());
MmapV1RecordHeader* r = recordFor(loc.getValue());
fassert(17210, r->lengthWithHeaders() >= lenWHdr);
// copy the data
r = reinterpret_cast<MmapV1RecordHeader*>(txn->recoveryUnit()->writingPtr(r, lenWHdr));
memcpy(r->data(), data, len);
_addRecordToRecListInExtent(txn, r, loc.getValue());
_details->incrementStats(txn, r->netLength(), 1);
return StatusWith<RecordId>(loc.getValue().toRecordId());
}
StatusWith<RecordId> RecordStoreV1Base::insertRecord(OperationContext* txn,
const DocWriter* doc,
bool enforceQuota) {
int docSize = doc->documentSize();
if (docSize < 4) {
return StatusWith<RecordId>(ErrorCodes::InvalidLength, "record has to be >= 4 bytes");
}
const int lenWHdr = docSize + MmapV1RecordHeader::HeaderSize;
if (lenWHdr > MaxAllowedAllocation) {
return StatusWith<RecordId>(ErrorCodes::InvalidLength, "record has to be <= 16.5MB");
}
const int lenToAlloc =
(doc->addPadding() && shouldPadInserts()) ? quantizeAllocationSpace(lenWHdr) : lenWHdr;
StatusWith<DiskLoc> loc = allocRecord(txn, lenToAlloc, enforceQuota);
if (!loc.isOK())
return StatusWith<RecordId>(loc.getStatus());
MmapV1RecordHeader* r = recordFor(loc.getValue());
fassert(17319, r->lengthWithHeaders() >= lenWHdr);
r = reinterpret_cast<MmapV1RecordHeader*>(txn->recoveryUnit()->writingPtr(r, lenWHdr));
doc->writeDocument(r->data());
_addRecordToRecListInExtent(txn, r, loc.getValue());
_details->incrementStats(txn, r->netLength(), 1);
return StatusWith<RecordId>(loc.getValue().toRecordId());
}
void RecordStoreV1Base::IntraExtentIterator::advance() {
if (_curr.isNull())
return;
const MmapV1RecordHeader* rec = recordFor(_curr);
const int nextOfs = _forward ? rec->nextOfs() : rec->prevOfs();
_curr = (nextOfs == DiskLoc::NullOfs ? DiskLoc() : DiskLoc(_curr.a(), nextOfs));
}
void RecordStoreV1Base::_addRecordToRecListInExtent(OperationContext* txn,
Record *r,
DiskLoc loc) {
dassert( recordFor(loc) == r );
Extent *e = _getExtent( _getExtentLocForRecord( loc ) );
if ( e->lastRecord.isNull() ) {
*txn->recoveryUnit()->writing(&e->firstRecord) = loc;
*txn->recoveryUnit()->writing(&e->lastRecord) = loc;
r->prevOfs() = r->nextOfs() = DiskLoc::NullOfs;
}
else {
Record *oldlast = recordFor(e->lastRecord);
r->prevOfs() = e->lastRecord.getOfs();
r->nextOfs() = DiskLoc::NullOfs;
txn->recoveryUnit()->writingInt(oldlast->nextOfs()) = loc.getOfs();
*txn->recoveryUnit()->writing(&e->lastRecord) = loc;
}
}
DiskLoc RecordStoreV1Base::getNextRecordInExtent( const DiskLoc& loc ) const {
int nextOffset = recordFor( loc )->nextOfs();
if ( nextOffset == DiskLoc::NullOfs )
return DiskLoc();
fassert( 17441, abs(nextOffset) >= 8 ); // defensive
return DiskLoc( loc.a(), nextOffset );
}
DiskLoc ExtentManager::getNextRecordInExtent( const DiskLoc& loc ) const {
int nextOffset = recordFor( loc )->nextOfs();
if ( nextOffset == DiskLoc::NullOfs )
return DiskLoc();
fassert( 16967, abs(nextOffset) >= 8 ); // defensive
return DiskLoc( loc.a(), nextOffset );
}
DiskLoc ExtentManager::getPrevRecordInExtent( const DiskLoc& loc ) const {
int prevOffset = recordFor( loc )->prevOfs();
if ( prevOffset == DiskLoc::NullOfs )
return DiskLoc();
fassert( 16968, abs(prevOffset) >= 8 ); // defensive
return DiskLoc( loc.a(), prevOffset );
}
DiskLoc RecordStoreV1Base::getPrevRecordInExtent( OperationContext* txn, const DiskLoc& loc ) const {
int prevOffset = recordFor( loc )->prevOfs();
if ( prevOffset == DiskLoc::NullOfs )
return DiskLoc();
fassert( 17442, abs(prevOffset) >= 8 ); // defensive
DiskLoc result( loc.a(), prevOffset );
return result;
}
DiskLoc RecordStoreV1Base::getPrevRecordInExtent( const DiskLoc& loc ) const {
int prevOffset = recordFor( loc )->prevOfs();
if ( prevOffset == DiskLoc::NullOfs )
return DiskLoc();
fassert( 17442, abs(prevOffset) >= 8 ); // defensive
return DiskLoc( loc.a(), prevOffset );
}
DiskLoc RecordStoreV1Base::IntraExtentIterator::getNext() {
if (_curr.isNull())
return DiskLoc();
const DiskLoc out = _curr; // we always return where we were, not where we will be.
const Record* rec = recordFor(_curr);
const int nextOfs = _forward ? rec->nextOfs() : rec->prevOfs();
_curr = (nextOfs == DiskLoc::NullOfs ? DiskLoc() : DiskLoc(_curr.a(), nextOfs));
return out;
}
bool RecordStoreV1Base::findRecord( OperationContext* txn,
const DiskLoc& loc, RecordData* rd ) const {
// this is a bit odd, as the semantics of using the storage engine imply it _has_ to be.
// And in fact we can't actually check.
// So we assume the best.
Record* rec = recordFor(loc);
if ( !rec ) {
return false;
}
*rd = rec->toRecordData();
return true;
}
Status HeapRecordStore::updateWithDamages( OperationContext* txn,
const DiskLoc& loc,
const char* damangeSource,
const mutablebson::DamageVector& damages ) {
HeapRecord* rec = recordFor( loc );
char* root = rec->data();
// All updates were in place. Apply them via durability and writing pointer.
mutablebson::DamageVector::const_iterator where = damages.begin();
const mutablebson::DamageVector::const_iterator end = damages.end();
for( ; where != end; ++where ) {
const char* sourcePtr = damangeSource + where->sourceOffset;
char* targetPtr = root + where->targetOffset;
std::memcpy(targetPtr, sourcePtr, where->size);
}
return Status::OK();
}
StatusWith<RecordId> RecordStoreV1Base::updateRecord(OperationContext* txn,
const RecordId& oldLocation,
const char* data,
int dataSize,
bool enforceQuota,
UpdateNotifier* notifier) {
MmapV1RecordHeader* oldRecord = recordFor(DiskLoc::fromRecordId(oldLocation));
if (oldRecord->netLength() >= dataSize) {
// Make sure to notify other queries before we do an in-place update.
if (notifier) {
Status callbackStatus = notifier->recordStoreGoingToUpdateInPlace(txn, oldLocation);
if (!callbackStatus.isOK())
return StatusWith<RecordId>(callbackStatus);
}
// we fit
memcpy(txn->recoveryUnit()->writingPtr(oldRecord->data(), dataSize), data, dataSize);
return StatusWith<RecordId>(oldLocation);
}
// We enforce the restriction of unchanging capped doc sizes above the storage layer.
invariant(!isCapped());
// we have to move
if (dataSize + MmapV1RecordHeader::HeaderSize > MaxAllowedAllocation) {
return StatusWith<RecordId>(ErrorCodes::InvalidLength, "record has to be <= 16.5MB");
}
StatusWith<RecordId> newLocation = _insertRecord(txn, data, dataSize, enforceQuota);
if (!newLocation.isOK())
return newLocation;
// insert worked, so we delete old record
if (notifier) {
Status moveStatus = notifier->recordStoreGoingToMove(
txn, oldLocation, oldRecord->data(), oldRecord->netLength());
if (!moveStatus.isOK())
return StatusWith<RecordId>(moveStatus);
}
deleteRecord(txn, oldLocation);
return newLocation;
}
StatusWith<RecordData> RecordStoreV1Base::updateWithDamages(
OperationContext* txn,
const RecordId& loc,
const RecordData& oldRec,
const char* damageSource,
const mutablebson::DamageVector& damages) {
MmapV1RecordHeader* rec = recordFor(DiskLoc::fromRecordId(loc));
char* root = rec->data();
// All updates were in place. Apply them via durability and writing pointer.
mutablebson::DamageVector::const_iterator where = damages.begin();
const mutablebson::DamageVector::const_iterator end = damages.end();
for (; where != end; ++where) {
const char* sourcePtr = damageSource + where->sourceOffset;
void* targetPtr = txn->recoveryUnit()->writingPtr(root + where->targetOffset, where->size);
std::memcpy(targetPtr, sourcePtr, where->size);
}
return rec->toRecordData();
}
StatusWith<DiskLoc> RecordStoreV1Base::updateRecord( OperationContext* txn,
const DiskLoc& oldLocation,
const char* data,
int dataSize,
bool enforceQuota,
UpdateMoveNotifier* notifier ) {
Record* oldRecord = recordFor( oldLocation );
if ( oldRecord->netLength() >= dataSize ) {
// we fit
_paddingFits( txn );
memcpy( txn->recoveryUnit()->writingPtr( oldRecord->data(), dataSize ), data, dataSize );
return StatusWith<DiskLoc>( oldLocation );
}
if ( isCapped() )
return StatusWith<DiskLoc>( ErrorCodes::InternalError,
"failing update: objects in a capped ns cannot grow",
10003 );
// we have to move
_paddingTooSmall( txn );
StatusWith<DiskLoc> newLocation = _insertRecord( txn, data, dataSize, enforceQuota );
if ( !newLocation.isOK() )
return newLocation;
// insert worked, so we delete old record
if ( notifier ) {
Status moveStatus = notifier->recordStoreGoingToMove( txn,
oldLocation,
oldRecord->data(),
oldRecord->netLength() );
if ( !moveStatus.isOK() )
return StatusWith<DiskLoc>( moveStatus );
}
deleteRecord( txn, oldLocation );
return newLocation;
}
StatusWith<DiskLoc> HeapRecordStore::updateRecord(OperationContext* txn,
const DiskLoc& oldLocation,
const char* data,
int len,
bool enforceQuota,
UpdateMoveNotifier* notifier ) {
HeapRecord* oldRecord = recordFor( oldLocation );
int oldLen = oldRecord->netLength();
// If the length of the new data is <= the length of the old data then just
// memcopy into the old space
if ( len <= oldLen) {
memcpy(oldRecord->data(), data, len);
_dataSize += len - oldLen;
return StatusWith<DiskLoc>(oldLocation);
}
if ( _isCapped ) {
return StatusWith<DiskLoc>( ErrorCodes::InternalError,
"failing update: objects in a capped ns cannot grow",
10003 );
}
// If the length of the new data exceeds the size of the old Record, we need to allocate
// a new Record, and delete the old one
const int lengthWithHeaders = len + HeapRecord::HeaderSize;
boost::shared_array<char> buf(new char[lengthWithHeaders]);
HeapRecord* rec = reinterpret_cast<HeapRecord*>(buf.get());
rec->lengthWithHeaders() = lengthWithHeaders;
memcpy(rec->data(), data, len);
_records[oldLocation] = buf;
_dataSize += len - oldLen;
cappedDeleteAsNeeded(txn);
return StatusWith<DiskLoc>(oldLocation);
}
StatusWith<DiskLoc> RecordStoreV1Base::insertRecord( TransactionExperiment* txn,
const DocWriter* doc,
int quotaMax ) {
int lenWHdr = doc->documentSize() + Record::HeaderSize;
if ( doc->addPadding() )
lenWHdr = getRecordAllocationSize( lenWHdr );
StatusWith<DiskLoc> loc = allocRecord( txn, lenWHdr, quotaMax );
if ( !loc.isOK() )
return loc;
Record *r = recordFor( loc.getValue() );
fassert( 17319, r->lengthWithHeaders() >= lenWHdr );
r = reinterpret_cast<Record*>( txn->writingPtr(r, lenWHdr) );
doc->writeDocument( r->data() );
_addRecordToRecListInExtent(txn, r, loc.getValue());
_details->incrementStats( txn, r->netLength(), 1 );
return loc;
}
void RecordStoreV1Base::deleteRecord( OperationContext* txn, const DiskLoc& dl ) {
Record* todelete = recordFor( dl );
invariant( todelete->netLength() >= 4 ); // this is required for defensive code
/* remove ourself from the record next/prev chain */
{
if ( todelete->prevOfs() != DiskLoc::NullOfs ) {
DiskLoc prev = getPrevRecordInExtent( txn, dl );
Record* prevRecord = recordFor( prev );
txn->recoveryUnit()->writingInt( prevRecord->nextOfs() ) = todelete->nextOfs();
}
if ( todelete->nextOfs() != DiskLoc::NullOfs ) {
DiskLoc next = getNextRecord( txn, dl );
Record* nextRecord = recordFor( next );
txn->recoveryUnit()->writingInt( nextRecord->prevOfs() ) = todelete->prevOfs();
}
}
/* remove ourself from extent pointers */
{
DiskLoc extentLoc = todelete->myExtentLoc(dl);
Extent *e = _getExtent( txn, extentLoc );
if ( e->firstRecord == dl ) {
txn->recoveryUnit()->writing(&e->firstRecord);
if ( todelete->nextOfs() == DiskLoc::NullOfs )
e->firstRecord.Null();
else
e->firstRecord.set(dl.a(), todelete->nextOfs() );
}
if ( e->lastRecord == dl ) {
txn->recoveryUnit()->writing(&e->lastRecord);
if ( todelete->prevOfs() == DiskLoc::NullOfs )
e->lastRecord.Null();
else
e->lastRecord.set(dl.a(), todelete->prevOfs() );
}
}
/* add to the free list */
{
_details->incrementStats( txn, -1 * todelete->netLength(), -1 );
if ( _isSystemIndexes ) {
/* temp: if in system.indexes, don't reuse, and zero out: we want to be
careful until validated more, as IndexDetails has pointers
to this disk location. so an incorrectly done remove would cause
a lot of problems.
*/
memset( txn->recoveryUnit()->writingPtr(todelete, todelete->lengthWithHeaders() ),
0, todelete->lengthWithHeaders() );
}
else {
// this is defensive so we can detect if we are still using a location
// that was deleted
memset(txn->recoveryUnit()->writingPtr(todelete->data(), 4), 0xee, 4);
addDeletedRec(txn, dl);
}
}
}
Extent* ExtentManager::extentFor( const DiskLoc& loc ) const {
Record* record = recordFor( loc );
DiskLoc extentLoc( loc.a(), record->extentOfs() );
return getExtent( extentLoc );
}
void SimpleRecordStoreV1::_compactExtent(OperationContext* txn,
const DiskLoc extentLoc,
int extentNumber,
RecordStoreCompactAdaptor* adaptor,
const CompactOptions* compactOptions,
CompactStats* stats ) {
log() << "compact begin extent #" << extentNumber
<< " for namespace " << _ns << " " << extentLoc;
unsigned oldObjSize = 0; // we'll report what the old padding was
unsigned oldObjSizeWithPadding = 0;
Extent* const sourceExtent = _extentManager->getExtent( extentLoc );
sourceExtent->assertOk();
fassert( 17437, sourceExtent->validates(extentLoc) );
{
// The next/prev Record pointers within the Extent might not be in order so we first
// page in the whole Extent sequentially.
// TODO benchmark on slow storage to verify this is measurably faster.
log() << "compact paging in len=" << sourceExtent->length/1000000.0 << "MB" << endl;
Timer t;
size_t length = sourceExtent->length;
touch_pages( reinterpret_cast<const char*>(sourceExtent), length );
int ms = t.millis();
if( ms > 1000 )
log() << "compact end paging in " << ms << "ms "
<< sourceExtent->length/1000000.0/t.seconds() << "MB/sec" << endl;
}
{
// Move each Record out of this extent and insert it in to the "new" extents.
log() << "compact copying records" << endl;
long long totalNetSize = 0;
long long nrecords = 0;
DiskLoc nextSourceLoc = sourceExtent->firstRecord;
while (!nextSourceLoc.isNull()) {
txn->checkForInterrupt();
WriteUnitOfWork wunit(txn);
Record* recOld = recordFor(nextSourceLoc);
RecordData oldData = recOld->toRecordData();
nextSourceLoc = getNextRecordInExtent(txn, nextSourceLoc);
if ( compactOptions->validateDocuments && !adaptor->isDataValid( oldData ) ) {
// object is corrupt!
log() << "compact removing corrupt document!";
stats->corruptDocuments++;
}
else {
// How much data is in the record. Excludes padding and Record headers.
const unsigned rawDataSize = adaptor->dataSize( oldData );
nrecords++;
oldObjSize += rawDataSize;
oldObjSizeWithPadding += recOld->netLength();
// Allocation sizes include the headers and possibly some padding.
const unsigned minAllocationSize = rawDataSize + Record::HeaderSize;
unsigned allocationSize = minAllocationSize;
switch( compactOptions->paddingMode ) {
case CompactOptions::NONE: // no padding, unless using powerOf2Sizes
if ( _details->isUserFlagSet(Flag_UsePowerOf2Sizes) )
allocationSize = quantizePowerOf2AllocationSpace(minAllocationSize);
else
allocationSize = minAllocationSize;
break;
case CompactOptions::PRESERVE: // keep original padding
allocationSize = recOld->lengthWithHeaders();
break;
case CompactOptions::MANUAL: // user specified how much padding to use
allocationSize = compactOptions->computeRecordSize(minAllocationSize);
if (allocationSize < minAllocationSize
|| allocationSize > BSONObjMaxUserSize / 2 ) {
allocationSize = minAllocationSize;
}
break;
}
invariant(allocationSize >= minAllocationSize);
// Copy the data to a new record. Because we orphaned the record freelist at the
// start of the compact, this insert will allocate a record in a new extent.
// See the comment in compact() for more details.
CompactDocWriter writer( recOld, rawDataSize, allocationSize );
StatusWith<DiskLoc> status = insertRecord( txn, &writer, false );
uassertStatusOK( status.getStatus() );
const Record* newRec = recordFor(status.getValue());
invariant(unsigned(newRec->netLength()) >= rawDataSize);
totalNetSize += newRec->netLength();
// Tells the caller that the record has been moved, so it can do things such as
// add it to indexes.
adaptor->inserted(newRec->toRecordData(), status.getValue());
}
// Remove the old record from the linked list of records withing the sourceExtent.
// The old record is not added to the freelist as we will be freeing the whole
// extent at the end.
*txn->recoveryUnit()->writing(&sourceExtent->firstRecord) = nextSourceLoc;
if (nextSourceLoc.isNull()) {
// Just moved the last record out of the extent. Mark extent as empty.
*txn->recoveryUnit()->writing(&sourceExtent->lastRecord) = DiskLoc();
}
else {
Record* newFirstRecord = recordFor(nextSourceLoc);
txn->recoveryUnit()->writingInt(newFirstRecord->prevOfs()) = DiskLoc::NullOfs;
}
// Adjust the stats to reflect the removal of the old record. The insert above
// handled adjusting the stats for the new record.
_details->incrementStats(txn, -(recOld->netLength()), -1);
wunit.commit();
}
// The extent must now be empty.
invariant(sourceExtent->firstRecord.isNull());
invariant(sourceExtent->lastRecord.isNull());
// We are still the first extent, but we must not be the only extent.
invariant( _details->firstExtent(txn) == extentLoc );
invariant( _details->lastExtent(txn) != extentLoc );
// Remove the newly emptied sourceExtent from the extent linked list and return it to
// the extent manager.
WriteUnitOfWork wunit(txn);
const DiskLoc newFirst = sourceExtent->xnext;
_details->setFirstExtent( txn, newFirst );
*txn->recoveryUnit()->writing(&_extentManager->getExtent( newFirst )->xprev) = DiskLoc();
_extentManager->freeExtent( txn, extentLoc );
wunit.commit();
{
const double oldPadding = oldObjSize ? double(oldObjSizeWithPadding) / oldObjSize
: 1.0; // defining 0/0 as 1 for this.
log() << "compact finished extent #" << extentNumber << " containing " << nrecords
<< " documents (" << totalNetSize / (1024*1024.0) << "MB)"
<< " oldPadding: " << oldPadding;
}
}
}
void HeapRecordStore::deleteRecord(OperationContext* txn, const DiskLoc& loc) {
HeapRecord* rec = recordFor(loc);
_dataSize -= rec->netLength();
invariant(_records.erase(loc) == 1);
}
DiskLoc ExtentManager::extentLocFor( const DiskLoc& loc ) const {
Record* record = recordFor( loc );
return DiskLoc( loc.a(), record->extentOfs() );
}
void HeapRecordStore::deleteRecord(OperationContext* txn, const DiskLoc& loc) {
HeapRecord* rec = recordFor(loc);
txn->recoveryUnit()->registerChange(new RemoveChange(_data, loc, *rec));
_data->dataSize -= rec->size;
invariant(_data->records.erase(loc) == 1);
}
void SimpleRecordStoreV1::_compactExtent(OperationContext* txn,
const DiskLoc diskloc,
int extentNumber,
RecordStoreCompactAdaptor* adaptor,
const CompactOptions* compactOptions,
CompactStats* stats ) {
log() << "compact begin extent #" << extentNumber
<< " for namespace " << _ns << " " << diskloc;
unsigned oldObjSize = 0; // we'll report what the old padding was
unsigned oldObjSizeWithPadding = 0;
Extent *e = _extentManager->getExtent( diskloc );
e->assertOk();
fassert( 17437, e->validates(diskloc) );
{
// the next/prev pointers within the extent might not be in order so we first
// page the whole thing in sequentially
log() << "compact paging in len=" << e->length/1000000.0 << "MB" << endl;
Timer t;
size_t length = e->length;
touch_pages( reinterpret_cast<const char*>(e), length );
int ms = t.millis();
if( ms > 1000 )
log() << "compact end paging in " << ms << "ms "
<< e->length/1000000.0/t.seconds() << "MB/sec" << endl;
}
{
log() << "compact copying records" << endl;
long long datasize = 0;
long long nrecords = 0;
DiskLoc L = e->firstRecord;
if( !L.isNull() ) {
while( 1 ) {
Record *recOld = recordFor(L);
L = getNextRecordInExtent(L);
if ( compactOptions->validateDocuments && !adaptor->isDataValid(recOld) ) {
// object is corrupt!
log() << "compact skipping corrupt document!";
stats->corruptDocuments++;
}
else {
unsigned dataSize = adaptor->dataSize( recOld );
unsigned docSize = dataSize;
nrecords++;
oldObjSize += docSize;
oldObjSizeWithPadding += recOld->netLength();
unsigned lenWHdr = docSize + Record::HeaderSize;
unsigned lenWPadding = lenWHdr;
switch( compactOptions->paddingMode ) {
case CompactOptions::NONE:
if ( _details->isUserFlagSet(Flag_UsePowerOf2Sizes) )
lenWPadding = quantizePowerOf2AllocationSpace(lenWPadding);
break;
case CompactOptions::PRESERVE:
// if we are preserving the padding, the record should not change size
lenWPadding = recOld->lengthWithHeaders();
break;
case CompactOptions::MANUAL:
lenWPadding = compactOptions->computeRecordSize(lenWPadding);
if (lenWPadding < lenWHdr || lenWPadding > BSONObjMaxUserSize / 2 ) {
lenWPadding = lenWHdr;
}
break;
}
CompactDocWriter writer( recOld, dataSize, lenWPadding );
StatusWith<DiskLoc> status = insertRecord( txn, &writer, 0 );
uassertStatusOK( status.getStatus() );
datasize += recordFor( status.getValue() )->netLength();
adaptor->inserted( recordFor( status.getValue() ), status.getValue() );
}
if( L.isNull() ) {
// we just did the very last record from the old extent. it's still pointed to
// by the old extent ext, but that will be fixed below after this loop
break;
}
// remove the old records (orphan them) periodically so our commit block doesn't get too large
bool stopping = false;
RARELY stopping = !txn->checkForInterruptNoAssert().isOK();
if( stopping || txn->recoveryUnit()->isCommitNeeded() ) {
*txn->recoveryUnit()->writing(&e->firstRecord) = L;
Record *r = recordFor(L);
txn->recoveryUnit()->writingInt(r->prevOfs()) = DiskLoc::NullOfs;
txn->recoveryUnit()->commitIfNeeded();
txn->checkForInterrupt();
}
}
} // if !L.isNull()
invariant( _details->firstExtent() == diskloc );
invariant( _details->lastExtent() != diskloc );
DiskLoc newFirst = e->xnext;
_details->setFirstExtent( txn, newFirst );
*txn->recoveryUnit()->writing(&_extentManager->getExtent( newFirst )->xprev) = DiskLoc();
_extentManager->freeExtent( txn, diskloc );
txn->recoveryUnit()->commitIfNeeded();
{
double op = 1.0;
if( oldObjSize )
op = static_cast<double>(oldObjSizeWithPadding)/oldObjSize;
log() << "compact finished extent #" << extentNumber << " containing " << nrecords
<< " documents (" << datasize/1000000.0 << "MB)"
<< " oldPadding: " << op << ' ' << static_cast<unsigned>(op*100.0)/100;
}
}
}
Status RecordStoreV1Base::validate( OperationContext* txn,
bool full, bool scanData,
ValidateAdaptor* adaptor,
ValidateResults* results, BSONObjBuilder* output ) const {
// 1) basic status that require no iteration
// 2) extent level info
// 3) check extent start and end
// 4) check each non-deleted record
// 5) check deleted list
// -------------
// 1111111111111111111
if ( isCapped() ){
output->appendBool("capped", true);
output->appendNumber("max", _details->maxCappedDocs());
}
output->appendNumber("datasize", _details->dataSize());
output->appendNumber("nrecords", _details->numRecords());
output->appendNumber("lastExtentSize", _details->lastExtentSize(txn));
output->appendNumber("padding", _details->paddingFactor());
if ( _details->firstExtent(txn).isNull() )
output->append( "firstExtent", "null" );
else
output->append( "firstExtent",
str::stream() << _details->firstExtent(txn).toString()
<< " ns:"
<< _getExtent( txn, _details->firstExtent(txn) )->nsDiagnostic.toString());
if ( _details->lastExtent(txn).isNull() )
output->append( "lastExtent", "null" );
else
output->append( "lastExtent", str::stream() << _details->lastExtent(txn).toString()
<< " ns:"
<< _getExtent( txn, _details->lastExtent(txn) )->nsDiagnostic.toString());
// 22222222222222222222222222
{ // validate extent basics
BSONArrayBuilder extentData;
int extentCount = 0;
DiskLoc extentDiskLoc;
try {
if ( !_details->firstExtent(txn).isNull() ) {
_getExtent( txn, _details->firstExtent(txn) )->assertOk();
_getExtent( txn, _details->lastExtent(txn) )->assertOk();
}
extentDiskLoc = _details->firstExtent(txn);
while (!extentDiskLoc.isNull()) {
Extent* thisExtent = _getExtent( txn, extentDiskLoc );
if (full) {
extentData << thisExtent->dump();
}
if (!thisExtent->validates(extentDiskLoc, &results->errors)) {
results->valid = false;
}
DiskLoc nextDiskLoc = thisExtent->xnext;
if (extentCount > 0 && !nextDiskLoc.isNull()
&& _getExtent( txn, nextDiskLoc )->xprev != extentDiskLoc) {
StringBuilder sb;
sb << "'xprev' pointer " << _getExtent( txn, nextDiskLoc )->xprev.toString()
<< " in extent " << nextDiskLoc.toString()
<< " does not point to extent " << extentDiskLoc.toString();
results->errors.push_back( sb.str() );
results->valid = false;
}
if (nextDiskLoc.isNull() && extentDiskLoc != _details->lastExtent(txn)) {
StringBuilder sb;
sb << "'lastExtent' pointer " << _details->lastExtent(txn).toString()
<< " does not point to last extent in list " << extentDiskLoc.toString();
results->errors.push_back( sb.str() );
results->valid = false;
}
extentDiskLoc = nextDiskLoc;
extentCount++;
txn->checkForInterrupt();
}
}
catch (const DBException& e) {
StringBuilder sb;
sb << "exception validating extent " << extentCount
<< ": " << e.what();
results->errors.push_back( sb.str() );
results->valid = false;
return Status::OK();
}
output->append("extentCount", extentCount);
if ( full )
output->appendArray( "extents" , extentData.arr() );
}
try {
// 333333333333333333333333333
bool testingLastExtent = false;
try {
DiskLoc firstExtentLoc = _details->firstExtent(txn);
if (firstExtentLoc.isNull()) {
// this is ok
}
else {
output->append("firstExtentDetails", _getExtent(txn, firstExtentLoc)->dump());
if (!_getExtent(txn, firstExtentLoc)->xprev.isNull()) {
StringBuilder sb;
sb << "'xprev' pointer in 'firstExtent' " << _details->firstExtent(txn).toString()
<< " is " << _getExtent(txn, firstExtentLoc)->xprev.toString()
<< ", should be null";
results->errors.push_back( sb.str() );
results->valid = false;
}
}
testingLastExtent = true;
DiskLoc lastExtentLoc = _details->lastExtent(txn);
if (lastExtentLoc.isNull()) {
// this is ok
}
else {
if (firstExtentLoc != lastExtentLoc) {
output->append("lastExtentDetails", _getExtent(txn, lastExtentLoc)->dump());
if (!_getExtent(txn, lastExtentLoc)->xnext.isNull()) {
StringBuilder sb;
sb << "'xnext' pointer in 'lastExtent' " << lastExtentLoc.toString()
<< " is " << _getExtent(txn, lastExtentLoc)->xnext.toString()
<< ", should be null";
results->errors.push_back( sb.str() );
results->valid = false;
}
}
}
}
catch (const DBException& e) {
StringBuilder sb;
sb << "exception processing '"
<< (testingLastExtent ? "lastExtent" : "firstExtent")
<< "': " << e.what();
results->errors.push_back( sb.str() );
results->valid = false;
}
// 4444444444444444444444444
set<DiskLoc> recs;
if( scanData ) {
int n = 0;
int nInvalid = 0;
long long nQuantizedSize = 0;
long long nPowerOf2QuantizedSize = 0;
long long len = 0;
long long nlen = 0;
long long bsonLen = 0;
int outOfOrder = 0;
DiskLoc cl_last;
scoped_ptr<RecordIterator> iterator( getIterator( txn,
DiskLoc(),
false,
CollectionScanParams::FORWARD ) );
DiskLoc cl;
while ( !( cl = iterator->getNext() ).isNull() ) {
n++;
if ( n < 1000000 )
recs.insert(cl);
if ( isCapped() ) {
if ( cl < cl_last )
outOfOrder++;
cl_last = cl;
}
Record *r = recordFor(cl);
len += r->lengthWithHeaders();
nlen += r->netLength();
if ( r->lengthWithHeaders() ==
quantizeAllocationSpace( r->lengthWithHeaders() ) ) {
// Count the number of records having a size consistent with
// the quantizeAllocationSpace quantization implementation.
++nQuantizedSize;
}
if ( r->lengthWithHeaders() ==
quantizePowerOf2AllocationSpace( r->lengthWithHeaders() ) ) {
// Count the number of records having a size consistent with the
// quantizePowerOf2AllocationSpace quantization implementation.
++nPowerOf2QuantizedSize;
}
if (full){
size_t dataSize = 0;
const Status status = adaptor->validate( r->toRecordData(), &dataSize );
if (!status.isOK()) {
results->valid = false;
if (nInvalid == 0) // only log once;
results->errors.push_back( "invalid object detected (see logs)" );
nInvalid++;
log() << "Invalid object detected in " << _ns
<< ": " << status.reason();
}
else {
bsonLen += dataSize;
}
}
}
if ( isCapped() && !_details->capLooped() ) {
output->append("cappedOutOfOrder", outOfOrder);
if ( outOfOrder > 1 ) {
results->valid = false;
results->errors.push_back( "too many out of order records" );
}
}
output->append("objectsFound", n);
if (full) {
output->append("invalidObjects", nInvalid);
}
output->appendNumber("nQuantizedSize", nQuantizedSize);
output->appendNumber("nPowerOf2QuantizedSize", nPowerOf2QuantizedSize);
output->appendNumber("bytesWithHeaders", len);
output->appendNumber("bytesWithoutHeaders", nlen);
if (full) {
output->appendNumber("bytesBson", bsonLen);
}
} // end scanData
// 55555555555555555555555555
BSONArrayBuilder deletedListArray;
for ( int i = 0; i < Buckets; i++ ) {
deletedListArray << _details->deletedListEntry(i).isNull();
}
int ndel = 0;
long long delSize = 0;
BSONArrayBuilder delBucketSizes;
int incorrect = 0;
for ( int i = 0; i < Buckets; i++ ) {
DiskLoc loc = _details->deletedListEntry(i);
try {
int k = 0;
while ( !loc.isNull() ) {
if ( recs.count(loc) )
incorrect++;
ndel++;
if ( loc.questionable() ) {
if( isCapped() && !loc.isValid() && i == 1 ) {
/* the constructor for NamespaceDetails intentionally sets deletedList[1] to invalid
see comments in namespace.h
*/
break;
}
string err( str::stream() << "bad pointer in deleted record list: "
<< loc.toString()
<< " bucket: " << i
<< " k: " << k );
results->errors.push_back( err );
results->valid = false;
break;
}
const DeletedRecord* d = deletedRecordFor(loc);
delSize += d->lengthWithHeaders();
loc = d->nextDeleted();
k++;
txn->checkForInterrupt();
}
delBucketSizes << k;
}
catch (...) {
results->errors.push_back( (string)"exception in deleted chain for bucket " +
BSONObjBuilder::numStr(i) );
results->valid = false;
}
}
output->appendNumber("deletedCount", ndel);
output->appendNumber("deletedSize", delSize);
if ( full ) {
output->append( "delBucketSizes", delBucketSizes.arr() );
}
if ( incorrect ) {
results->errors.push_back( BSONObjBuilder::numStr(incorrect) +
" records from datafile are in deleted list" );
results->valid = false;
}
}
catch (AssertionException) {
results->errors.push_back( "exception during validate" );
results->valid = false;
}
return Status::OK();
}