StatusWith<DiskLoc> RocksRecordStore::updateRecord( OperationContext* txn,
const DiskLoc& loc,
const char* data,
int len,
bool enforceQuota,
UpdateMoveNotifier* notifier ) {
RocksRecoveryUnit* ru = _getRecoveryUnit( txn );
std::string old_value;
// XXX Be sure to also first query the write batch once Facebook implements that
rocksdb::Status status = _db->Get( _readOptions( txn ),
_columnFamily,
_makeKey( loc ),
&old_value );
if ( !status.ok() ) {
return StatusWith<DiskLoc>( ErrorCodes::InternalError, status.ToString() );
}
int old_length = old_value.size();
ru->writeBatch()->Put( _columnFamily, _makeKey( loc ), rocksdb::Slice( data, len ) );
_increaseDataSize(txn, len - old_length);
cappedDeleteAsNeeded(txn);
return StatusWith<DiskLoc>( loc );
}
StatusWith<RecordId> WiredTigerRecordStore::updateRecord(OperationContext* txn,
const RecordId& loc,
const char* data,
int len,
bool enforceQuota,
UpdateNotifier* notifier) {
WiredTigerCursor curwrap(_uri, _tableId, true, txn);
curwrap.assertInActiveTxn();
WT_CURSOR* c = curwrap.get();
invariant(c);
c->set_key(c, _makeKey(loc));
int ret = WT_OP_CHECK(c->search(c));
invariantWTOK(ret);
WT_ITEM old_value;
ret = c->get_value(c, &old_value);
invariantWTOK(ret);
int old_length = old_value.size;
c->set_key(c, _makeKey(loc));
WiredTigerItem value(data, len);
c->set_value(c, value.Get());
ret = WT_OP_CHECK(c->insert(c));
invariantWTOK(ret);
_increaseDataSize(txn, len - old_length);
cappedDeleteAsNeeded(txn, loc);
return StatusWith<RecordId>(loc);
}
void RocksRecordStore::deleteRecord( OperationContext* txn, const DiskLoc& dl ) {
RocksRecoveryUnit* ru = _getRecoveryUnit( txn );
std::string oldValue;
_db->Get( _readOptions( txn ), _columnFamily, _makeKey( dl ), &oldValue );
int oldLength = oldValue.size();
ru->writeBatch()->Delete( _columnFamily, _makeKey( dl ) );
_changeNumRecords(txn, false);
_increaseDataSize(txn, -oldLength);
}
StatusWith<RecordId> WiredTigerRecordStore::insertRecord(OperationContext* txn,
const char* data,
int len,
bool enforceQuota) {
if (_isCapped && len > _cappedMaxSize) {
return StatusWith<RecordId>(ErrorCodes::BadValue, "object to insert exceeds cappedMaxSize");
}
RecordId loc;
if (_useOplogHack) {
StatusWith<RecordId> status = extractAndCheckLocForOplog(data, len);
if (!status.isOK())
return status;
loc = status.getValue();
if (loc > _oplog_highestSeen) {
stdx::lock_guard<stdx::mutex> lk(_uncommittedDiskLocsMutex);
if (loc > _oplog_highestSeen) {
_oplog_highestSeen = loc;
}
}
} else if (_isCapped) {
stdx::lock_guard<stdx::mutex> lk(_uncommittedDiskLocsMutex);
loc = _nextId();
_addUncommitedDiskLoc_inlock(txn, loc);
} else {
loc = _nextId();
}
WiredTigerCursor curwrap(_uri, _tableId, true, txn);
curwrap.assertInActiveTxn();
WT_CURSOR* c = curwrap.get();
invariant(c);
c->set_key(c, _makeKey(loc));
WiredTigerItem value(data, len);
c->set_value(c, value.Get());
int ret = WT_OP_CHECK(c->insert(c));
if (ret) {
return StatusWith<RecordId>(wtRCToStatus(ret, "WiredTigerRecordStore::insertRecord"));
}
_changeNumRecords(txn, 1);
_increaseDataSize(txn, len);
cappedDeleteAsNeeded(txn, loc);
return StatusWith<RecordId>(loc);
}
Status WiredTigerRecordStore::truncate(OperationContext* txn) {
WiredTigerCursor startWrap(_uri, _tableId, true, txn);
WT_CURSOR* start = startWrap.get();
int ret = WT_OP_CHECK(start->next(start));
// Empty collections don't have anything to truncate.
if (ret == WT_NOTFOUND) {
return Status::OK();
}
invariantWTOK(ret);
WT_SESSION* session = WiredTigerRecoveryUnit::get(txn)->getSession(txn)->getSession();
invariantWTOK(WT_OP_CHECK(session->truncate(session, NULL, start, NULL, NULL)));
_changeNumRecords(txn, -numRecords(txn));
_increaseDataSize(txn, -dataSize(txn));
return Status::OK();
}
void WiredTigerRecordStore::deleteRecord(OperationContext* txn, const RecordId& loc) {
WiredTigerCursor cursor(_uri, _tableId, true, txn);
cursor.assertInActiveTxn();
WT_CURSOR* c = cursor.get();
c->set_key(c, _makeKey(loc));
int ret = WT_OP_CHECK(c->search(c));
invariantWTOK(ret);
WT_ITEM old_value;
ret = c->get_value(c, &old_value);
invariantWTOK(ret);
int old_length = old_value.size;
ret = WT_OP_CHECK(c->remove(c));
invariantWTOK(ret);
_changeNumRecords(txn, -1);
_increaseDataSize(txn, -old_length);
}
StatusWith<DiskLoc> RocksRecordStore::insertRecord( OperationContext* txn,
const char* data,
int len,
bool enforceQuota ) {
if ( _isCapped && len > _cappedMaxSize ) {
return StatusWith<DiskLoc>( ErrorCodes::BadValue,
"object to insert exceeds cappedMaxSize" );
}
RocksRecoveryUnit* ru = _getRecoveryUnit( txn );
DiskLoc loc = _nextId();
ru->writeBatch()->Put( _columnFamily, _makeKey( loc ), rocksdb::Slice( data, len ) );
_changeNumRecords( txn, true );
_increaseDataSize( txn, len );
cappedDeleteAsNeeded(txn);
return StatusWith<DiskLoc>( loc );
}
int64_t WiredTigerRecordStore::cappedDeleteAsNeeded_inlock(OperationContext* txn,
const RecordId& justInserted) {
// we do this is a side transaction in case it aborts
WiredTigerRecoveryUnit* realRecoveryUnit =
checked_cast<WiredTigerRecoveryUnit*>(txn->releaseRecoveryUnit());
invariant(realRecoveryUnit);
WiredTigerSessionCache* sc = realRecoveryUnit->getSessionCache();
OperationContext::RecoveryUnitState const realRUstate =
txn->setRecoveryUnit(new WiredTigerRecoveryUnit(sc), OperationContext::kNotInUnitOfWork);
WiredTigerRecoveryUnit::get(txn)->markNoTicketRequired(); // realRecoveryUnit already has
WT_SESSION* session = WiredTigerRecoveryUnit::get(txn)->getSession(txn)->getSession();
int64_t dataSize = _dataSize.load();
int64_t numRecords = _numRecords.load();
int64_t sizeOverCap = (dataSize > _cappedMaxSize) ? dataSize - _cappedMaxSize : 0;
int64_t sizeSaved = 0;
int64_t docsOverCap = 0, docsRemoved = 0;
if (_cappedMaxDocs != -1 && numRecords > _cappedMaxDocs)
docsOverCap = numRecords - _cappedMaxDocs;
try {
WriteUnitOfWork wuow(txn);
WiredTigerCursor curwrap(_uri, _tableId, true, txn);
WT_CURSOR* c = curwrap.get();
RecordId newestOld;
int ret = 0;
while ((sizeSaved < sizeOverCap || docsRemoved < docsOverCap) && (docsRemoved < 20000) &&
(ret = WT_OP_CHECK(c->next(c))) == 0) {
int64_t key;
ret = c->get_key(c, &key);
invariantWTOK(ret);
// don't go past the record we just inserted
newestOld = _fromKey(key);
if (newestOld >= justInserted) // TODO: use oldest uncommitted instead
break;
if (_shuttingDown)
break;
WT_ITEM old_value;
invariantWTOK(c->get_value(c, &old_value));
++docsRemoved;
sizeSaved += old_value.size;
if (_cappedDeleteCallback) {
uassertStatusOK(_cappedDeleteCallback->aboutToDeleteCapped(
txn,
newestOld,
RecordData(static_cast<const char*>(old_value.data), old_value.size)));
}
}
if (ret != WT_NOTFOUND) {
invariantWTOK(ret);
}
if (docsRemoved > 0) {
// if we scanned to the end of the collection or past our insert, go back one
if (ret == WT_NOTFOUND || newestOld >= justInserted) {
ret = WT_OP_CHECK(c->prev(c));
}
invariantWTOK(ret);
WiredTigerCursor startWrap(_uri, _tableId, true, txn);
WT_CURSOR* start = startWrap.get();
ret = WT_OP_CHECK(start->next(start));
invariantWTOK(ret);
ret = session->truncate(session, NULL, start, c, NULL);
if (ret == ENOENT || ret == WT_NOTFOUND) {
// TODO we should remove this case once SERVER-17141 is resolved
log() << "Soft failure truncating capped collection. Will try again later.";
docsRemoved = 0;
} else {
invariantWTOK(ret);
_changeNumRecords(txn, -docsRemoved);
_increaseDataSize(txn, -sizeSaved);
wuow.commit();
}
}
} catch (const WriteConflictException& wce) {
delete txn->releaseRecoveryUnit();
txn->setRecoveryUnit(realRecoveryUnit, realRUstate);
log() << "got conflict truncating capped, ignoring";
return 0;
} catch (...) {
delete txn->releaseRecoveryUnit();
txn->setRecoveryUnit(realRecoveryUnit, realRUstate);
throw;
}
delete txn->releaseRecoveryUnit();
txn->setRecoveryUnit(realRecoveryUnit, realRUstate);
return docsRemoved;
}
