v8::Handle<v8::Value> ThreadReturnData(V8Scope* scope, const v8::Arguments& args) {
BSONObj data = thisConfig(scope, args)->returnData();
return scope->mongoToV8Element(data.firstElement(), true);
}
Status MMAPV1DatabaseCatalogEntry::renameCollection(OperationContext* txn,
StringData fromNS,
StringData toNS,
bool stayTemp) {
Status s = _renameSingleNamespace(txn, fromNS, toNS, stayTemp);
if (!s.isOK())
return s;
NamespaceDetails* details = _namespaceIndex.details(toNS);
invariant(details);
RecordStoreV1Base* systemIndexRecordStore = _getIndexRecordStore();
auto cursor = systemIndexRecordStore->getCursor(txn);
while (auto record = cursor->next()) {
BSONObj oldIndexSpec = record->data.releaseToBson();
if (fromNS != oldIndexSpec["ns"].valuestrsafe())
continue;
BSONObj newIndexSpec;
{
BSONObjBuilder b;
BSONObjIterator i(oldIndexSpec);
while (i.more()) {
BSONElement e = i.next();
if (strcmp(e.fieldName(), "ns") != 0)
b.append(e);
else
b << "ns" << toNS;
}
newIndexSpec = b.obj();
}
StatusWith<RecordId> newIndexSpecLoc = systemIndexRecordStore->insertRecord(
txn, newIndexSpec.objdata(), newIndexSpec.objsize(), false);
if (!newIndexSpecLoc.isOK())
return newIndexSpecLoc.getStatus();
const std::string& indexName = oldIndexSpec.getStringField("name");
{
// Fix the IndexDetails pointer.
int indexI = getCollectionCatalogEntry(toNS)->_findIndexNumber(txn, indexName);
IndexDetails& indexDetails = details->idx(indexI);
*txn->recoveryUnit()->writing(&indexDetails.info) =
DiskLoc::fromRecordId(newIndexSpecLoc.getValue());
}
{
// Move the underlying namespace.
std::string oldIndexNs = IndexDescriptor::makeIndexNamespace(fromNS, indexName);
std::string newIndexNs = IndexDescriptor::makeIndexNamespace(toNS, indexName);
Status s = _renameSingleNamespace(txn, oldIndexNs, newIndexNs, false);
if (!s.isOK())
return s;
}
systemIndexRecordStore->deleteRecord(txn, record->id);
}
return Status::OK();
}
BSONObj* DBController::readBSON(StreamType* stream) {
auto_ptr<BSONInputStream> is(new BSONInputStream(stream));
BSONObj* res = is->readBSON();
if (_logger->isDebug()) _logger->debug(3, "DBController read bson from disc: %s", res->toChar());
return res;
}
TEST(MatchExpressionParserTest, ParseIntegerElementToNonNegativeLongRejectsNegative) {
BSONObj query = BSON("" << -2LL);
ASSERT_NOT_OK(
MatchExpressionParser::parseIntegerElementToNonNegativeLong(query.firstElement()));
}
void operator()( DBClientCursorBatchIterator &i ) {
mongolock l( true );
if ( context ) {
context->relocked();
}
while( i.moreInCurrentBatch() ) {
if ( n % 128 == 127 /*yield some*/ ) {
time_t now = time(0);
if( now - lastLog >= 60 ) {
// report progress
if( lastLog )
log() << "clone " << to_collection << ' ' << n << endl;
lastLog = now;
}
mayInterrupt( _mayBeInterrupted );
dbtempreleaseif t( _mayYield );
}
BSONObj tmp = i.nextSafe();
/* assure object is valid. note this will slow us down a little. */
if ( !tmp.valid() ) {
stringstream ss;
ss << "Cloner: skipping corrupt object from " << from_collection;
BSONElement e = tmp.firstElement();
try {
e.validate();
ss << " firstElement: " << e;
}
catch( ... ) {
ss << " firstElement corrupt";
}
out() << ss.str() << endl;
continue;
}
++n;
BSONObj js = tmp;
if ( isindex ) {
assert( strstr(from_collection, "system.indexes") );
js = fixindex(tmp);
storedForLater->push_back( js.getOwned() );
continue;
}
try {
theDataFileMgr.insertWithObjMod(to_collection, js);
if ( logForRepl )
logOp("i", to_collection, js);
getDur().commitIfNeeded();
}
catch( UserException& e ) {
log() << "warning: exception cloning object in " << from_collection << ' ' << e.what() << " obj:" << js.toString() << '\n';
}
RARELY if ( time( 0 ) - saveLast > 60 ) {
log() << n << " objects cloned so far from collection " << from_collection << endl;
saveLast = time( 0 );
}
}
}
TEST(MatchExpressionParserTest, ParseIntegerElementToLongRejectsTooLargeNegativeDouble) {
BSONObj query = BSON("" << std::numeric_limits<double>::min());
ASSERT_NOT_OK(MatchExpressionParser::parseIntegerElementToLong(query.firstElement()));
}
TEST(MatchExpressionParserTest, ParseIntegerElementToLongRejectsNonIntegralDecimal) {
BSONObj query = BSON("" << Decimal128("2.5"));
ASSERT_NOT_OK(MatchExpressionParser::parseIntegerElementToLong(query.firstElement()));
}
bool run( const string& dbname,
BSONObj& cmdObj,
int,
string& errmsg,
BSONObjBuilder& result,
bool ) {
string ns;
if ( !FieldParser::extract( cmdObj, nsField, &ns, &errmsg ) ) {
return false;
}
if ( ns.size() == 0 ) {
errmsg = "no namespace specified";
return false;
}
vector<BSONObj> bounds;
if ( !FieldParser::extract( cmdObj, boundsField, &bounds, &errmsg ) ) {
return false;
}
if ( bounds.size() == 0 ) {
errmsg = "no bounds were specified";
return false;
}
if ( bounds.size() != 2 ) {
errmsg = "only a min and max bound may be specified";
return false;
}
BSONObj minKey = bounds[0];
BSONObj maxKey = bounds[1];
if ( minKey.isEmpty() ) {
errmsg = "no min key specified";
return false;
}
if ( maxKey.isEmpty() ) {
errmsg = "no max key specified";
return false;
}
//
// This might be the first call from mongos, so we may need to pass the config and shard
// information to initialize the shardingState.
//
string config;
FieldParser::FieldState extracted = FieldParser::extract( cmdObj,
configField,
&config,
&errmsg );
if ( !extracted ) return false;
if ( extracted != FieldParser::FIELD_NONE ) {
ShardingState::initialize( config );
}
else if ( !shardingState.enabled() ) {
errmsg =
"sharding state must be enabled or config server specified to merge chunks";
return false;
}
// ShardName is optional, but might not be set yet
string shardName;
extracted = FieldParser::extract( cmdObj, shardNameField, &shardName, &errmsg );
if ( !extracted ) return false;
if ( extracted != FieldParser::FIELD_NONE ) {
shardingState.gotShardName( shardName );
}
//
// Epoch is optional, and if not set indicates we should use the latest epoch
//
OID epoch;
if ( !FieldParser::extract( cmdObj, epochField, &epoch, &errmsg ) ) {
return false;
}
return mergeChunks( NamespaceString( ns ), minKey, maxKey, epoch, true, &errmsg );
}
/* ****************************************************************************
*
* mongoUnsubscribeContextAvailability -
*/
HttpStatusCode mongoUnsubscribeContextAvailability
(
UnsubscribeContextAvailabilityRequest* requestP,
UnsubscribeContextAvailabilityResponse* responseP,
const std::string& tenant
)
{
bool reqSemTaken;
std::string err;
reqSemTake(__FUNCTION__, "ngsi9 unsubscribe request", SemWriteOp, &reqSemTaken);
LM_T(LmtMongo, ("Unsubscribe Context Availability"));
/* No matter if success or failure, the subscriptionId in the response is always the one
* in the request */
responseP->subscriptionId = requestP->subscriptionId;
/* Look for document */
BSONObj sub;
OID id;
if (!safeGetSubId(requestP->subscriptionId, &id, &(responseP->statusCode)))
{
reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (safeGetSubId fail)", reqSemTaken);
if (responseP->statusCode.code == SccContextElementNotFound)
{
// FIXME: doubt: invalid OID format? Or, subscription not found?
std::string details = std::string("invalid OID format: '") + requestP->subscriptionId.get() + "'";
alarmMgr.badInput(clientIp, details);
}
else // SccReceiverInternalError
{
LM_E(("Runtime Error (exception getting OID: %s)", responseP->statusCode.details.c_str()));
}
return SccOk;
}
if (!collectionFindOne(getSubscribeContextAvailabilityCollectionName(tenant), BSON("_id" << id), &sub, &err))
{
reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo db exception)", reqSemTaken);
responseP->statusCode.fill(SccReceiverInternalError, err);
return SccOk;
}
alarmMgr.dbErrorReset();
if (sub.isEmpty())
{
responseP->statusCode.fill(SccContextElementNotFound);
reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (no subscriptions)", reqSemTaken);
return SccOk;
}
/* Remove document in MongoDB */
// FIXME: I would prefer to do the find and remove in a single operation. Is the some similar
// to findAndModify for this?
if (!collectionRemove(getSubscribeContextAvailabilityCollectionName(tenant), BSON("_id" << OID(requestP->subscriptionId.get())), &err))
{
reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo db exception)", reqSemTaken);
responseP->statusCode.fill(SccReceiverInternalError, err);
return SccOk;
}
reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request", reqSemTaken);
responseP->statusCode.fill(SccOk);
return SccOk;
}
UpdateResult update(UpdateRequest& request, UpdateDriver* driver) {
const NamespaceString& nsString = request.getNamespaceString();
validateUpdate( nsString.ns().c_str(), request.getUpdates(), request.getQuery() );
NamespaceDetails* nsDetails = nsdetails( nsString.ns() );
NamespaceDetailsTransient* nsDetailsTransient =
&NamespaceDetailsTransient::get( nsString.ns().c_str() );
OpDebug& debug = request.getDebug();
// TODO: This seems a bit circuitious.
debug.updateobj = request.getUpdates();
driver->refreshIndexKeys( nsDetailsTransient->indexKeys() );
shared_ptr<Cursor> cursor = getOptimizedCursor(
nsString.ns(), request.getQuery(), BSONObj(), request.getQueryPlanSelectionPolicy() );
// If the update was marked with '$isolated' (a.k.a '$atomic'), we are not allowed to
// yield while evaluating the update loop below.
//
// TODO: Old code checks this repeatedly within the update loop. Is that necessary? It seems
// that once atomic should be always atomic.
const bool isolated =
cursor->ok() &&
cursor->matcher() &&
cursor->matcher()->docMatcher().atomic();
// The 'cursor' the optimizer gave us may contain query plans that generate duplicate
// diskloc's. We set up here the mechanims that will prevent us from processing those
// twice if we see them. We also set up a 'ClientCursor' so that we can support
// yielding.
//
// TODO: Is it valid to call this on a non-ok cursor?
const bool dedupHere = cursor->autoDedup();
//
// We'll start assuming we have one or more documents for this update. (Othwerwise,
// we'll fallback to upserting.)
//
// We record that this will not be an upsert, in case a mod doesn't want to be applied
// when in strict update mode.
driver->setContext( ModifierInterface::ExecInfo::UPDATE_CONTEXT );
// Let's fetch each of them and pipe them through the update expression, making sure to
// keep track of the necessary stats. Recall that we'll be pulling documents out of
// cursors and some of them do not deduplicate the entries they generate. We have
// deduping logic in here, too -- for now.
unordered_set<DiskLoc, DiskLoc::Hasher> seenLocs;
int numMatched = 0;
debug.nscanned = 0;
Client& client = cc();
mutablebson::Document doc;
// If we are going to be yielding, we will need a ClientCursor scoped to this loop. We
// only loop as long as the underlying cursor is OK.
for ( auto_ptr<ClientCursor> clientCursor; cursor->ok(); ) {
// If we haven't constructed a ClientCursor, and if the client allows us to throw
// page faults, and if we are referring to a location that is likely not in
// physical memory, then throw a PageFaultException. The entire operation will be
// restarted.
if ( clientCursor.get() == NULL &&
client.allowedToThrowPageFaultException() &&
!cursor->currLoc().isNull() &&
!cursor->currLoc().rec()->likelyInPhysicalMemory() ) {
// We should never throw a PFE if we have already updated items.
dassert((numMatched == 0) || (numMatched == debug.nupdateNoops));
throw PageFaultException( cursor->currLoc().rec() );
}
if ( !isolated && debug.nscanned != 0 ) {
// We are permitted to yield. To do so we need a ClientCursor, so create one
// now if we have not yet done so.
if ( !clientCursor.get() )
clientCursor.reset(
new ClientCursor( QueryOption_NoCursorTimeout, cursor, nsString.ns() ) );
// Ask the client cursor to yield. We get two bits of state back: whether or not
// we yielded, and whether or not we correctly recovered from yielding.
bool yielded = false;
const bool recovered = clientCursor->yieldSometimes(
ClientCursor::WillNeed, &yielded );
if ( !recovered ) {
// If we failed to recover from the yield, then the ClientCursor is already
// gone. Release it so we don't destroy it a second time.
clientCursor.release();
break;
}
if ( !cursor->ok() ) {
// If the cursor died while we were yielded, just get out of the update loop.
break;
}
if ( yielded ) {
// We yielded and recovered OK, and our cursor is still good. Details about
// our namespace may have changed while we were yielded, so we re-acquire
// them here. If we can't do so, escape the update loop. Otherwise, refresh
// the driver so that it knows about what is currently indexed.
nsDetails = nsdetails( nsString.ns() );
if ( !nsDetails )
break;
nsDetailsTransient = &NamespaceDetailsTransient::get( nsString.ns().c_str() );
// TODO: This copies the index keys, but it may not need to do so.
driver->refreshIndexKeys( nsDetailsTransient->indexKeys() );
}
}
// Let's fetch the next candidate object for this update.
Record* record = cursor->_current();
DiskLoc loc = cursor->currLoc();
const BSONObj oldObj = loc.obj();
// We count how many documents we scanned even though we may skip those that are
// deemed duplicated. The final 'numUpdated' and 'nscanned' numbers may differ for
// that reason.
debug.nscanned++;
// Skips this document if it:
// a) doesn't match the query portion of the update
// b) was deemed duplicate by the underlying cursor machinery
//
// Now, if we are going to update the document,
// c) we don't want to do so while the cursor is at it, as that may invalidate
// the cursor. So, we advance to next document, before issuing the update.
MatchDetails matchDetails;
matchDetails.requestElemMatchKey();
if ( !cursor->currentMatches( &matchDetails ) ) {
// a)
cursor->advance();
continue;
}
else if ( cursor->getsetdup( loc ) && dedupHere ) {
// b)
cursor->advance();
continue;
}
else if (!driver->isDocReplacement() && request.isMulti()) {
// c)
cursor->advance();
if ( dedupHere ) {
if ( seenLocs.count( loc ) ) {
continue;
}
}
// There are certain kind of cursors that hold multiple pointers to data
// underneath. $or cursors is one example. In a $or cursor, it may be the case
// that when we did the last advance(), we finished consuming documents from
// one of $or child and started consuming the next one. In that case, it is
// possible that the last document of the previous child is the same as the
// first document of the next (see SERVER-5198 and jstests/orp.js).
//
// So we advance the cursor here until we see a new diskloc.
//
// Note that we won't be yielding, and we may not do so for a while if we find
// a particularly duplicated sequence of loc's. That is highly unlikely,
// though. (See SERVER-5725, if curious, but "stage" based $or will make that
// ticket moot).
while( cursor->ok() && loc == cursor->currLoc() ) {
cursor->advance();
}
}
// For some (unfortunate) historical reasons, not all cursors would be valid after
// a write simply because we advanced them to a document not affected by the write.
// To protect in those cases, not only we engaged in the advance() logic above, but
// we also tell the cursor we're about to write a document that we've just seen.
// prepareToTouchEarlierIterate() requires calling later
// recoverFromTouchingEarlierIterate(), so we make a note here to do so.
bool touchPreviousDoc = request.isMulti() && cursor->ok();
if ( touchPreviousDoc ) {
if ( clientCursor.get() )
clientCursor->setDoingDeletes( true );
cursor->prepareToTouchEarlierIterate();
}
// Found a matching document
numMatched++;
// Ask the driver to apply the mods. It may be that the driver can apply those "in
// place", that is, some values of the old document just get adjusted without any
// change to the binary layout on the bson layer. It may be that a whole new
// document is needed to accomodate the new bson layout of the resulting document.
doc.reset( oldObj, mutablebson::Document::kInPlaceEnabled );
BSONObj logObj;
// If there was a matched field, obtain it.
string matchedField;
if (matchDetails.hasElemMatchKey())
matchedField = matchDetails.elemMatchKey();
Status status = driver->update( matchedField, &doc, &logObj );
if ( !status.isOK() ) {
uasserted( 16837, status.reason() );
}
// If the driver applied the mods in place, we can ask the mutable for what
// changed. We call those changes "damages". :) We use the damages to inform the
// journal what was changed, and then apply them to the original document
// ourselves. If, however, the driver applied the mods out of place, we ask it to
// generate a new, modified document for us. In that case, the file manager will
// take care of the journaling details for us.
//
// This code flow is admittedly odd. But, right now, journaling is baked in the file
// manager. And if we aren't using the file manager, we have to do jounaling
// ourselves.
bool objectWasChanged = false;
BSONObj newObj;
const char* source = NULL;
mutablebson::DamageVector damages;
bool inPlace = doc.getInPlaceUpdates(&damages, &source);
if ( inPlace && !damages.empty() && !driver->modsAffectIndices() ) {
nsDetails->paddingFits();
// 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 = source + where->sourceOffset;
void* targetPtr = getDur().writingPtr(
const_cast<char*>(oldObj.objdata()) + where->targetOffset,
where->size);
std::memcpy(targetPtr, sourcePtr, where->size);
}
newObj = oldObj;
debug.fastmod = true;
objectWasChanged = true;
}
else {
// The updates were not in place. Apply them through the file manager.
newObj = doc.getObject();
DiskLoc newLoc = theDataFileMgr.updateRecord(nsString.ns().c_str(),
nsDetails,
nsDetailsTransient,
record,
loc,
newObj.objdata(),
newObj.objsize(),
debug);
// If we've moved this object to a new location, make sure we don't apply
// that update again if our traversal picks the objecta again.
//
// We also take note that the diskloc if the updates are affecting indices.
// Chances are that we're traversing one of them and they may be multi key and
// therefore duplicate disklocs.
if ( newLoc != loc || driver->modsAffectIndices() ) {
seenLocs.insert( newLoc );
}
objectWasChanged = true;
}
// Log Obj
if ( request.shouldUpdateOpLog() ) {
if ( driver->isDocReplacement() || !logObj.isEmpty() ) {
BSONObj idQuery = driver->makeOplogEntryQuery(newObj, request.isMulti());
logOp("u", nsString.ns().c_str(), logObj , &idQuery,
NULL, request.isFromMigration(), &newObj);
}
}
// If it was noop since the document didn't change, record that.
if (!objectWasChanged)
debug.nupdateNoops++;
if (!request.isMulti()) {
break;
}
// If we used the cursor mechanism that prepares an earlier seen document for a
// write we need to tell such mechanisms that the write is over.
if ( touchPreviousDoc ) {
cursor->recoverFromTouchingEarlierIterate();
}
getDur().commitIfNeeded();
}
// TODO: Can this be simplified?
if ((numMatched > 0) || (numMatched == 0 && !request.isUpsert()) ) {
debug.nupdated = numMatched;
return UpdateResult( numMatched > 0 /* updated existing object(s) */,
!driver->isDocReplacement() /* $mod or obj replacement */,
numMatched /* # of docments update, even no-ops */,
BSONObj() );
}
//
// We haven't found any existing document so an insert is done
// (upsert is true).
//
debug.upsert = true;
// Since this is an insert (no docs found and upsert:true), we will be logging it
// as an insert in the oplog. We don't need the driver's help to build the
// oplog record, then. We also set the context of the update driver to the INSERT_CONTEXT.
// Some mods may only work in that context (e.g. $setOnInsert).
driver->setLogOp( false );
driver->setContext( ModifierInterface::ExecInfo::INSERT_CONTEXT );
BSONObj baseObj;
// Reset the document we will be writing to
doc.reset( baseObj, mutablebson::Document::kInPlaceDisabled );
if ( request.getQuery().hasElement("_id") ) {
uassertStatusOK(doc.root().appendElement(request.getQuery().getField("_id")));
}
// If this is a $mod base update, we need to generate a document by examining the
// query and the mods. Otherwise, we can use the object replacement sent by the user
// update command that was parsed by the driver before.
// In the following block we handle the query part, and then do the regular mods after.
if ( *request.getUpdates().firstElementFieldName() == '$' ) {
uassertStatusOK(UpdateDriver::createFromQuery(request.getQuery(), doc));
debug.fastmodinsert = true;
}
// Apply the update modifications and then log the update as an insert manually.
Status status = driver->update( StringData(), &doc, NULL /* no oplog record */);
if ( !status.isOK() ) {
uasserted( 16836, status.reason() );
}
BSONObj newObj = doc.getObject();
theDataFileMgr.insertWithObjMod( nsString.ns().c_str(), newObj, false, request.isGod() );
if ( request.shouldUpdateOpLog() ) {
logOp( "i", nsString.ns().c_str(), newObj,
NULL, NULL, request.isFromMigration(), &newObj );
}
debug.nupdated = 1;
return UpdateResult( false /* updated a non existing document */,
!driver->isDocReplacement() /* $mod or obj replacement? */,
1 /* count of updated documents */,
newObj /* object that was upserted */ );
}
// static
Status ClearFilters::clear(OperationContext* txn,
QuerySettings* querySettings,
PlanCache* planCache,
const std::string& ns,
const BSONObj& cmdObj) {
invariant(querySettings);
// According to the specification, the planCacheClearFilters command runs in two modes:
// - clear all hints; or
// - clear hints for single query shape when a query shape is described in the
// command arguments.
if (cmdObj.hasField("query")) {
auto statusWithCQ = PlanCacheCommand::canonicalize(txn, ns, cmdObj);
if (!statusWithCQ.isOK()) {
return statusWithCQ.getStatus();
}
unique_ptr<CanonicalQuery> cq = std::move(statusWithCQ.getValue());
querySettings->removeAllowedIndices(planCache->computeKey(*cq));
// Remove entry from plan cache
planCache->remove(*cq);
LOG(0) << "Removed index filter on " << ns << " " << cq->toStringShort();
return Status::OK();
}
// If query is not provided, make sure sort and projection are not in arguments.
// We do not want to clear the entire cache inadvertently when the user
// forgot to provide a value for "query".
if (cmdObj.hasField("sort") || cmdObj.hasField("projection")) {
return Status(ErrorCodes::BadValue, "sort or projection provided without query");
}
// Get entries from query settings. We need to remove corresponding entries from the plan
// cache shortly.
OwnedPointerVector<AllowedIndexEntry> entries;
entries.mutableVector() = querySettings->getAllAllowedIndices();
// OK to proceed with clearing entire cache.
querySettings->clearAllowedIndices();
const NamespaceString nss(ns);
const ExtensionsCallbackReal extensionsCallback(txn, &nss);
// Remove corresponding entries from plan cache.
// Admin hints affect the planning process directly. If there were
// plans generated as a result of applying index filter, these need to be
// invalidated. This allows the planner to re-populate the plan cache with
// non-filtered indexed solutions next time the query is run.
// Resolve plan cache key from (query, sort, projection) in query settings entry.
// Concurrency note: There's no harm in removing plan cache entries one at at time.
// Only way that PlanCache::remove() can fail is when the query shape has been removed from
// the cache by some other means (re-index, collection info reset, ...). This is OK since
// that's the intended effect of calling the remove() function with the key from the hint entry.
for (vector<AllowedIndexEntry*>::const_iterator i = entries.begin(); i != entries.end(); ++i) {
AllowedIndexEntry* entry = *i;
invariant(entry);
// Create canonical query.
auto qr = stdx::make_unique<QueryRequest>(nss);
qr->setFilter(entry->query);
qr->setSort(entry->sort);
qr->setProj(entry->projection);
auto statusWithCQ = CanonicalQuery::canonicalize(txn, std::move(qr), extensionsCallback);
invariantOK(statusWithCQ.getStatus());
std::unique_ptr<CanonicalQuery> cq = std::move(statusWithCQ.getValue());
// Remove plan cache entry.
planCache->remove(*cq);
}
LOG(0) << "Removed all index filters for collection: " << ns;
return Status::OK();
}
void handle_modify(DBClientBase *client, DBOperation *operation)
{
// First, we have to format our findandmodify.
BSONObjBuilder sets;
bool has_sets = false;
BSONObjBuilder unsets;
bool has_unsets = false;
for(auto it = operation->set_fields().begin();
it != operation->set_fields().end(); ++it) {
stringstream fieldname;
fieldname << "fields." << it->first->get_name();
if(it->second.empty()) {
unsets << fieldname.str() << true;
has_unsets = true;
} else {
DatagramPtr dg = Datagram::create();
dg->add_data(it->second);
DatagramIterator dgi(dg);
sets << fieldname.str() << bamboo2bson(it->first->get_type(), dgi)["_"];
has_sets = true;
}
}
BSONObjBuilder updates_b;
if(has_sets) {
updates_b << "$set" << sets.obj();
}
if(has_unsets) {
updates_b << "$unset" << unsets.obj();
}
BSONObj updates = updates_b.obj();
// Also format any criteria for the change:
BSONObjBuilder query_b;
query_b << "_id" << operation->doid();
for(auto it = operation->criteria_fields().begin();
it != operation->criteria_fields().end(); ++it) {
stringstream fieldname;
fieldname << "fields." << it->first->get_name();
if(it->second.empty()) {
query_b << fieldname.str() << BSON("$exists" << false);
} else {
DatagramPtr dg = Datagram::create();
dg->add_data(it->second);
DatagramIterator dgi(dg);
query_b << fieldname.str() << bamboo2bson(it->first->get_type(), dgi)["_"];
}
}
BSONObj query = query_b.obj();
m_log->trace() << "Performing updates to " << operation->doid()
<< ": " << updates << endl;
m_log->trace() << "Query is: " << query << endl;
BSONObj result;
bool success;
try {
success = client->runCommand(
m_db,
BSON("findandmodify" << "astron.objects"
<< "query" << query
<< "update" << updates),
result);
} catch(mongo::DBException &e) {
m_log->error() << "Unexpected error while modifying "
<< operation->doid() << ": " << e.what() << endl;
operation->on_failure();
return;
}
m_log->trace() << "Update result: " << result << endl;
BSONObj obj;
if(!success || result["value"].isNull()) {
// Okay, something didn't work right. If we had criteria, let's
// try to fetch the object without the criteria to see if it's a
// criteria mismatch or a missing DOID.
if(!operation->criteria_fields().empty()) {
try {
obj = client->findOne(m_obj_collection,
BSON("_id" << operation->doid()));
} catch(mongo::DBException &e) {
m_log->error() << "Unexpected error while modifying "
<< operation->doid() << ": " << e.what() << endl;
operation->on_failure();
return;
}
if(!obj.isEmpty()) {
// There's the problem. Now we can send back a snapshot:
DBObjectSnapshot *snap = format_snapshot(operation->doid(), obj);
if(snap && operation->verify_class(snap->m_dclass)) {
operation->on_criteria_mismatch(snap);
return;
} else {
// Something else weird happened with our snapshot;
// either the class wasn't recognized or it was the
// wrong class. Either way, an error has been logged,
// and we need to fail the operation.
operation->on_failure();
return;
}
}
}
// Nope, not that. We're missing the DOID.
m_log->error() << "Attempted to modify unknown DOID: "
<< operation->doid() << endl;
operation->on_failure();
return;
}
// If we've gotten to this point: Hooray! The change has gone
// through to the database.
// Let's, however, double-check our changes. (Specifically, we should
// run verify_class so that we know the frontend is happy with what
// kind of object we just modified.)
obj = result["value"].Obj();
try {
string dclass_name = obj["dclass"].String();
const dclass::Class *dclass = g_dcf->get_class_by_name(dclass_name);
if(!dclass) {
m_log->error() << "Encountered unknown database object: "
<< dclass_name << "(" << operation->doid() << ")" << endl;
} else if(operation->verify_class(dclass)) {
// Yep, it all checks out. Complete the operation:
operation->on_complete();
return;
}
} catch(mongo::DBException &e) { }
// If we've gotten here, something is seriously wrong. We've just
// mucked with an object without knowing the consequences! What have
// we done?! We've created an abomination in the database! Kill it!
// Kill it with fire!
// All we really can do to mitigate this is scream at the user (which
// the above verification has already done by now) and revert the
// object back to how it was when we found it.
// NOTE: This DOES have the potential for data loss, because we're
// wiping out any changes that conceivably could have happened
// between the findandmodify and now. In dev environments, (which we
// are probably in right now, if other components are making
// outlandish requests like this) this shouldn't be a huge issue.
m_log->trace() << "Reverting changes made to " << operation->doid() << endl;
try {
client->update(
m_obj_collection,
BSON("_id" << operation->doid()),
obj);
} catch(mongo::DBException &e) {
// Wow, we REALLY fail at life.
m_log->error() << "Could not revert corrupting changes to "
<< operation->doid() << ": " << e.what() << endl;
}
operation->on_failure();
}
void operator()( DBClientCursorBatchIterator &i ) {
Lock::GlobalWrite lk;
if ( context ) {
context->relocked();
}
while( i.moreInCurrentBatch() ) {
if ( n % 128 == 127 /*yield some*/ ) {
time_t now = time(0);
if( now - lastLog >= 60 ) {
// report progress
if( lastLog )
log() << "clone " << to_collection << ' ' << n << endl;
lastLog = now;
}
mayInterrupt( _mayBeInterrupted );
dbtempreleaseif t( _mayYield );
}
BSONObj tmp = i.nextSafe();
/* assure object is valid. note this will slow us down a little. */
if ( !tmp.valid() ) {
stringstream ss;
ss << "Cloner: skipping corrupt object from " << from_collection;
BSONElement e = tmp.firstElement();
try {
e.validate();
ss << " firstElement: " << e;
}
catch( ... ) {
ss << " firstElement corrupt";
}
out() << ss.str() << endl;
continue;
}
++n;
BSONObj js = tmp;
if ( isindex ) {
verify( strstr(from_collection, "system.indexes") );
js = fixindex(tmp);
storedForLater->push_back( js.getOwned() );
continue;
}
try {
// add keys for presorting
DiskLoc loc = theDataFileMgr.insertWithObjMod(to_collection, js);
loc.assertOk();
if (_sortersForIndex != NULL) {
// add key to SortersForNS
for (SortersForIndex::iterator iSorter = _sortersForIndex->begin();
iSorter != _sortersForIndex->end();
++iSorter) {
iSorter->second.preSortPhase.addKeys(iSorter->second.spec, js,
loc, false);
}
}
if ( logForRepl )
logOp("i", to_collection, js);
getDur().commitIfNeeded();
}
catch( UserException& e ) {
error() << "error: exception cloning object in " << from_collection << ' ' << e.what() << " obj:" << js.toString() << '\n';
throw;
}
RARELY if ( time( 0 ) - saveLast > 60 ) {
log() << n << " objects cloned so far from collection " << from_collection << endl;
saveLast = time( 0 );
}
}
}
void FTSIndexFormat::getKeys(const FTSSpec& spec, const BSONObj& obj, BSONObjSet* keys) {
int extraSize = 0;
vector<BSONElement> extrasBefore;
vector<BSONElement> extrasAfter;
// Compute the non FTS key elements for the prefix.
for (unsigned i = 0; i < spec.numExtraBefore(); i++) {
auto indexedElement = extractNonFTSKeyElement(obj, spec.extraBefore(i));
extrasBefore.push_back(indexedElement);
extraSize += indexedElement.size();
}
// Compute the non FTS key elements for the suffix.
for (unsigned i = 0; i < spec.numExtraAfter(); i++) {
auto indexedElement = extractNonFTSKeyElement(obj, spec.extraAfter(i));
extrasAfter.push_back(indexedElement);
extraSize += indexedElement.size();
}
TermFrequencyMap term_freqs;
spec.scoreDocument(obj, &term_freqs);
// create index keys from raw scores
// only 1 per string
// TODO SERVER-36440: Completely remove this limit in 4.3.
if (serverGlobalParams.featureCompatibility.isVersionInitialized() &&
serverGlobalParams.featureCompatibility.getVersion() ==
ServerGlobalParams::FeatureCompatibility::Version::kFullyDowngradedTo40) {
uassert(16732,
str::stream() << "too many unique keys for a single document to"
<< " have a text index, max is "
<< term_freqs.size()
<< obj["_id"],
term_freqs.size() <= 400000);
}
long long keyBSONSize = 0;
const int MaxKeyBSONSizeMB = 4;
for (TermFrequencyMap::const_iterator i = term_freqs.begin(); i != term_freqs.end(); ++i) {
const string& term = i->first;
double weight = i->second;
// guess the total size of the btree entry based on the size of the weight, term tuple
int guess = 5 /* bson overhead */ + 10 /* weight */ + 8 /* term overhead */ +
/* term size (could be truncated/hashed) */
guessTermSize(term, spec.getTextIndexVersion()) + extraSize;
BSONObjBuilder b(guess); // builds a BSON object with guess length.
for (unsigned k = 0; k < extrasBefore.size(); k++) {
b.appendAs(extrasBefore[k], "");
}
_appendIndexKey(b, weight, term, spec.getTextIndexVersion());
for (unsigned k = 0; k < extrasAfter.size(); k++) {
b.appendAs(extrasAfter[k], "");
}
BSONObj res = b.obj();
verify(guess >= res.objsize());
keys->insert(res);
keyBSONSize += res.objsize();
// TODO SERVER-36440: Completely remove this limit in 4.3.
if (serverGlobalParams.featureCompatibility.isVersionInitialized() &&
serverGlobalParams.featureCompatibility.getVersion() ==
ServerGlobalParams::FeatureCompatibility::Version::kFullyDowngradedTo40) {
uassert(16733,
str::stream() << "trying to index text where term list is too big, max is "
<< MaxKeyBSONSizeMB
<< "mb "
<< obj["_id"],
keyBSONSize <= (MaxKeyBSONSizeMB * 1024 * 1024));
}
}
}
Status ChunkManagerTargeter::targetUpdate(OperationContext* txn,
const BatchedUpdateDocument& updateDoc,
vector<ShardEndpoint*>* endpoints) const {
//
// Update targeting may use either the query or the update. This is to support save-style
// updates, of the form:
//
// coll.update({ _id : xxx }, { _id : xxx, shardKey : 1, foo : bar }, { upsert : true })
//
// Because drivers do not know the shard key, they can't pull the shard key automatically
// into the query doc, and to correctly support upsert we must target a single shard.
//
// The rule is simple - If the update is replacement style (no '$set'), we target using the
// update. If the update is replacement style, we target using the query.
//
// If we have the exact shard key in either the query or replacement doc, we target using
// that extracted key.
//
BSONObj query = updateDoc.getQuery();
BSONObj updateExpr = updateDoc.getUpdateExpr();
UpdateType updateType = getUpdateExprType(updateDoc.getUpdateExpr());
if (updateType == UpdateType_Unknown) {
return Status(ErrorCodes::UnsupportedFormat,
stream() << "update document " << updateExpr
<< " has mixed $operator and non-$operator style fields");
}
BSONObj shardKey;
if (_manager) {
//
// Sharded collections have the following futher requirements for targeting:
//
// Upserts must be targeted exactly by shard key.
// Non-multi updates must be targeted exactly by shard key *or* exact _id.
//
// Get the shard key
if (updateType == UpdateType_OpStyle) {
// Target using the query
StatusWith<BSONObj> status =
_manager->getShardKeyPattern().extractShardKeyFromQuery(query);
// Bad query
if (!status.isOK())
return status.getStatus();
shardKey = status.getValue();
} else {
// Target using the replacement document
shardKey = _manager->getShardKeyPattern().extractShardKeyFromDoc(updateExpr);
}
//
// Extra sharded update validation
//
if (updateDoc.getUpsert()) {
// Sharded upserts *always* need to be exactly targeted by shard key
if (shardKey.isEmpty()) {
return Status(ErrorCodes::ShardKeyNotFound,
stream() << "upsert " << updateDoc.toBSON()
<< " does not contain shard key for pattern "
<< _manager->getShardKeyPattern().toString());
}
// Also check shard key size on upsert
Status status = ShardKeyPattern::checkShardKeySize(shardKey);
if (!status.isOK())
return status;
}
// Validate that single (non-multi) sharded updates are targeted by shard key or _id
if (!updateDoc.getMulti() && shardKey.isEmpty() && !isExactIdQuery(updateDoc.getQuery())) {
return Status(ErrorCodes::ShardKeyNotFound,
stream() << "update " << updateDoc.toBSON()
<< " does not contain _id or shard key for pattern "
<< _manager->getShardKeyPattern().toString());
}
}
// Target the shard key, query, or replacement doc
if (!shardKey.isEmpty()) {
// We can't rely on our query targeting to be exact
ShardEndpoint* endpoint = NULL;
Status result =
targetShardKey(txn, shardKey, (query.objsize() + updateExpr.objsize()), &endpoint);
endpoints->push_back(endpoint);
return result;
} else if (updateType == UpdateType_OpStyle) {
return targetQuery(query, endpoints);
} else {
return targetDoc(updateExpr, endpoints);
}
}
DiskLoc _repairExtent( Database* db , string ns, bool forward , DiskLoc eLoc , Writer& w ) {
LogIndentLevel lil;
if ( eLoc.getOfs() <= 0 ) {
error() << "invalid extent ofs: " << eLoc.getOfs() << endl;
return DiskLoc();
}
MongoDataFile * mdf = db->getFile( eLoc.a() );
Extent * e = mdf->debug_getExtent( eLoc );
if ( ! e->isOk() ) {
warning() << "Extent not ok magic: " << e->magic << " going to try to continue" << endl;
}
log() << "length:" << e->length << endl;
LogIndentLevel lil2;
set<DiskLoc> seen;
DiskLoc loc = forward ? e->firstRecord : e->lastRecord;
while ( ! loc.isNull() ) {
if ( ! seen.insert( loc ).second ) {
error() << "infinite loop in extend, seen: " << loc << " before" << endl;
break;
}
if ( loc.getOfs() <= 0 ) {
error() << "offset is 0 for record which should be impossible" << endl;
break;
}
log(1) << loc << endl;
Record* rec = loc.rec();
BSONObj obj;
try {
obj = loc.obj();
assert( obj.valid() );
LOG(1) << obj << endl;
w( obj );
}
catch ( std::exception& e ) {
log() << "found invalid document @ " << loc << " " << e.what() << endl;
if ( ! obj.isEmpty() ) {
try {
BSONElement e = obj.firstElement();
stringstream ss;
ss << "first element: " << e;
log() << ss.str();
}
catch ( std::exception& ) {
}
}
}
loc = forward ? rec->getNext( loc ) : rec->getPrev( loc );
}
return forward ? e->xnext : e->xprev;
}
TEST(MatchExpressionParserTest, ParseIntegerElementToLongRejectsTooLargeDouble) {
BSONObj query = BSON("" << MatchExpressionParser::kLongLongMaxPlusOneAsDouble);
ASSERT_NOT_OK(MatchExpressionParser::parseIntegerElementToLong(query.firstElement()));
}
int run() {
if ( hasParam( "repair" ) ) {
warning() << "repair is a work in progress" << endl;
return repair();
}
{
string q = getParam("query");
if ( q.size() )
_query = fromjson( q );
}
string opLogName = "";
unsigned long long opLogStart = 0;
if (hasParam("oplog")) {
if (hasParam("query") || hasParam("db") || hasParam("collection")) {
cout << "oplog mode is only supported on full dumps" << endl;
return -1;
}
BSONObj isMaster;
conn("true").simpleCommand("admin", &isMaster, "isMaster");
if (isMaster.hasField("hosts")) { // if connected to replica set member
opLogName = "local.oplog.rs";
}
else {
opLogName = "local.oplog.$main";
if ( ! isMaster["ismaster"].trueValue() ) {
cout << "oplog mode is only supported on master or replica set member" << endl;
return -1;
}
}
auth("local");
BSONObj op = conn(true).findOne(opLogName, Query().sort("$natural", -1), 0, QueryOption_SlaveOk);
if (op.isEmpty()) {
cout << "No operations in oplog. Please ensure you are connecting to a master." << endl;
return -1;
}
assert(op["ts"].type() == Timestamp);
opLogStart = op["ts"]._numberLong();
}
// check if we're outputting to stdout
string out = getParam("out");
if ( out == "-" ) {
if ( _db != "" && _coll != "" ) {
writeCollectionStdout( _db+"."+_coll );
return 0;
}
else {
cout << "You must specify database and collection to print to stdout" << endl;
return -1;
}
}
_usingMongos = isMongos();
path root( out );
string db = _db;
if ( db == "" ) {
cout << "all dbs" << endl;
auth( "admin" );
BSONObj res = conn( true ).findOne( "admin.$cmd" , BSON( "listDatabases" << 1 ) );
if ( ! res["databases"].isABSONObj() ) {
error() << "output of listDatabases isn't what we expected, no 'databases' field:\n" << res << endl;
return -2;
}
BSONObj dbs = res["databases"].embeddedObjectUserCheck();
set<string> keys;
dbs.getFieldNames( keys );
for ( set<string>::iterator i = keys.begin() ; i != keys.end() ; i++ ) {
string key = *i;
if ( ! dbs[key].isABSONObj() ) {
error() << "database field not an object key: " << key << " value: " << dbs[key] << endl;
return -3;
}
BSONObj dbobj = dbs[key].embeddedObjectUserCheck();
const char * dbName = dbobj.getField( "name" ).valuestr();
if ( (string)dbName == "local" )
continue;
go ( dbName , root / dbName );
}
}
else {
auth( db );
go( db , root / db );
}
if (!opLogName.empty()) {
BSONObjBuilder b;
b.appendTimestamp("$gt", opLogStart);
_query = BSON("ts" << b.obj());
writeCollectionFile( opLogName , root / "oplog.bson" );
}
return 0;
}
TEST(MatchExpressionParserTest, ParseIntegerElementToLongRejectsString) {
BSONObj query = BSON(""
<< "1");
ASSERT_NOT_OK(MatchExpressionParser::parseIntegerElementToLong(query.firstElement()));
}
virtual bool run(OperationContext* txn, const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl) {
if ( cmdObj.firstElement().type() != Array ) {
errmsg = "ops has to be an array";
return false;
}
BSONObj ops = cmdObj.firstElement().Obj();
{
// check input
BSONObjIterator i( ops );
while ( i.more() ) {
BSONElement e = i.next();
if (!_checkOperation(e, errmsg)) {
return false;
}
}
}
// SERVER-4328 todo : is global ok or does this take a long time? i believe multiple
// ns used so locking individually requires more analysis
ScopedTransaction scopedXact(txn, MODE_X);
Lock::GlobalWrite globalWriteLock(txn->lockState());
if (!fromRepl &&
!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase(dbname)) {
return appendCommandStatus(result, Status(ErrorCodes::NotMaster, str::stream()
<< "Not primary while applying ops to database " << dbname));
}
// Preconditions check reads the database state, so needs to be done locked
if ( cmdObj["preCondition"].type() == Array ) {
BSONObjIterator i( cmdObj["preCondition"].Obj() );
while ( i.more() ) {
BSONObj f = i.next().Obj();
DBDirectClient db( txn );
BSONObj realres = db.findOne( f["ns"].String() , f["q"].Obj() );
// Apply-ops would never have a $where matcher, so use the default callback,
// which will throw an error if $where is found.
Matcher m(f["res"].Obj());
if ( ! m.matches( realres ) ) {
result.append( "got" , realres );
result.append( "whatFailed" , f );
errmsg = "pre-condition failed";
return false;
}
}
}
// apply
int num = 0;
int errors = 0;
BSONObjIterator i( ops );
BSONArrayBuilder ab;
const bool alwaysUpsert = cmdObj.hasField("alwaysUpsert") ?
cmdObj["alwaysUpsert"].trueValue() : true;
while ( i.more() ) {
BSONElement e = i.next();
const BSONObj& temp = e.Obj();
// Ignore 'n' operations.
const char *opType = temp["op"].valuestrsafe();
if (*opType == 'n') continue;
const string ns = temp["ns"].String();
// Run operations under a nested lock as a hack to prevent yielding.
//
// The list of operations is supposed to be applied atomically; yielding
// would break atomicity by allowing an interruption or a shutdown to occur
// after only some operations are applied. We are already locked globally
// at this point, so taking a DBLock on the namespace creates a nested lock,
// and yields are disallowed for operations that hold a nested lock.
//
// We do not have a wrapping WriteUnitOfWork so it is possible for a journal
// commit to happen with a subset of ops applied.
// TODO figure out what to do about this.
Lock::GlobalWrite globalWriteLockDisallowTempRelease(txn->lockState());
// Ensures that yielding will not happen (see the comment above).
DEV {
Locker::LockSnapshot lockSnapshot;
invariant(!txn->lockState()->saveLockStateAndUnlock(&lockSnapshot));
};
OldClientContext ctx(txn, ns);
Status status(ErrorCodes::InternalError, "");
while (true) {
try {
// We assume that in the WriteConflict retry case, either the op rolls back
// any changes it makes or is otherwise safe to rerun.
status =
repl::applyOperation_inlock(txn, ctx.db(), temp, false, alwaysUpsert);
break;
}
catch (const WriteConflictException& wce) {
LOG(2) << "WriteConflictException in applyOps command, retrying.";
txn->recoveryUnit()->commitAndRestart();
continue;
}
}
ab.append(status.isOK());
if (!status.isOK()) {
errors++;
}
num++;
WriteUnitOfWork wuow(txn);
logOpForDbHash(txn, ns.c_str());
wuow.commit();
}
result.append( "applied" , num );
result.append( "results" , ab.arr() );
if ( ! fromRepl ) {
// We want this applied atomically on slaves
// so we re-wrap without the pre-condition for speed
string tempNS = str::stream() << dbname << ".$cmd";
// TODO: possibly use mutable BSON to remove preCondition field
// once it is available
BSONObjIterator iter(cmdObj);
BSONObjBuilder cmdBuilder;
while (iter.more()) {
BSONElement elem(iter.next());
if (strcmp(elem.fieldName(), "preCondition") != 0) {
cmdBuilder.append(elem);
}
}
const BSONObj cmdRewritten = cmdBuilder.done();
// We currently always logOp the command regardless of whether the individial ops
// succeeded and rely on any failures to also happen on secondaries. This isn't
// perfect, but it's what the command has always done and is part of its "correct"
// behavior.
while (true) {
try {
WriteUnitOfWork wunit(txn);
getGlobalEnvironment()->getOpObserver()->onApplyOps(txn,
tempNS,
cmdRewritten);
wunit.commit();
break;
}
catch (const WriteConflictException& wce) {
LOG(2) <<
"WriteConflictException while logging applyOps command, retrying.";
txn->recoveryUnit()->commitAndRestart();
continue;
}
}
}
if (errors != 0) {
return false;
}
return true;
}
TEST(MatchExpressionParserTest, ParseIntegerElementToLongRejectsLargestDecimal) {
BSONObj query = BSON("" << Decimal128(Decimal128::kLargestPositive));
ASSERT_NOT_OK(MatchExpressionParser::parseIntegerElementToLong(query.firstElement()));
}
long long DeleteExecutor::execute(OperationContext* txn, Database* db) {
uassertStatusOK(prepare());
uassert(17417,
mongoutils::str::stream() <<
"DeleteExecutor::prepare() failed to parse query " << _request->getQuery(),
_isQueryParsed);
const bool logop = _request->shouldCallLogOp();
const NamespaceString& ns(_request->getNamespaceString());
if (!_request->isGod()) {
if (ns.isSystem()) {
uassert(12050,
"cannot delete from system namespace",
legalClientSystemNS(ns.ns(), true));
}
if (ns.ns().find('$') != string::npos) {
log() << "cannot delete from collection with reserved $ in name: " << ns << endl;
uasserted( 10100, "cannot delete from collection with reserved $ in name" );
}
}
Collection* collection = db->getCollection(ns.ns());
if (NULL == collection) {
return 0;
}
uassert(10101,
str::stream() << "cannot remove from a capped collection: " << ns.ns(),
!collection->isCapped());
uassert(ErrorCodes::NotMaster,
str::stream() << "Not primary while removing from " << ns.ns(),
!logop || replset::isMasterNs(ns.ns().c_str()));
long long nDeleted = 0;
Runner* rawRunner;
if (_canonicalQuery.get()) {
uassertStatusOK(getRunner(collection, _canonicalQuery.release(), &rawRunner));
}
else {
CanonicalQuery* ignored;
uassertStatusOK(getRunner(collection,
ns.ns(),
_request->getQuery(),
&rawRunner,
&ignored));
}
auto_ptr<Runner> runner(rawRunner);
ScopedRunnerRegistration safety(runner.get());
DiskLoc rloc;
Runner::RunnerState state;
CurOp* curOp = cc().curop();
int oldYieldCount = curOp->numYields();
while (Runner::RUNNER_ADVANCED == (state = runner->getNext(NULL, &rloc))) {
if (oldYieldCount != curOp->numYields()) {
uassert(ErrorCodes::NotMaster,
str::stream() << "No longer primary while removing from " << ns.ns(),
!logop || replset::isMasterNs(ns.ns().c_str()));
oldYieldCount = curOp->numYields();
}
BSONObj toDelete;
// TODO: do we want to buffer docs and delete them in a group rather than
// saving/restoring state repeatedly?
runner->saveState();
collection->deleteDocument(txn, rloc, false, false, logop ? &toDelete : NULL );
runner->restoreState(txn);
nDeleted++;
if (logop) {
if ( toDelete.isEmpty() ) {
problem() << "Deleted object without id in collection " << collection->ns()
<< ", not logging.";
}
else {
bool replJustOne = true;
replset::logOp(txn, "d", ns.ns().c_str(), toDelete, 0, &replJustOne);
}
}
if (!_request->isMulti()) {
break;
}
if (!_request->isGod()) {
txn->recoveryUnit()->commitIfNeeded();
}
if (debug && _request->isGod() && nDeleted == 100) {
log() << "warning high number of deletes with god=true "
<< " which could use significant memory b/c we don't commit journal";
}
}
return nDeleted;
}
TEST(MatchExpressionParserTest, ParseIntegerElementToLongAcceptsNegative) {
BSONObj query = BSON("" << -2LL);
auto result = MatchExpressionParser::parseIntegerElementToLong(query.firstElement());
ASSERT_OK(result.getStatus());
ASSERT_EQ(-2LL, result.getValue());
}
virtual bool run(OperationContext* txn,
const string& dbname,
BSONObj& cmdObj,
int,
string& errmsg,
BSONObjBuilder& result,
bool fromRepl) {
ScopedTransaction transaction(txn, MODE_X);
Lock::GlobalWrite globalWriteLock(txn->lockState());
string source = cmdObj.getStringField( name.c_str() );
string target = cmdObj.getStringField( "to" );
if (!fromRepl &&
!repl::getGlobalReplicationCoordinator()->canAcceptWritesForDatabase(dbname)) {
return appendCommandStatus(result, Status(ErrorCodes::NotMaster, str::stream()
<< "Not primary while renaming collection " << source << " to " << target));
}
// We stay in source context the whole time. This is mostly to set the CurOp namespace.
Client::Context ctx(txn, source);
if ( !NamespaceString::validCollectionComponent(target.c_str()) ) {
errmsg = "invalid collection name: " + target;
return false;
}
if ( source.empty() || target.empty() ) {
errmsg = "invalid command syntax";
return false;
}
if ((repl::getGlobalReplicationCoordinator()->getReplicationMode() !=
repl::ReplicationCoordinator::modeNone)) {
if (NamespaceString(source).isOplog()) {
errmsg = "can't rename live oplog while replicating";
return false;
}
if (NamespaceString(target).isOplog()) {
errmsg = "can't rename to live oplog while replicating";
return false;
}
}
if (NamespaceString::oplog(source) != NamespaceString::oplog(target)) {
errmsg =
"If either the source or target of a rename is an oplog name, both must be";
return false;
}
if (!fromRepl) { // If it got through on the master, need to allow it here too
Status sourceStatus = userAllowedWriteNS(source);
if (!sourceStatus.isOK()) {
errmsg = "error with source namespace: " + sourceStatus.reason();
return false;
}
Status targetStatus = userAllowedWriteNS(target);
if (!targetStatus.isOK()) {
errmsg = "error with target namespace: " + targetStatus.reason();
return false;
}
}
if (NamespaceString(source).coll() == "system.indexes"
|| NamespaceString(target).coll() == "system.indexes") {
errmsg = "renaming system.indexes is not allowed";
return false;
}
Database* const sourceDB = dbHolder().get(txn, nsToDatabase(source));
Collection* const sourceColl = sourceDB ? sourceDB->getCollection(source)
: NULL;
if (!sourceColl) {
errmsg = "source namespace does not exist";
return false;
}
{
// Ensure that collection name does not exceed maximum length.
// Ensure that index names do not push the length over the max.
// Iterator includes unfinished indexes.
IndexCatalog::IndexIterator sourceIndIt =
sourceColl->getIndexCatalog()->getIndexIterator( txn, true );
int longestIndexNameLength = 0;
while ( sourceIndIt.more() ) {
int thisLength = sourceIndIt.next()->indexName().length();
if ( thisLength > longestIndexNameLength )
longestIndexNameLength = thisLength;
}
unsigned int longestAllowed =
min(int(NamespaceString::MaxNsCollectionLen),
int(NamespaceString::MaxNsLen) - 2/*strlen(".$")*/ - longestIndexNameLength);
if (target.size() > longestAllowed) {
StringBuilder sb;
sb << "collection name length of " << target.size()
<< " exceeds maximum length of " << longestAllowed
<< ", allowing for index names";
errmsg = sb.str();
return false;
}
}
const std::vector<BSONObj> indexesInProg = stopIndexBuilds(txn, sourceDB, cmdObj);
// Dismissed on success
ScopeGuard indexBuildRestorer = MakeGuard(IndexBuilder::restoreIndexes,
txn,
indexesInProg);
Database* const targetDB = dbHolder().openDb(txn, nsToDatabase(target));
{
WriteUnitOfWork wunit(txn);
// Check if the target namespace exists and if dropTarget is true.
// If target exists and dropTarget is not true, return false.
if (targetDB->getCollection(target)) {
if (!cmdObj["dropTarget"].trueValue()) {
errmsg = "target namespace exists";
return false;
}
Status s = targetDB->dropCollection(txn, target);
if ( !s.isOK() ) {
errmsg = s.toString();
return false;
}
}
// If we are renaming in the same database, just
// rename the namespace and we're done.
if (sourceDB == targetDB) {
Status s = targetDB->renameCollection(txn,
source,
target,
cmdObj["stayTemp"].trueValue() );
if (!s.isOK()) {
return appendCommandStatus(result, s);
}
if (!fromRepl) {
repl::logOp(txn, "c", (dbname + ".$cmd").c_str(), cmdObj);
}
wunit.commit();
indexBuildRestorer.Dismiss();
return true;
}
wunit.commit();
}
// If we get here, we are renaming across databases, so we must copy all the data and
// indexes, then remove the source collection.
// Create the target collection. It will be removed if we fail to copy the collection.
// TODO use a temp collection and unset the temp flag on success.
Collection* targetColl = NULL;
{
CollectionOptions options;
options.setNoIdIndex();
if (sourceColl->isCapped()) {
const CollectionOptions sourceOpts =
sourceColl->getCatalogEntry()->getCollectionOptions(txn);
options.capped = true;
options.cappedSize = sourceOpts.cappedSize;
options.cappedMaxDocs = sourceOpts.cappedMaxDocs;
}
WriteUnitOfWork wunit(txn);
// No logOp necessary because the entire renameCollection command is one logOp.
targetColl = targetDB->createCollection(txn, target, options);
if (!targetColl) {
errmsg = "Failed to create target collection.";
return false;
}
wunit.commit();
}
// Dismissed on success
ScopeGuard targetCollectionDropper = MakeGuard(dropCollection, txn, targetDB, target);
MultiIndexBlock indexer(txn, targetColl);
indexer.allowInterruption();
// Copy the index descriptions from the source collection, adjusting the ns field.
{
std::vector<BSONObj> indexesToCopy;
IndexCatalog::IndexIterator sourceIndIt =
sourceColl->getIndexCatalog()->getIndexIterator( txn, true );
while (sourceIndIt.more()) {
const BSONObj currIndex = sourceIndIt.next()->infoObj();
// Process the source index.
BSONObjBuilder newIndex;
newIndex.append("ns", target);
newIndex.appendElementsUnique(currIndex);
indexesToCopy.push_back(newIndex.obj());
}
indexer.init(indexesToCopy);
}
{
// Copy over all the data from source collection to target collection.
boost::scoped_ptr<RecordIterator> sourceIt(sourceColl->getIterator(txn));
while (!sourceIt->isEOF()) {
txn->checkForInterrupt();
const Snapshotted<BSONObj> obj = sourceColl->docFor(txn, sourceIt->getNext());
WriteUnitOfWork wunit(txn);
// No logOp necessary because the entire renameCollection command is one logOp.
Status status =
targetColl->insertDocument(txn, obj.value(), &indexer, true).getStatus();
if (!status.isOK())
return appendCommandStatus(result, status);
wunit.commit();
}
}
Status status = indexer.doneInserting();
if (!status.isOK())
return appendCommandStatus(result, status);
{
// Getting here means we successfully built the target copy. We now remove the
// source collection and finalize the rename.
WriteUnitOfWork wunit(txn);
Status status = sourceDB->dropCollection(txn, source);
if (!status.isOK())
return appendCommandStatus(result, status);
indexer.commit();
if (!fromRepl) {
repl::logOp(txn, "c", (dbname + ".$cmd").c_str(), cmdObj);
}
wunit.commit();
}
indexBuildRestorer.Dismiss();
targetCollectionDropper.Dismiss();
return true;
}
std::string runQuery(OperationContext* txn,
QueryMessage& q,
const NamespaceString& nss,
Message& result) {
CurOp& curop = *CurOp::get(txn);
// Validate the namespace.
uassert(16256, str::stream() << "Invalid ns [" << nss.ns() << "]", nss.isValid());
invariant(!nss.isCommand());
// Set curop information.
beginQueryOp(txn, nss, q.query, q.ntoreturn, q.ntoskip);
// Parse the qm into a CanonicalQuery.
auto statusWithCQ = CanonicalQuery::canonicalize(q, WhereCallbackReal(txn, nss.db()));
if (!statusWithCQ.isOK()) {
uasserted(
17287,
str::stream() << "Can't canonicalize query: " << statusWithCQ.getStatus().toString());
}
unique_ptr<CanonicalQuery> cq = std::move(statusWithCQ.getValue());
invariant(cq.get());
LOG(5) << "Running query:\n" << cq->toString();
LOG(2) << "Running query: " << cq->toStringShort();
// Parse, canonicalize, plan, transcribe, and get a plan executor.
AutoGetCollectionForRead ctx(txn, nss);
Collection* collection = ctx.getCollection();
const int dbProfilingLevel =
ctx.getDb() ? ctx.getDb()->getProfilingLevel() : serverGlobalParams.defaultProfile;
// We have a parsed query. Time to get the execution plan for it.
std::unique_ptr<PlanExecutor> exec = uassertStatusOK(
getExecutorFind(txn, collection, nss, std::move(cq), PlanExecutor::YIELD_AUTO));
const LiteParsedQuery& pq = exec->getCanonicalQuery()->getParsed();
// If it's actually an explain, do the explain and return rather than falling through
// to the normal query execution loop.
if (pq.isExplain()) {
BufBuilder bb;
bb.skip(sizeof(QueryResult::Value));
BSONObjBuilder explainBob;
Explain::explainStages(exec.get(), ExplainCommon::EXEC_ALL_PLANS, &explainBob);
// Add the resulting object to the return buffer.
BSONObj explainObj = explainBob.obj();
bb.appendBuf((void*)explainObj.objdata(), explainObj.objsize());
// TODO: Does this get overwritten/do we really need to set this twice?
curop.debug().query = q.query;
// Set query result fields.
QueryResult::View qr = bb.buf();
bb.decouple();
qr.setResultFlagsToOk();
qr.msgdata().setLen(bb.len());
curop.debug().responseLength = bb.len();
qr.msgdata().setOperation(opReply);
qr.setCursorId(0);
qr.setStartingFrom(0);
qr.setNReturned(1);
result.setData(qr.view2ptr(), true);
return "";
}
ShardingState* const shardingState = ShardingState::get(txn);
// We freak out later if this changes before we're done with the query.
const ChunkVersion shardingVersionAtStart = shardingState->getVersion(nss.ns());
// Handle query option $maxTimeMS (not used with commands).
curop.setMaxTimeMicros(static_cast<unsigned long long>(pq.getMaxTimeMS()) * 1000);
txn->checkForInterrupt(); // May trigger maxTimeAlwaysTimeOut fail point.
// uassert if we are not on a primary, and not a secondary with SlaveOk query parameter set.
bool slaveOK = pq.isSlaveOk() || pq.hasReadPref();
Status serveReadsStatus =
repl::getGlobalReplicationCoordinator()->checkCanServeReadsFor(txn, nss, slaveOK);
uassertStatusOK(serveReadsStatus);
// Run the query.
// bb is used to hold query results
// this buffer should contain either requested documents per query or
// explain information, but not both
BufBuilder bb(32768);
bb.skip(sizeof(QueryResult::Value));
// How many results have we obtained from the executor?
int numResults = 0;
// If we're replaying the oplog, we save the last time that we read.
Timestamp slaveReadTill;
BSONObj obj;
PlanExecutor::ExecState state;
// uint64_t numMisplacedDocs = 0;
// Get summary info about which plan the executor is using.
curop.debug().planSummary = Explain::getPlanSummary(exec.get());
while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) {
// Add result to output buffer.
bb.appendBuf((void*)obj.objdata(), obj.objsize());
// Count the result.
++numResults;
// Possibly note slave's position in the oplog.
if (pq.isOplogReplay()) {
BSONElement e = obj["ts"];
if (Date == e.type() || bsonTimestamp == e.type()) {
slaveReadTill = e.timestamp();
}
}
if (FindCommon::enoughForFirstBatch(pq, numResults, bb.len())) {
LOG(5) << "Enough for first batch, wantMore=" << pq.wantMore()
<< " ntoreturn=" << pq.getNToReturn().value_or(0) << " numResults=" << numResults
<< endl;
break;
}
}
// If we cache the executor later, we want to deregister it as it receives notifications
// anyway by virtue of being cached.
//
// If we don't cache the executor later, we are deleting it, so it must be deregistered.
//
// So, no matter what, deregister the executor.
exec->deregisterExec();
// Caller expects exceptions thrown in certain cases.
if (PlanExecutor::FAILURE == state || PlanExecutor::DEAD == state) {
const unique_ptr<PlanStageStats> stats(exec->getStats());
error() << "Plan executor error during find: " << PlanExecutor::statestr(state)
<< ", stats: " << Explain::statsToBSON(*stats);
uasserted(17144, "Executor error: " + WorkingSetCommon::toStatusString(obj));
}
// TODO: Currently, chunk ranges are kept around until all ClientCursors created while the
// chunk belonged on this node are gone. Separating chunk lifetime management from
// ClientCursor should allow this check to go away.
if (!shardingState->getVersion(nss.ns()).isWriteCompatibleWith(shardingVersionAtStart)) {
// if the version changed during the query we might be missing some data and its safe to
// send this as mongols can resend at this point
throw SendStaleConfigException(nss.ns(),
"version changed during initial query",
shardingVersionAtStart,
shardingState->getVersion(nss.ns()));
}
// Fill out curop based on query results. If we have a cursorid, we will fill out curop with
// this cursorid later.
long long ccId = 0;
if (shouldSaveCursor(txn, collection, state, exec.get())) {
// We won't use the executor until it's getMore'd.
exec->saveState();
exec->detachFromOperationContext();
// Allocate a new ClientCursor. We don't have to worry about leaking it as it's
// inserted into a global map by its ctor.
ClientCursor* cc =
new ClientCursor(collection->getCursorManager(),
exec.release(),
nss.ns(),
txn->recoveryUnit()->isReadingFromMajorityCommittedSnapshot(),
pq.getOptions(),
pq.getFilter());
ccId = cc->cursorid();
LOG(5) << "caching executor with cursorid " << ccId << " after returning " << numResults
<< " results" << endl;
// TODO document
if (pq.isOplogReplay() && !slaveReadTill.isNull()) {
cc->slaveReadTill(slaveReadTill);
}
// TODO document
if (pq.isExhaust()) {
curop.debug().exhaust = true;
}
cc->setPos(numResults);
// If the query had a time limit, remaining time is "rolled over" to the cursor (for
// use by future getmore ops).
cc->setLeftoverMaxTimeMicros(curop.getRemainingMaxTimeMicros());
endQueryOp(txn, collection, *cc->getExecutor(), dbProfilingLevel, numResults, ccId);
} else {
LOG(5) << "Not caching executor but returning " << numResults << " results.\n";
endQueryOp(txn, collection, *exec, dbProfilingLevel, numResults, ccId);
}
// Add the results from the query into the output buffer.
result.appendData(bb.buf(), bb.len());
bb.decouple();
// Fill out the output buffer's header.
QueryResult::View qr = result.header().view2ptr();
qr.setCursorId(ccId);
qr.setResultFlagsToOk();
qr.msgdata().setOperation(opReply);
qr.setStartingFrom(0);
qr.setNReturned(numResults);
// curop.debug().exhaust is set above.
return curop.debug().exhaust ? nss.ns() : "";
}
/**
* The core config write functionality.
*
* Config writes run in two passes - the first is a quick check to ensure the config servers
* are all reachable, the second runs the actual write.
*
* TODO: Upgrade and move this logic to the config servers, a state machine implementation
* is probably the next step.
*/
void ConfigCoordinator::executeBatch(const BatchedCommandRequest& clientRequest,
BatchedCommandResponse* clientResponse) {
const NamespaceString nss(clientRequest.getNS());
// Should never use it for anything other than DBs residing on the config server
dassert(nss.db() == "config" || nss.db() == "admin");
dassert(clientRequest.sizeWriteOps() == 1u);
// This is an opportunistic check that all config servers look healthy by calling
// getLastError on each one of them. If there was some form of write/journaling error, get
// last error would fail.
{
for (vector<ConnectionString>::iterator it = _configHosts.begin();
it != _configHosts.end();
++it) {
_dispatcher->addCommand(*it,
"admin",
RawBSONSerializable(BSON("getLastError" << true <<
"fsync" << true)));
}
_dispatcher->sendAll();
bool error = false;
while (_dispatcher->numPending()) {
ConnectionString host;
RawBSONSerializable response;
Status status = _dispatcher->recvAny(&host, &response);
if (status.isOK()) {
BSONObj obj = response.toBSON();
LOG(3) << "Response " << obj.toString();
// If the ok field is anything other than 1, count it as error
if (!obj["ok"].trueValue()) {
error = true;
log() << "Config server check for host " << host
<< " returned error: " << response;
}
}
else {
error = true;
log() << "Config server check for host " << host
<< " failed with status: " << status;
}
}
// All responses should have been gathered by this point
if (error) {
clientResponse->setOk(false);
clientResponse->setErrCode(ErrorCodes::RemoteValidationError);
clientResponse->setErrMessage("Could not verify that config servers were active"
" and reachable before write");
return;
}
}
if (!_checkConfigString(clientResponse)) {
return;
}
//
// Do the actual writes
//
BatchedCommandRequest configRequest( clientRequest.getBatchType() );
clientRequest.cloneTo( &configRequest );
configRequest.setNS( nss.coll() );
OwnedPointerVector<ConfigResponse> responsesOwned;
vector<ConfigResponse*>& responses = responsesOwned.mutableVector();
//
// Send the actual config writes
//
// Get as many batches as we can at once
for (vector<ConnectionString>::const_iterator it = _configHosts.begin();
it != _configHosts.end();
++it) {
const ConnectionString& configHost = *it;
_dispatcher->addCommand(configHost, nss.db(), configRequest);
}
// Send them all out
_dispatcher->sendAll();
//
// Recv side
//
while (_dispatcher->numPending() > 0) {
// Get the response
responses.push_back(new ConfigResponse());
ConfigResponse& configResponse = *responses.back();
Status dispatchStatus = _dispatcher->recvAny(&configResponse.configHost,
&configResponse.response);
if (!dispatchStatus.isOK()) {
buildErrorFrom(dispatchStatus, &configResponse.response);
}
}
combineResponses(responses, clientResponse);
}
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();
}
bool ClientInfo::getLastError( const BSONObj& options , BSONObjBuilder& result , bool fromWriteBackListener ) {
set<string> * shards = getPrev();
if ( shards->size() == 0 ) {
result.appendNull( "err" );
return true;
}
vector<WBInfo> writebacks;
// handle single server
if ( shards->size() == 1 ) {
string theShard = *(shards->begin() );
ShardConnection conn( theShard , "" );
BSONObj res;
bool ok = false;
try {
ok = conn->runCommand( "admin" , options , res );
}
catch( std::exception &e ) {
warning() << "could not get last error from shard " << theShard << causedBy( e ) << endl;
// Catch everything that happens here, since we need to ensure we return our connection when we're
// finished.
conn.done();
return false;
}
res = res.getOwned();
conn.done();
_addWriteBack( writebacks , res );
// hit other machines just to block
for ( set<string>::const_iterator i=sinceLastGetError().begin(); i!=sinceLastGetError().end(); ++i ) {
string temp = *i;
if ( temp == theShard )
continue;
ShardConnection conn( temp , "" );
try {
_addWriteBack( writebacks , conn->getLastErrorDetailed() );
}
catch( std::exception &e ) {
warning() << "could not clear last error from shard " << temp << causedBy( e ) << endl;
}
conn.done();
}
clearSinceLastGetError();
if ( writebacks.size() ) {
vector<BSONObj> v = _handleWriteBacks( writebacks , fromWriteBackListener );
if ( v.size() == 0 && fromWriteBackListener ) {
// ok
}
else {
// this will usually be 1
// it can be greater than 1 if a write to a different shard
// than the last write op had a writeback
// all we're going to report is the first
// since that's the current write
// but we block for all
assert( v.size() >= 1 );
result.appendElements( v[0] );
result.appendElementsUnique( res );
result.append( "writebackGLE" , v[0] );
result.append( "initialGLEHost" , theShard );
}
}
else {
result.append( "singleShard" , theShard );
result.appendElements( res );
}
return ok;
}
BSONArrayBuilder bbb( result.subarrayStart( "shards" ) );
BSONObjBuilder shardRawGLE;
long long n = 0;
int updatedExistingStat = 0; // 0 is none, -1 has but false, 1 has true
// hit each shard
vector<string> errors;
vector<BSONObj> errorObjects;
for ( set<string>::iterator i = shards->begin(); i != shards->end(); i++ ) {
string theShard = *i;
bbb.append( theShard );
ShardConnection conn( theShard , "" );
BSONObj res;
bool ok = false;
try {
ok = conn->runCommand( "admin" , options , res );
shardRawGLE.append( theShard , res );
}
catch( std::exception &e ) {
// Safe to return here, since we haven't started any extra processing yet, just collecting
// responses.
warning() << "could not get last error from a shard " << theShard << causedBy( e ) << endl;
conn.done();
return false;
}
_addWriteBack( writebacks, res );
string temp = DBClientWithCommands::getLastErrorString( res );
if ( conn->type() != ConnectionString::SYNC && ( ok == false || temp.size() ) ) {
errors.push_back( temp );
errorObjects.push_back( res );
}
n += res["n"].numberLong();
if ( res["updatedExisting"].type() ) {
if ( res["updatedExisting"].trueValue() )
updatedExistingStat = 1;
else if ( updatedExistingStat == 0 )
updatedExistingStat = -1;
}
conn.done();
}
bbb.done();
result.append( "shardRawGLE" , shardRawGLE.obj() );
result.appendNumber( "n" , n );
if ( updatedExistingStat )
result.appendBool( "updatedExisting" , updatedExistingStat > 0 );
// hit other machines just to block
for ( set<string>::const_iterator i=sinceLastGetError().begin(); i!=sinceLastGetError().end(); ++i ) {
string temp = *i;
if ( shards->count( temp ) )
continue;
ShardConnection conn( temp , "" );
try {
_addWriteBack( writebacks, conn->getLastErrorDetailed() );
}
catch( std::exception &e ) {
warning() << "could not clear last error from a shard " << temp << causedBy( e ) << endl;
}
conn.done();
}
clearSinceLastGetError();
if ( errors.size() == 0 ) {
result.appendNull( "err" );
_handleWriteBacks( writebacks , fromWriteBackListener );
return true;
}
result.append( "err" , errors[0].c_str() );
{
// errs
BSONArrayBuilder all( result.subarrayStart( "errs" ) );
for ( unsigned i=0; i<errors.size(); i++ ) {
all.append( errors[i].c_str() );
}
all.done();
}
{
// errObjects
BSONArrayBuilder all( result.subarrayStart( "errObjects" ) );
for ( unsigned i=0; i<errorObjects.size(); i++ ) {
all.append( errorObjects[i] );
}
all.done();
}
_handleWriteBacks( writebacks , fromWriteBackListener );
return true;
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
if ( ! okForConfigChanges( errmsg ) )
return false;
ShardConnection::sync();
string ns = cmdObj.firstElement().valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "no ns";
return false;
}
DBConfigPtr config = grid.getDBConfig( ns );
if ( ! config->isSharded( ns ) ) {
config->reload();
if ( ! config->isSharded( ns ) ) {
errmsg = "ns not sharded. have to shard before can split";
return false;
}
}
BSONObj find = cmdObj.getObjectField( "find" );
if ( find.isEmpty() ) {
find = cmdObj.getObjectField( "middle" );
if ( find.isEmpty() ) {
errmsg = "need to specify find or middle";
return false;
}
}
ChunkManagerPtr info = config->getChunkManager( ns );
ChunkPtr chunk = info->findChunk( find );
BSONObj middle = cmdObj.getObjectField( "middle" );
assert( chunk.get() );
log() << "splitting: " << ns << " shard: " << chunk << endl;
BSONObj res;
bool worked;
if ( middle.isEmpty() ) {
BSONObj ret = chunk->singleSplit( true /* force a split even if not enough data */ , res );
worked = !ret.isEmpty();
}
else {
// sanity check if the key provided is a valid split point
if ( ( middle == chunk->getMin() ) || ( middle == chunk->getMax() ) ) {
errmsg = "cannot split on initial or final chunk's key";
return false;
}
if (!fieldsMatch(middle, info->getShardKey().key())){
errmsg = "middle has different fields (or different order) than shard key";
return false;
}
vector<BSONObj> splitPoints;
splitPoints.push_back( middle );
worked = chunk->multiSplit( splitPoints , res );
}
if ( !worked ) {
errmsg = "split failed";
result.append( "cause" , res );
return false;
}
config->getChunkManager( ns , true );
return true;
}
/* ****************************************************************************
*
* runCollectionCommand -
*
* NOTE
* Different from other functions in this module, this function can get the connection
* in the params, instead of using getMongoConnection().
* This is only done from DB connection bootstrapping code .
*
*/
bool runCollectionCommand
(
DBClientBase* connection,
const std::string& col,
const BSONObj& command,
BSONObj* result,
std::string* err
)
{
bool releaseConnection = false;
//
// The call to TIME_STAT_MONGO_COMMAND_WAIT_START must be on toplevel so that local variables
// are visible for TIME_STAT_MONGO_COMMAND_WAIT_STOP()
//
TIME_STAT_MONGO_COMMAND_WAIT_START();
if (connection == NULL)
{
connection = getMongoConnection();
releaseConnection = true;
if (connection == NULL)
{
TIME_STAT_MONGO_COMMAND_WAIT_STOP();
LM_E(("Fatal Error (null DB connection)"));
return false;
}
}
LM_T(LmtMongo, ("runCommand() in '%s' collection: '%s'", col.c_str(), command.toString().c_str()));
try
{
connection->runCommand(col.c_str(), command, *result);
if (releaseConnection)
{
releaseMongoConnection(connection);
TIME_STAT_MONGO_COMMAND_WAIT_STOP();
}
LM_I(("Database Operation Successful (command: %s)", command.toString().c_str()));
}
catch (const std::exception &e)
{
if (releaseConnection)
{
releaseMongoConnection(connection);
TIME_STAT_MONGO_COMMAND_WAIT_STOP();
}
std::string msg = std::string("collection: ") + col.c_str() +
" - runCommand(): " + command.toString() +
" - exception: " + e.what();
*err = "Database Error (" + msg + ")";
LM_E((err->c_str()));
return false;
}
catch (...)
{
if (releaseConnection)
{
releaseMongoConnection(connection);
TIME_STAT_MONGO_COMMAND_WAIT_STOP();
}
std::string msg = std::string("collection: ") + col.c_str() +
" - runCommand(): " + command.toString() +
" - exception: generic";
*err = "Database Error (" + msg + ")";
LM_E((err->c_str()));
return false;
}
return true;
}