PlanStage::StageState MultiIteratorStage::work(WorkingSetID* out) {
if (_collection == NULL)
return PlanStage::DEAD;
// The RecordId we're about to look at it might not be in memory. In this case
// we request a yield while we fetch the document.
if (!_iterators.empty()) {
RecordId curr = _iterators.back()->curr();
if (!curr.isNull()) {
std::auto_ptr<RecordFetcher> fetcher(_collection->documentNeedsFetch(_txn, curr));
if (NULL != fetcher.get()) {
WorkingSetMember* member = _ws->get(_wsidForFetch);
member->loc = curr;
// Pass the RecordFetcher off to the WSM on which we're performing the fetch.
member->setFetcher(fetcher.release());
*out = _wsidForFetch;
return NEED_FETCH;
}
}
}
RecordId next = _advance();
if (next.isNull())
return PlanStage::IS_EOF;
*out = _ws->allocate();
WorkingSetMember* member = _ws->get(*out);
member->loc = next;
member->obj = _collection->docFor(_txn, next);
member->state = WorkingSetMember::LOC_AND_OBJ;
return PlanStage::ADVANCED;
}
PlanStage::StageState OplogStart::workExtentHopping(WorkingSetID* out) {
if (_done || _subIterators.empty()) {
return PlanStage::IS_EOF;
}
// we work from the back to the front since the back has the newest data.
const RecordId loc = _subIterators.back()->curr();
if (loc.isNull()) return PlanStage::NEED_TIME;
// TODO: should we ever try and return NEED_FETCH here?
const BSONObj obj = _subIterators.back()->dataFor(loc).releaseToBson();
if (!_filter->matchesBSON(obj)) {
_done = true;
WorkingSetID id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->loc = loc;
member->obj = Snapshotted<BSONObj>(_txn->recoveryUnit()->getSnapshotId(), obj);
member->state = WorkingSetMember::LOC_AND_UNOWNED_OBJ;
*out = id;
return PlanStage::ADVANCED;
}
_subIterators.popAndDeleteBack();
return PlanStage::NEED_TIME;
}
bool Helpers::findById(OperationContext* opCtx,
Database* database,
StringData ns,
BSONObj query,
BSONObj& result,
bool* nsFound,
bool* indexFound) {
invariant(database);
Collection* collection = database->getCollection(opCtx, ns);
if (!collection) {
return false;
}
if (nsFound)
*nsFound = true;
IndexCatalog* catalog = collection->getIndexCatalog();
const IndexDescriptor* desc = catalog->findIdIndex(opCtx);
if (!desc)
return false;
if (indexFound)
*indexFound = 1;
RecordId loc = catalog->getIndex(desc)->findSingle(opCtx, query["_id"].wrap());
if (loc.isNull())
return false;
result = collection->docFor(opCtx, loc).value();
return true;
}
boost::optional<Record> next() final {
if (_eof)
return {};
WT_CURSOR* c = _cursor->get();
bool mustAdvance = true;
if (_lastReturnedId.isNull() && !_forward && _rs._isCapped) {
// In this case we need to seek to the highest visible record.
const RecordId reverseCappedInitialSeekPoint =
_readUntilForOplog.isNull() ? _rs.lowestCappedHiddenRecord() : _readUntilForOplog;
if (!reverseCappedInitialSeekPoint.isNull()) {
c->set_key(c, _makeKey(reverseCappedInitialSeekPoint));
int cmp;
int seekRet = WT_OP_CHECK(c->search_near(c, &cmp));
if (seekRet == WT_NOTFOUND) {
_eof = true;
return {};
}
invariantWTOK(seekRet);
// If we landed at or past the lowest hidden record, we must advance to be in
// the visible range.
mustAdvance = _rs.isCappedHidden(reverseCappedInitialSeekPoint)
? (cmp >= 0)
: (cmp > 0); // No longer hidden.
}
}
if (mustAdvance) {
// Nothing after the next line can throw WCEs.
// Note that an unpositioned (or eof) WT_CURSOR returns the first/last entry in the
// table when you call next/prev.
int advanceRet = WT_OP_CHECK(_forward ? c->next(c) : c->prev(c));
if (advanceRet == WT_NOTFOUND) {
_eof = true;
return {};
}
invariantWTOK(advanceRet);
}
int64_t key;
invariantWTOK(c->get_key(c, &key));
const RecordId id = _fromKey(key);
if (!isVisible(id)) {
_eof = true;
return {};
}
WT_ITEM value;
invariantWTOK(c->get_value(c, &value));
_lastReturnedId = id;
return {{id, {static_cast<const char*>(value.data), static_cast<int>(value.size)}}};
}
int nExtents() const {
int count = 0;
for ( RecordId extLoc = nsd()->firstExtent();
!extLoc.isNull();
extLoc = extentManager()->getExtent(extLoc)->xnext ) {
++count;
}
return count;
}
RecordId MultiIteratorStage::_advance() {
while (!_iterators.empty()) {
RecordId out = _iterators.back()->getNext();
if (!out.isNull())
return out;
_iterators.popAndDeleteBack();
}
return RecordId();
}
/* fetch a single object from collection ns that matches query
set your db SavedContext first
*/
bool Helpers::findOne(OperationContext* txn,
Collection* collection,
const BSONObj &query,
BSONObj& result,
bool requireIndex) {
RecordId loc = findOne( txn, collection, query, requireIndex );
if ( loc.isNull() )
return false;
result = collection->docFor(txn, loc).value();
return true;
}
void ReplicationRecoveryImpl::_truncateOplogTo(OperationContext* opCtx,
Timestamp truncateTimestamp) {
Timer timer;
const NamespaceString oplogNss(NamespaceString::kRsOplogNamespace);
AutoGetDb autoDb(opCtx, oplogNss.db(), MODE_IX);
Lock::CollectionLock oplogCollectionLoc(opCtx->lockState(), oplogNss.ns(), MODE_X);
Collection* oplogCollection = autoDb.getDb()->getCollection(opCtx, oplogNss);
if (!oplogCollection) {
fassertFailedWithStatusNoTrace(
34418,
Status(ErrorCodes::NamespaceNotFound,
str::stream() << "Can't find " << NamespaceString::kRsOplogNamespace.ns()));
}
// Scan through oplog in reverse, from latest entry to first, to find the truncateTimestamp.
RecordId oldestIDToDelete; // Non-null if there is something to delete.
auto oplogRs = oplogCollection->getRecordStore();
auto oplogReverseCursor = oplogRs->getCursor(opCtx, /*forward=*/false);
size_t count = 0;
while (auto next = oplogReverseCursor->next()) {
const BSONObj entry = next->data.releaseToBson();
const RecordId id = next->id;
count++;
const auto tsElem = entry["ts"];
if (count == 1) {
if (tsElem.eoo())
LOG(2) << "Oplog tail entry: " << redact(entry);
else
LOG(2) << "Oplog tail entry ts field: " << tsElem;
}
if (tsElem.timestamp() < truncateTimestamp) {
// If count == 1, that means that we have nothing to delete because everything in the
// oplog is < truncateTimestamp.
if (count != 1) {
invariant(!oldestIDToDelete.isNull());
oplogCollection->cappedTruncateAfter(opCtx, oldestIDToDelete, /*inclusive=*/true);
}
log() << "Replication recovery oplog truncation finished in: " << timer.millis()
<< "ms";
return;
}
oldestIDToDelete = id;
}
severe() << "Reached end of oplog looking for oplog entry before " << truncateTimestamp.toBSON()
<< " but couldn't find any after looking through " << count << " entries.";
fassertFailedNoTrace(40296);
}
PlanStage::StageState IDHackStage::doWork(WorkingSetID* out) {
if (_done) {
return PlanStage::IS_EOF;
}
WorkingSetID id = WorkingSet::INVALID_ID;
try {
// Look up the key by going directly to the index.
RecordId recordId = indexAccessMethod()->findSingle(getOpCtx(), _key);
// Key not found.
if (recordId.isNull()) {
_done = true;
return PlanStage::IS_EOF;
}
++_specificStats.keysExamined;
++_specificStats.docsExamined;
// Create a new WSM for the result document.
id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->recordId = recordId;
_workingSet->transitionToRecordIdAndIdx(id);
if (!_recordCursor)
_recordCursor = collection()->getCursor(getOpCtx());
// Find the document associated with 'id' in the collection's record store.
if (!WorkingSetCommon::fetch(getOpCtx(), _workingSet, id, _recordCursor)) {
// We didn't find a document with RecordId 'id'.
_workingSet->free(id);
_commonStats.isEOF = true;
_done = true;
return IS_EOF;
}
return advance(id, member, out);
} catch (const WriteConflictException&) {
// Restart at the beginning on retry.
_recordCursor.reset();
if (id != WorkingSet::INVALID_ID)
_workingSet->free(id);
*out = WorkingSet::INVALID_ID;
return NEED_YIELD;
}
}
bool TerarkDbRecordStore::findRecord(OperationContext* txn,
const RecordId& id,
RecordData* out) const {
if (id.isNull())
return false;
llong recIdx = id.repr() - 1;
CompositeTable* tab = m_table->m_tab.get();
auto& td = m_table->getMyThreadData();
tab->getValue(recIdx, &td.m_buf, &*td.m_dbCtx);
SharedBuffer bson = td.m_coder.decode(&tab->rowSchema(), td.m_buf);
// size_t bufsize = sizeof(SharedBuffer::Holder) + bson.objsize();
int bufsize = ConstDataView(bson.get()).read<LittleEndian<int>>();
*out = RecordData(bson, bufsize);
return true;
}
int nRecords() const {
int count = 0;
const Extent* ext;
for ( RecordId extLoc = nsd()->firstExtent();
!extLoc.isNull();
extLoc = ext->xnext) {
ext = extentManager()->getExtent(extLoc);
int fileNo = ext->firstRecord.a();
if ( fileNo == -1 )
continue;
for ( int recOfs = ext->firstRecord.getOfs(); recOfs != RecordId::NullOfs;
recOfs = recordStore()->recordFor(RecordId(fileNo, recOfs))->nextOfs() ) {
++count;
}
}
ASSERT_EQUALS( count, nsd()->numRecords() );
return count;
}
InMemoryRecordIterator::InMemoryRecordIterator(OperationContext* txn,
const InMemoryRecordStore::Records& records,
const InMemoryRecordStore& rs,
RecordId start,
bool tailable)
: _txn(txn),
_tailable(tailable),
_lastLoc(RecordId::min()),
_killedByInvalidate(false),
_records(records),
_rs(rs) {
if (start.isNull()) {
_it = _records.begin();
}
else {
_it = _records.find(start);
invariant(_it != _records.end());
}
}
CollectionOptions MMAPV1DatabaseCatalogEntry::getCollectionOptions(OperationContext* txn,
RecordId rid) const {
CollectionOptions options;
if (rid.isNull()) {
return options;
}
RecordStoreV1Base* rs = _getNamespaceRecordStore();
invariant(rs);
RecordData data;
invariant(rs->findRecord(txn, rid, &data));
if (data.releaseToBson()["options"].isABSONObj()) {
Status status = options.parse(data.releaseToBson()["options"].Obj());
fassert(18523, status);
}
return options;
}
InMemoryRecordReverseIterator::InMemoryRecordReverseIterator(
OperationContext* txn,
const InMemoryRecordStore::Records& records,
const InMemoryRecordStore& rs,
RecordId start) : _txn(txn),
_killedByInvalidate(false),
_records(records),
_rs(rs) {
if (start.isNull()) {
_it = _records.rbegin();
}
else {
// The reverse iterator will point to the preceding element, so we
// increment the base iterator to make it point past the found element
InMemoryRecordStore::Records::const_iterator baseIt(++_records.find(start));
_it = InMemoryRecordStore::Records::const_reverse_iterator(baseIt);
invariant(_it != _records.rend());
}
}
void run() {
create();
nsd()->deletedListEntry( 2 ) = nsd()->cappedListOfAllDeletedRecords().drec()->nextDeleted().drec()->nextDeleted();
nsd()->cappedListOfAllDeletedRecords().drec()->nextDeleted().drec()->nextDeleted().writing() = RecordId();
nsd()->cappedLastDelRecLastExtent().Null();
NamespaceDetails *d = nsd();
zero( &d->capExtent() );
zero( &d->capFirstNewRecord() );
// this has a side effect of called NamespaceDetails::cappedCheckMigrate
db()->namespaceIndex().details( ns() );
ASSERT( nsd()->firstExtent() == nsd()->capExtent() );
ASSERT( nsd()->capExtent().getOfs() != 0 );
ASSERT( !nsd()->capFirstNewRecord().isValid() );
int nDeleted = 0;
for ( RecordId i = nsd()->cappedListOfAllDeletedRecords(); !i.isNull(); i = i.drec()->nextDeleted(), ++nDeleted );
ASSERT_EQUALS( 10, nDeleted );
ASSERT( nsd()->cappedLastDelRecLastExtent().isNull() );
}
PlanStage::StageState OplogStart::workExtentHopping(WorkingSetID* out) {
if (_done || _subIterators.empty()) {
return PlanStage::IS_EOF;
}
// we work from the back to the front since the back has the newest data.
const RecordId loc = _subIterators.back()->getNext();
_subIterators.popAndDeleteBack();
// TODO: should we ever try and return NEED_FETCH here?
if (!loc.isNull() && !_filter->matchesBSON(_collection->docFor(_txn, loc))) {
_done = true;
WorkingSetID id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->loc = loc;
member->obj = _collection->docFor(_txn, member->loc);
member->state = WorkingSetMember::LOC_AND_UNOWNED_OBJ;
*out = id;
return PlanStage::ADVANCED;
}
return PlanStage::NEED_TIME;
}
bool Helpers::findById(OperationContext* txn,
Database* database,
const char *ns,
BSONObj query,
BSONObj& result,
bool* nsFound,
bool* indexFound) {
invariant(database);
Collection* collection = database->getCollection( ns );
if ( !collection ) {
return false;
}
if ( nsFound )
*nsFound = true;
IndexCatalog* catalog = collection->getIndexCatalog();
const IndexDescriptor* desc = catalog->findIdIndex( txn );
if ( !desc )
return false;
if ( indexFound )
*indexFound = 1;
// See SERVER-12397. This may not always be true.
BtreeBasedAccessMethod* accessMethod =
static_cast<BtreeBasedAccessMethod*>(catalog->getIndex( desc ));
RecordId loc = accessMethod->findSingle( txn, query["_id"].wrap() );
if ( loc.isNull() )
return false;
result = collection->docFor(txn, loc).value();
return true;
}
PlanStage::StageState IDHackStage::doWork(WorkingSetID* out) {
if (_done) {
return PlanStage::IS_EOF;
}
if (WorkingSet::INVALID_ID != _idBeingPagedIn) {
invariant(_recordCursor);
WorkingSetID id = _idBeingPagedIn;
_idBeingPagedIn = WorkingSet::INVALID_ID;
invariant(WorkingSetCommon::fetchIfUnfetched(getOpCtx(), _workingSet, id, _recordCursor));
WorkingSetMember* member = _workingSet->get(id);
return advance(id, member, out);
}
WorkingSetID id = WorkingSet::INVALID_ID;
try {
// Look up the key by going directly to the index.
RecordId recordId = _accessMethod->findSingle(getOpCtx(), _key);
// Key not found.
if (recordId.isNull()) {
_done = true;
return PlanStage::IS_EOF;
}
++_specificStats.keysExamined;
++_specificStats.docsExamined;
// Create a new WSM for the result document.
id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->recordId = recordId;
_workingSet->transitionToRecordIdAndIdx(id);
if (!_recordCursor)
_recordCursor = _collection->getCursor(getOpCtx());
// We may need to request a yield while we fetch the document.
if (auto fetcher = _recordCursor->fetcherForId(recordId)) {
// There's something to fetch. Hand the fetcher off to the WSM, and pass up a
// fetch request.
_idBeingPagedIn = id;
member->setFetcher(fetcher.release());
*out = id;
return NEED_YIELD;
}
// The doc was already in memory, so we go ahead and return it.
if (!WorkingSetCommon::fetch(getOpCtx(), _workingSet, id, _recordCursor)) {
// _id is immutable so the index would return the only record that could
// possibly match the query.
_workingSet->free(id);
_commonStats.isEOF = true;
_done = true;
return IS_EOF;
}
return advance(id, member, out);
} catch (const WriteConflictException& wce) {
// Restart at the beginning on retry.
_recordCursor.reset();
if (id != WorkingSet::INVALID_ID)
_workingSet->free(id);
*out = WorkingSet::INVALID_ID;
return NEED_YIELD;
}
}
void syncFixUp(OperationContext* txn,
FixUpInfo& fixUpInfo,
OplogReader* oplogreader,
ReplicationCoordinator* replCoord) {
DBClientConnection* them = oplogreader->conn();
// fetch all first so we needn't handle interruption in a fancy way
unsigned long long totalSize = 0;
list< pair<DocID, BSONObj> > goodVersions;
BSONObj newMinValid;
// fetch all the goodVersions of each document from current primary
DocID doc;
unsigned long long numFetched = 0;
try {
for (set<DocID>::iterator it = fixUpInfo.toRefetch.begin();
it != fixUpInfo.toRefetch.end();
it++) {
doc = *it;
verify(!doc._id.eoo());
{
// TODO : slow. lots of round trips.
numFetched++;
BSONObj good = them->findOne(doc.ns, doc._id.wrap(),
NULL, QueryOption_SlaveOk).getOwned();
totalSize += good.objsize();
uassert(13410, "replSet too much data to roll back",
totalSize < 300 * 1024 * 1024);
// note good might be eoo, indicating we should delete it
goodVersions.push_back(pair<DocID, BSONObj>(doc,good));
}
}
newMinValid = oplogreader->getLastOp(rsOplogName);
if (newMinValid.isEmpty()) {
error() << "rollback error newMinValid empty?";
return;
}
}
catch (DBException& e) {
LOG(1) << "rollback re-get objects: " << e.toString();
error() << "rollback couldn't re-get ns:" << doc.ns << " _id:" << doc._id << ' '
<< numFetched << '/' << fixUpInfo.toRefetch.size();
throw e;
}
log() << "rollback 3.5";
if (fixUpInfo.rbid != getRBID(oplogreader->conn())) {
// our source rolled back itself. so the data we received isn't necessarily consistent.
warning() << "rollback rbid on source changed during rollback, cancelling this attempt";
return;
}
// update them
log() << "rollback 4 n:" << goodVersions.size();
bool warn = false;
invariant(!fixUpInfo.commonPointOurDiskloc.isNull());
invariant(txn->lockState()->isW());
// we have items we are writing that aren't from a point-in-time. thus best not to come
// online until we get to that point in freshness.
Timestamp minValid = newMinValid["ts"].timestamp();
log() << "minvalid=" << minValid.toStringLong();
setMinValid(txn, minValid);
// any full collection resyncs required?
if (!fixUpInfo.collectionsToResyncData.empty()
|| !fixUpInfo.collectionsToResyncMetadata.empty()) {
for (const string& ns : fixUpInfo.collectionsToResyncData) {
log() << "rollback 4.1.1 coll resync " << ns;
fixUpInfo.collectionsToResyncMetadata.erase(ns);
const NamespaceString nss(ns);
Database* db = dbHolder().openDb(txn, nss.db().toString());
invariant(db);
{
WriteUnitOfWork wunit(txn);
db->dropCollection(txn, ns);
wunit.commit();
}
{
string errmsg;
// This comes as a GlobalWrite lock, so there is no DB to be acquired after
// resume, so we can skip the DB stability checks. Also
// copyCollectionFromRemote will acquire its own database pointer, under the
// appropriate locks, so just releasing and acquiring the lock is safe.
invariant(txn->lockState()->isW());
Lock::TempRelease release(txn->lockState());
bool ok = copyCollectionFromRemote(txn, them->getServerAddress(), ns, errmsg);
uassert(15909, str::stream() << "replSet rollback error resyncing collection "
<< ns << ' ' << errmsg, ok);
}
}
for (const string& ns : fixUpInfo.collectionsToResyncMetadata) {
log() << "rollback 4.1.2 coll metadata resync " << ns;
const NamespaceString nss(ns);
auto db = dbHolder().openDb(txn, nss.db().toString());
invariant(db);
auto collection = db->getCollection(ns);
invariant(collection);
auto cce = collection->getCatalogEntry();
const std::list<BSONObj> info =
them->getCollectionInfos(nss.db().toString(), BSON("name" << nss.coll()));
if (info.empty()) {
// Collection dropped by "them" so we should drop it too.
log() << ns << " not found on remote host, dropping";
fixUpInfo.toDrop.insert(ns);
continue;
}
invariant(info.size() == 1);
CollectionOptions options;
auto status = options.parse(info.front());
if (!status.isOK()) {
throw RSFatalException(str::stream() << "Failed to parse options "
<< info.front() << ": "
<< status.toString());
}
WriteUnitOfWork wuow(txn);
if (options.flagsSet || cce->getCollectionOptions(txn).flagsSet) {
cce->updateFlags(txn, options.flags);
}
status = collection->setValidator(txn, options.validator);
if (!status.isOK()) {
throw RSFatalException(str::stream() << "Failed to set validator: "
<< status.toString());
}
wuow.commit();
}
// we did more reading from primary, so check it again for a rollback (which would mess
// us up), and make minValid newer.
log() << "rollback 4.2";
string err;
try {
newMinValid = oplogreader->getLastOp(rsOplogName);
if (newMinValid.isEmpty()) {
err = "can't get minvalid from sync source";
}
else {
Timestamp minValid = newMinValid["ts"].timestamp();
log() << "minvalid=" << minValid.toStringLong();
setMinValid(txn, minValid);
}
}
catch (DBException& e) {
err = "can't get/set minvalid: ";
err += e.what();
}
if (fixUpInfo.rbid != getRBID(oplogreader->conn())) {
// our source rolled back itself. so the data we received isn't necessarily
// consistent. however, we've now done writes. thus we have a problem.
err += "rbid at primary changed during resync/rollback";
}
if (!err.empty()) {
severe() << "rolling back : " << err
<< ". A full resync will be necessary.";
// TODO: reset minvalid so that we are permanently in fatal state
// TODO: don't be fatal, but rather, get all the data first.
throw RSFatalException();
}
log() << "rollback 4.3";
}
map<string,shared_ptr<Helpers::RemoveSaver> > removeSavers;
log() << "rollback 4.6";
// drop collections to drop before doing individual fixups - that might make things faster
// below actually if there were subsequent inserts to rollback
for (set<string>::iterator it = fixUpInfo.toDrop.begin();
it != fixUpInfo.toDrop.end();
it++) {
log() << "rollback drop: " << *it;
Database* db = dbHolder().get(txn, nsToDatabaseSubstring(*it));
if (db) {
WriteUnitOfWork wunit(txn);
shared_ptr<Helpers::RemoveSaver>& removeSaver = removeSavers[*it];
if (!removeSaver)
removeSaver.reset(new Helpers::RemoveSaver("rollback", "", *it));
// perform a collection scan and write all documents in the collection to disk
boost::scoped_ptr<PlanExecutor> exec(
InternalPlanner::collectionScan(txn,
*it,
db->getCollection(*it)));
BSONObj curObj;
PlanExecutor::ExecState execState;
while (PlanExecutor::ADVANCED == (execState = exec->getNext(&curObj, NULL))) {
removeSaver->goingToDelete(curObj);
}
if (execState != PlanExecutor::IS_EOF) {
if (execState == PlanExecutor::FAILURE &&
WorkingSetCommon::isValidStatusMemberObject(curObj)) {
Status errorStatus = WorkingSetCommon::getMemberObjectStatus(curObj);
severe() << "rolling back createCollection on " << *it
<< " failed with " << errorStatus
<< ". A full resync is necessary.";
}
else {
severe() << "rolling back createCollection on " << *it
<< " failed. A full resync is necessary.";
}
throw RSFatalException();
}
db->dropCollection(txn, *it);
wunit.commit();
}
}
log() << "rollback 4.7";
OldClientContext ctx(txn, rsOplogName);
Collection* oplogCollection = ctx.db()->getCollection(rsOplogName);
uassert(13423,
str::stream() << "replSet error in rollback can't find " << rsOplogName,
oplogCollection);
unsigned deletes = 0, updates = 0;
time_t lastProgressUpdate = time(0);
time_t progressUpdateGap = 10;
for (list<pair<DocID, BSONObj> >::iterator it = goodVersions.begin();
it != goodVersions.end();
it++) {
time_t now = time(0);
if (now - lastProgressUpdate > progressUpdateGap) {
log() << deletes << " delete and "
<< updates << " update operations processed out of "
<< goodVersions.size() << " total operations";
lastProgressUpdate = now;
}
const DocID& doc = it->first;
BSONObj pattern = doc._id.wrap(); // { _id : ... }
try {
verify(doc.ns && *doc.ns);
if (fixUpInfo.collectionsToResyncData.count(doc.ns)) {
// we just synced this entire collection
continue;
}
// keep an archive of items rolled back
shared_ptr<Helpers::RemoveSaver>& removeSaver = removeSavers[doc.ns];
if (!removeSaver)
removeSaver.reset(new Helpers::RemoveSaver("rollback", "", doc.ns));
// todo: lots of overhead in context, this can be faster
OldClientContext ctx(txn, doc.ns);
// Add the doc to our rollback file
BSONObj obj;
Collection* collection = ctx.db()->getCollection(doc.ns);
// Do not log an error when undoing an insert on a no longer existent collection.
// It is likely that the collection was dropped as part of rolling back a
// createCollection command and regardless, the document no longer exists.
if (collection) {
bool found = Helpers::findOne(txn, collection, pattern, obj, false);
if (found) {
removeSaver->goingToDelete(obj);
}
else {
error() << "rollback cannot find object: " << pattern
<< " in namespace " << doc.ns;
}
}
if (it->second.isEmpty()) {
// wasn't on the primary; delete.
// TODO 1.6 : can't delete from a capped collection. need to handle that here.
deletes++;
if (collection) {
if (collection->isCapped()) {
// can't delete from a capped collection - so we truncate instead. if
// this item must go, so must all successors!!!
try {
// TODO: IIRC cappedTruncateAfter does not handle completely empty.
// this will crazy slow if no _id index.
long long start = Listener::getElapsedTimeMillis();
RecordId loc = Helpers::findOne(txn, collection, pattern, false);
if (Listener::getElapsedTimeMillis() - start > 200)
warning() << "roll back slow no _id index for "
<< doc.ns << " perhaps?";
// would be faster but requires index:
// RecordId loc = Helpers::findById(nsd, pattern);
if (!loc.isNull()) {
try {
collection->temp_cappedTruncateAfter(txn, loc, true);
}
catch (DBException& e) {
if (e.getCode() == 13415) {
// hack: need to just make cappedTruncate do this...
MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN {
WriteUnitOfWork wunit(txn);
uassertStatusOK(collection->truncate(txn));
wunit.commit();
} MONGO_WRITE_CONFLICT_RETRY_LOOP_END(
txn,
"truncate",
collection->ns().ns());
}
else {
throw e;
}
}
}
}
catch (DBException& e) {
error() << "rolling back capped collection rec "
<< doc.ns << ' ' << e.toString();
}
}
else {
PlanStage::StageState IDHackStage::work(WorkingSetID* out) {
++_commonStats.works;
// Adds the amount of time taken by work() to executionTimeMillis.
ScopedTimer timer(&_commonStats.executionTimeMillis);
if (_done) {
return PlanStage::IS_EOF;
}
if (WorkingSet::INVALID_ID != _idBeingPagedIn) {
invariant(_recordCursor);
WorkingSetID id = _idBeingPagedIn;
_idBeingPagedIn = WorkingSet::INVALID_ID;
WorkingSetMember* member = _workingSet->get(id);
invariant(WorkingSetCommon::fetchIfUnfetched(_txn, member, _recordCursor));
return advance(id, member, out);
}
WorkingSetID id = WorkingSet::INVALID_ID;
try {
// Use the index catalog to get the id index.
const IndexCatalog* catalog = _collection->getIndexCatalog();
// Find the index we use.
IndexDescriptor* idDesc = catalog->findIdIndex(_txn);
if (NULL == idDesc) {
_done = true;
return PlanStage::IS_EOF;
}
// Look up the key by going directly to the index.
RecordId loc = catalog->getIndex(idDesc)->findSingle(_txn, _key);
// Key not found.
if (loc.isNull()) {
_done = true;
return PlanStage::IS_EOF;
}
++_specificStats.keysExamined;
++_specificStats.docsExamined;
// Create a new WSM for the result document.
id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->state = WorkingSetMember::LOC_AND_IDX;
member->loc = loc;
if (!_recordCursor)
_recordCursor = _collection->getCursor(_txn);
// We may need to request a yield while we fetch the document.
if (auto fetcher = _recordCursor->fetcherForId(loc)) {
// There's something to fetch. Hand the fetcher off to the WSM, and pass up a
// fetch request.
_idBeingPagedIn = id;
member->setFetcher(fetcher.release());
*out = id;
_commonStats.needYield++;
return NEED_YIELD;
}
// The doc was already in memory, so we go ahead and return it.
if (!WorkingSetCommon::fetch(_txn, member, _recordCursor)) {
// _id is immutable so the index would return the only record that could
// possibly match the query.
_workingSet->free(id);
_commonStats.isEOF = true;
_done = true;
return IS_EOF;
}
return advance(id, member, out);
} catch (const WriteConflictException& wce) {
// Restart at the beginning on retry.
_recordCursor.reset();
if (id != WorkingSet::INVALID_ID)
_workingSet->free(id);
*out = WorkingSet::INVALID_ID;
_commonStats.needYield++;
return NEED_YIELD;
}
}
PlanStage::StageState IDHackStage::work(WorkingSetID* out) {
++_commonStats.works;
// Adds the amount of time taken by work() to executionTimeMillis.
ScopedTimer timer(&_commonStats.executionTimeMillis);
if (_done) { return PlanStage::IS_EOF; }
if (WorkingSet::INVALID_ID != _idBeingPagedIn) {
WorkingSetID id = _idBeingPagedIn;
_idBeingPagedIn = WorkingSet::INVALID_ID;
WorkingSetMember* member = _workingSet->get(id);
WorkingSetCommon::completeFetch(_txn, member, _collection);
return advance(id, member, out);
}
// Use the index catalog to get the id index.
const IndexCatalog* catalog = _collection->getIndexCatalog();
// Find the index we use.
IndexDescriptor* idDesc = catalog->findIdIndex(_txn);
if (NULL == idDesc) {
_done = true;
return PlanStage::IS_EOF;
}
// This may not be valid always. See SERVER-12397.
const BtreeBasedAccessMethod* accessMethod =
static_cast<const BtreeBasedAccessMethod*>(catalog->getIndex(idDesc));
// Look up the key by going directly to the Btree.
RecordId loc = accessMethod->findSingle(_txn, _key);
// Key not found.
if (loc.isNull()) {
_done = true;
return PlanStage::IS_EOF;
}
++_specificStats.keysExamined;
++_specificStats.docsExamined;
// Create a new WSM for the result document.
WorkingSetID id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->loc = loc;
member->state = WorkingSetMember::LOC_AND_UNOWNED_OBJ;
// We may need to request a yield while we fetch the document.
std::auto_ptr<RecordFetcher> fetcher(_collection->documentNeedsFetch(_txn, loc));
if (NULL != fetcher.get()) {
// There's something to fetch. Hand the fetcher off to the WSM, and pass up a
// fetch request.
_idBeingPagedIn = id;
member->setFetcher(fetcher.release());
*out = id;
_commonStats.needFetch++;
return NEED_FETCH;
}
// The doc was already in memory, so we go ahead and return it.
member->obj = _collection->docFor(_txn, member->loc);
return advance(id, member, out);
}
Status MMAPV1DatabaseCatalogEntry::_renameSingleNamespace(OperationContext* txn,
StringData fromNS,
StringData toNS,
bool stayTemp) {
// some sanity checking
NamespaceDetails* fromDetails = _namespaceIndex.details(fromNS);
if (!fromDetails)
return Status(ErrorCodes::BadValue, "from namespace doesn't exist");
if (_namespaceIndex.details(toNS))
return Status(ErrorCodes::BadValue, "to namespace already exists");
// at this point, we haven't done anything destructive yet
// ----
// actually start moving
// ----
// this could throw, but if it does we're ok
_namespaceIndex.add_ns(txn, toNS, fromDetails);
NamespaceDetails* toDetails = _namespaceIndex.details(toNS);
try {
toDetails->copyingFrom(txn, toNS, _namespaceIndex, fromDetails); // fixes extraOffset
} catch (DBException&) {
// could end up here if .ns is full - if so try to clean up / roll back a little
_namespaceIndex.kill_ns(txn, toNS);
throw;
}
// at this point, code .ns stuff moved
_namespaceIndex.kill_ns(txn, fromNS);
fromDetails = NULL;
// fix system.namespaces
BSONObj newSpec;
RecordId oldSpecLocation = getCollectionCatalogEntry(fromNS)->getNamespacesRecordId();
invariant(!oldSpecLocation.isNull());
{
BSONObj oldSpec = _getNamespaceRecordStore()->dataFor(txn, oldSpecLocation).releaseToBson();
invariant(!oldSpec.isEmpty());
BSONObjBuilder b;
BSONObjIterator i(oldSpec.getObjectField("options"));
while (i.more()) {
BSONElement e = i.next();
if (strcmp(e.fieldName(), "create") != 0) {
if (stayTemp || (strcmp(e.fieldName(), "temp") != 0))
b.append(e);
} else {
b << "create" << toNS;
}
}
newSpec = b.obj();
}
RecordId rid = _addNamespaceToNamespaceCollection(txn, toNS, newSpec.isEmpty() ? 0 : &newSpec);
_getNamespaceRecordStore()->deleteRecord(txn, oldSpecLocation);
Entry*& entry = _collections[toNS.toString()];
invariant(entry == NULL);
txn->recoveryUnit()->registerChange(new EntryInsertion(toNS, this));
entry = new Entry();
_removeFromCache(txn->recoveryUnit(), fromNS);
_insertInCache(txn, toNS, rid, entry);
return Status::OK();
}
void pass(int p) {
OperationContextImpl txn;
create();
ASSERT_EQUALS( 2, nExtents() );
BSONObj b = bigObj();
int N = MinExtentSize / b.objsize() * nExtents() + 5;
int T = N - 4;
RecordId truncAt;
//RecordId l[ 8 ];
for ( int i = 0; i < N; ++i ) {
BSONObj bb = bigObj();
StatusWith<RecordId> status = collection()->insertDocument( &txn, bb, true );
ASSERT( status.isOK() );
RecordId a = status.getValue();
if( T == i )
truncAt = a;
ASSERT( !a.isNull() );
/*ASSERT_EQUALS( i < 2 ? i + 1 : 3 + i % 2, nRecords() );
if ( i > 3 )
ASSERT( l[ i ] == l[ i - 4 ] );*/
}
ASSERT( nRecords() < N );
RecordId last, first;
{
auto_ptr<Runner> runner(InternalPlanner::collectionScan(&txn,
ns(),
collection(),
InternalPlanner::BACKWARD));
runner->getNext(NULL, &last);
ASSERT( !last.isNull() );
}
{
auto_ptr<Runner> runner(InternalPlanner::collectionScan(&txn,
ns(),
collection(),
InternalPlanner::FORWARD));
runner->getNext(NULL, &first);
ASSERT( !first.isNull() );
ASSERT( first != last ) ;
}
collection()->temp_cappedTruncateAfter(&txn, truncAt, false);
ASSERT_EQUALS( collection()->numRecords() , 28u );
{
RecordId loc;
auto_ptr<Runner> runner(InternalPlanner::collectionScan(&txn,
ns(),
collection(),
InternalPlanner::FORWARD));
runner->getNext(NULL, &loc);
ASSERT( first == loc);
}
{
auto_ptr<Runner> runner(InternalPlanner::collectionScan(&txn,
ns(),
collection(),
InternalPlanner::BACKWARD));
RecordId loc;
runner->getNext(NULL, &loc);
ASSERT( last != loc );
ASSERT( !last.isNull() );
}
// Too big
BSONObjBuilder bob;
bob.appendOID("_id", 0, true);
bob.append( "a", string( MinExtentSize + 300, 'a' ) );
BSONObj bigger = bob.done();
StatusWith<RecordId> status = collection()->insertDocument( &txn, bigger, true );
ASSERT( !status.isOK() );
ASSERT_EQUALS( 0, nRecords() );
}
PlanStage::StageState CollectionScan::work(WorkingSetID* out) {
++_commonStats.works;
// Adds the amount of time taken by work() to executionTimeMillis.
ScopedTimer timer(&_commonStats.executionTimeMillis);
if (_isDead) { return PlanStage::DEAD; }
// Do some init if we haven't already.
if (NULL == _iter) {
if ( _params.collection == NULL ) {
_isDead = true;
return PlanStage::DEAD;
}
try {
if (_lastSeenLoc.isNull()) {
_iter.reset( _params.collection->getIterator( _txn,
_params.start,
_params.direction ) );
}
else {
invariant(_params.tailable);
_iter.reset( _params.collection->getIterator( _txn,
_lastSeenLoc,
_params.direction ) );
// Advance _iter past where we were last time. If it returns something else,
// mark us as dead since we want to signal an error rather than silently
// dropping data from the stream. This is related to the _lastSeenLock handling
// in invalidate.
if (_iter->getNext() != _lastSeenLoc) {
_isDead = true;
return PlanStage::DEAD;
}
}
}
catch (const WriteConflictException& wce) {
// Leave us in a state to try again next time.
_iter.reset();
*out = WorkingSet::INVALID_ID;
return PlanStage::NEED_YIELD;
}
++_commonStats.needTime;
return PlanStage::NEED_TIME;
}
// Should we try getNext() on the underlying _iter?
if (isEOF())
return PlanStage::IS_EOF;
const RecordId curr = _iter->curr();
if (curr.isNull()) {
// We just hit EOF
if (_params.tailable)
_iter.reset(); // pick up where we left off on the next call to work()
return PlanStage::IS_EOF;
}
_lastSeenLoc = curr;
// See if the record we're about to access is in memory. If not, pass a fetch request up.
// Note that curr() does not touch the record. This way, we are able to yield before
// fetching the record.
{
std::auto_ptr<RecordFetcher> fetcher(
_params.collection->documentNeedsFetch(_txn, curr));
if (NULL != fetcher.get()) {
WorkingSetMember* member = _workingSet->get(_wsidForFetch);
member->loc = curr;
// Pass the RecordFetcher off to the WSM.
member->setFetcher(fetcher.release());
*out = _wsidForFetch;
_commonStats.needYield++;
return NEED_YIELD;
}
}
// Do this before advancing because it is more efficient while the iterator is still on this
// document.
const Snapshotted<BSONObj> obj = Snapshotted<BSONObj>(_txn->recoveryUnit()->getSnapshotId(),
_iter->dataFor(curr).releaseToBson());
// Advance the iterator.
try {
invariant(_iter->getNext() == curr);
}
catch (const WriteConflictException& wce) {
// If getNext thows, it leaves us on the original document.
invariant(_iter->curr() == curr);
*out = WorkingSet::INVALID_ID;
return PlanStage::NEED_YIELD;
}
WorkingSetID id = _workingSet->allocate();
WorkingSetMember* member = _workingSet->get(id);
member->loc = curr;
member->obj = obj;
member->state = WorkingSetMember::LOC_AND_UNOWNED_OBJ;
return returnIfMatches(member, id, out);
}
