/** caller handles locking */
static bool PREPLOGBUFFER(BufBuilder& bb) {
if( writes.empty() )
return false;
bb.reset();
unsigned *lenInBlockHeader;
{
// JSectHeader
bb.appendStr("\nHH\n", false);
lenInBlockHeader = (unsigned *) bb.skip(4);
}
string lastFilePath;
{
scoped_lock lk(privateViews._mutex());
for( vector<WriteIntent>::iterator i = writes.begin(); i != writes.end(); i++ ) {
size_t ofs;
MongoMMF *mmf = privateViews._find(i->p, ofs);
if( mmf == 0 ) {
journalingFailure("view pointer cannot be resolved");
}
else {
if( mmf->filePath() != lastFilePath ) {
lastFilePath = mmf->filePath();
JDbContext c;
bb.appendStruct(c);
bb.appendStr(lastFilePath);
}
JEntry e;
e.len = i->len;
e.fileNo = mmf->fileSuffixNo();
bb.appendStruct(e);
bb.appendBuf(i->p, i->len);
}
}
}
{
JSectFooter f;
f.hash = 0;
bb.appendStruct(f);
}
{
unsigned L = (bb.len() + 8191) & 0xffffe000; // fill to alignment
dassert( L >= (unsigned) bb.len() );
*lenInBlockHeader = L;
unsigned padding = L - bb.len();
bb.skip(padding);
dassert( bb.len() % 8192 == 0 );
}
writes.clear();
alreadyNoted.clear();
return true;
}
static bool receivedQuery(Client& c, DbResponse& dbresponse, Message& m ){
bool ok = true;
MSGID responseTo = m.header()->id;
DbMessage d(m);
QueryMessage q(d);
auto_ptr< Message > resp( new Message() );
CurOp& op = *(c.curop());
try {
dbresponse.exhaust = runQuery(m, q, op, *resp);
assert( !resp->empty() );
}
catch ( AssertionException& e ) {
ok = false;
op.debug().str << " exception ";
LOGSOME {
log() << "assertion " << e.toString() << " ns:" << q.ns << " query:" <<
(q.query.valid() ? q.query.toString() : "query object is corrupt") << endl;
if( q.ntoskip || q.ntoreturn )
log() << " ntoskip:" << q.ntoskip << " ntoreturn:" << q.ntoreturn << endl;
}
BSONObjBuilder err;
e.getInfo().append( err );
BSONObj errObj = err.done();
BufBuilder b;
b.skip(sizeof(QueryResult));
b.appendBuf((void*) errObj.objdata(), errObj.objsize());
// todo: call replyToQuery() from here instead of this!!! see dbmessage.h
QueryResult * msgdata = (QueryResult *) b.buf();
b.decouple();
QueryResult *qr = msgdata;
qr->_resultFlags() = ResultFlag_ErrSet;
if ( e.getCode() == StaleConfigInContextCode )
qr->_resultFlags() |= ResultFlag_ShardConfigStale;
qr->len = b.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
resp.reset( new Message() );
resp->setData( msgdata, true );
}
if ( op.shouldDBProfile( 0 ) ){
op.debug().str << " bytes:" << resp->header()->dataLen();
}
dbresponse.response = resp.release();
dbresponse.responseTo = responseTo;
return ok;
}
std::unique_ptr<Message> LegacyReplyBuilder::done() {
invariant(_state == State::kOutputDocs);
BSONObj reply = uassertStatusOK(rpc::downconvertReplyMetadata(_commandReply, _metadata));
BufBuilder bufBuilder;
bufBuilder.skip(sizeof(QueryResult::Value));
if (_allowAddingOutputDocs) {
BSONObjBuilder topBuilder(bufBuilder);
for (const auto& el : reply) {
if (kCursorTag != el.fieldNameStringData()) {
topBuilder.append(el);
continue;
}
invariant(el.isABSONObj());
BSONObjBuilder curBuilder(topBuilder.subobjStart(kCursorTag));
for (const auto& insideEl : el.Obj()) {
if (kFirstBatchTag != insideEl.fieldNameStringData()) {
curBuilder.append(insideEl);
continue;
}
invariant(insideEl.isABSONObj());
BSONArrayBuilder arrBuilder(curBuilder.subarrayStart(kFirstBatchTag));
for (const auto& doc : _outputDocs) {
arrBuilder.append(doc);
}
arrBuilder.doneFast();
}
curBuilder.doneFast();
}
topBuilder.doneFast();
} else {
reply.appendSelfToBufBuilder(bufBuilder);
}
auto msgHeaderSz = static_cast<std::size_t>(MsgData::MsgDataHeaderSize);
invariant(static_cast<std::size_t>(bufBuilder.len()) + msgHeaderSz <=
mongol::MaxMessageSizeBytes);
QueryResult::View qr = bufBuilder.buf();
qr.setResultFlagsToOk();
qr.msgdata().setLen(bufBuilder.len());
qr.msgdata().setOperation(opReply);
qr.setCursorId(0);
qr.setStartingFrom(0);
qr.setNReturned(1);
_message->setData(qr.view2ptr(), true);
bufBuilder.decouple();
_state = State::kDone;
return std::move(_message);
}
void generateLegacyQueryErrorResponse(const AssertionException* exception,
const QueryMessage& queryMessage,
CurOp* curop,
Message* response) {
curop->debug().exceptionInfo = exception->getInfo();
log(LogComponent::kQuery) << "assertion " << exception->toString() << " ns:" << queryMessage.ns
<< " query:" << (queryMessage.query.valid()
? queryMessage.query.toString()
: "query object is corrupt");
if (queryMessage.ntoskip || queryMessage.ntoreturn) {
log(LogComponent::kQuery) << " ntoskip:" << queryMessage.ntoskip
<< " ntoreturn:" << queryMessage.ntoreturn;
}
const SendStaleConfigException* scex = (exception->getCode() == ErrorCodes::SendStaleConfig)
? static_cast<const SendStaleConfigException*>(exception)
: NULL;
BSONObjBuilder err;
exception->getInfo().append(err);
if (scex) {
err.append("ok", 0.0);
err.append("ns", scex->getns());
scex->getVersionReceived().addToBSON(err, "vReceived");
scex->getVersionWanted().addToBSON(err, "vWanted");
}
BSONObj errObj = err.done();
if (scex) {
log(LogComponent::kQuery) << "stale version detected during query over " << queryMessage.ns
<< " : " << errObj;
}
BufBuilder bb;
bb.skip(sizeof(QueryResult::Value));
bb.appendBuf((void*)errObj.objdata(), errObj.objsize());
// TODO: call replyToQuery() from here instead of this!!! see dbmessage.h
QueryResult::View msgdata = bb.buf();
bb.decouple();
QueryResult::View qr = msgdata;
qr.setResultFlags(ResultFlag_ErrSet);
if (scex)
qr.setResultFlags(qr.getResultFlags() | ResultFlag_ShardConfigStale);
qr.msgdata().setLen(bb.len());
qr.msgdata().setOperation(opReply);
qr.setCursorId(0);
qr.setStartingFrom(0);
qr.setNReturned(1);
response->setData(msgdata.view2ptr(), true);
}
Message opCommandRequestFromOpMsgRequest(const OpMsgRequest& request) {
const auto commandName = request.getCommandName();
BufBuilder builder;
builder.skip(mongo::MsgData::MsgDataHeaderSize); // Leave room for message header.
builder.appendStr(request.getDatabase());
builder.appendStr(commandName);
// OP_COMMAND is only used when communicating with 3.4 nodes and they serialize their metadata
// fields differently. In addition to field-level differences, some generic arguments are pulled
// out to a metadata object, separate from the body. We do all down-conversion here so that the
// rest of the code only has to deal with the current format.
BSONObjBuilder metadataBuilder; // Will be appended to the message after we finish the body.
{
BSONObjBuilder bodyBuilder(builder);
for (auto elem : request.body) {
const auto fieldName = elem.fieldNameStringData();
if (fieldName == "$configServerState") {
metadataBuilder.appendAs(elem, "configsvr");
} else if (fieldName == "$readPreference") {
BSONObjBuilder ssmBuilder(metadataBuilder.subobjStart("$ssm"));
ssmBuilder.append(elem);
ssmBuilder.append("$secondaryOk",
uassertStatusOK(ReadPreferenceSetting::fromInnerBSON(elem))
.canRunOnSecondary());
} else if (fieldName == "$db") {
// skip
} else if (fieldGoesInMetadata(commandName, fieldName)) {
metadataBuilder.append(elem);
} else {
bodyBuilder.append(elem);
}
}
for (auto&& seq : request.sequences) {
invariant(seq.name.find('.') == std::string::npos); // Only support top-level for now.
dassert(!bodyBuilder.asTempObj().hasField(seq.name));
bodyBuilder.append(seq.name, seq.objs);
}
}
metadataBuilder.obj().appendSelfToBufBuilder(builder);
MsgData::View msg = builder.buf();
msg.setLen(builder.len());
msg.setOperation(dbCommand);
return Message(builder.release());
}
void replyToQuery(int queryResultFlags, Message &m, DbResponse &dbresponse, BSONObj obj) {
BufBuilder b;
b.skip(sizeof(QueryResult));
b.appendBuf((void*) obj.objdata(), obj.objsize());
QueryResult* msgdata = (QueryResult *) b.buf();
b.decouple();
QueryResult *qr = msgdata;
qr->_resultFlags() = queryResultFlags;
qr->len = b.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
Message *resp = new Message();
resp->setData(msgdata, true); // transport will free
dbresponse.response = resp;
dbresponse.responseTo = m.header()->id;
}
void replyToQuery( int queryResultFlags, Message& response, const BSONObj& resultObj ) {
BufBuilder bufBuilder;
bufBuilder.skip( sizeof( QueryResult::Value ));
bufBuilder.appendBuf( reinterpret_cast< void *>(
const_cast< char* >( resultObj.objdata() )), resultObj.objsize() );
QueryResult::View queryResult = bufBuilder.buf();
bufBuilder.decouple();
queryResult.setResultFlags(queryResultFlags);
queryResult.msgdata().setLen(bufBuilder.len());
queryResult.msgdata().setOperation( opReply );
queryResult.setCursorId(0);
queryResult.setStartingFrom(0);
queryResult.setNReturned(1);
response.setData( queryResult.view2ptr(), true ); // transport will free
}
void replyToQuery( int queryResultFlags, Message& response, const BSONObj& resultObj ) {
BufBuilder bufBuilder;
bufBuilder.skip( sizeof( QueryResult ));
bufBuilder.appendBuf( reinterpret_cast< void *>(
const_cast< char* >( resultObj.objdata() )), resultObj.objsize() );
QueryResult* queryResult = reinterpret_cast< QueryResult* >( bufBuilder.buf() );
bufBuilder.decouple();
queryResult->_resultFlags() = queryResultFlags;
queryResult->len = bufBuilder.len();
queryResult->setOperation( opReply );
queryResult->cursorId = 0;
queryResult->startingFrom = 0;
queryResult->nReturned = 1;
response.setData( queryResult, true ); // transport will free
}
static bool receivedQuery(Client& c, DbResponse& dbresponse, Message& m ) {
bool ok = true;
MSGID responseTo = m.header()->id;
DbMessage d(m);
QueryMessage q(d);
auto_ptr< Message > resp( new Message() );
CurOp& op = *(c.curop());
shared_ptr<AssertionException> ex;
try {
dbresponse.exhaust = runQuery(m, q, op, *resp);
assert( !resp->empty() );
}
catch ( SendStaleConfigException& e ){
ex.reset( new SendStaleConfigException( e.getns(), e.getInfo().msg ) );
ok = false;
}
catch ( AssertionException& e ) {
ex.reset( new AssertionException( e.getInfo().msg, e.getCode() ) );
ok = false;
}
if( ex ){
op.debug().exceptionInfo = ex->getInfo();
LOGWITHRATELIMIT {
log() << "assertion " << ex->toString() << " ns:" << q.ns << " query:" <<
(q.query.valid() ? q.query.toString() : "query object is corrupt") << endl;
if( q.ntoskip || q.ntoreturn )
log() << " ntoskip:" << q.ntoskip << " ntoreturn:" << q.ntoreturn << endl;
}
SendStaleConfigException* scex = NULL;
if ( ex->getCode() == SendStaleConfigCode ) scex = static_cast<SendStaleConfigException*>( ex.get() );
BSONObjBuilder err;
ex->getInfo().append( err );
if( scex ) err.append( "ns", scex->getns() );
BSONObj errObj = err.done();
log() << errObj << endl;
BufBuilder b;
b.skip(sizeof(QueryResult));
b.appendBuf((void*) errObj.objdata(), errObj.objsize());
// todo: call replyToQuery() from here instead of this!!! see dbmessage.h
QueryResult * msgdata = (QueryResult *) b.buf();
b.decouple();
QueryResult *qr = msgdata;
qr->_resultFlags() = ResultFlag_ErrSet;
if( scex ) qr->_resultFlags() |= ResultFlag_ShardConfigStale;
qr->len = b.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
resp.reset( new Message() );
resp->setData( msgdata, true );
}
op.debug().responseLength = resp->header()->dataLen();
dbresponse.response = resp.release();
dbresponse.responseTo = responseTo;
return ok;
}
void receivedQuery(DbResponse& dbresponse, /*AbstractMessagingPort& dbMsgPort, */Message& m, stringstream& ss, bool logit) {
MSGID responseTo = m.data->id;
DbMessage d(m);
QueryMessage q(d);
QueryResult* msgdata;
try {
if (q.fields.get() && q.fields->errmsg)
uassert(q.fields->errmsg, false);
/* note these are logged BEFORE authentication -- which is sort of ok */
if ( _diaglog.level && logit ) {
if ( strstr(q.ns, ".$cmd") ) {
/* $cmd queries are "commands" and usually best treated as write operations */
OPWRITE;
}
else {
OPREAD;
}
}
setClient( q.ns );
Client& client = cc();
client.top.setRead();
strncpy(client.curop()->ns, q.ns, Namespace::MaxNsLen);
msgdata = runQuery(m, ss ).release();
}
catch ( AssertionException& e ) {
ss << " exception ";
LOGSOME problem() << " Caught Assertion in runQuery ns:" << q.ns << ' ' << e.toString() << '\n';
log() << " ntoskip:" << q.ntoskip << " ntoreturn:" << q.ntoreturn << '\n';
if ( q.query.valid() )
log() << " query:" << q.query.toString() << endl;
else
log() << " query object is not valid!" << endl;
BSONObjBuilder err;
err.append("$err", e.msg.empty() ? "assertion during query" : e.msg);
BSONObj errObj = err.done();
BufBuilder b;
b.skip(sizeof(QueryResult));
b.append((void*) errObj.objdata(), errObj.objsize());
// todo: call replyToQuery() from here instead of this!!! see dbmessage.h
msgdata = (QueryResult *) b.buf();
b.decouple();
QueryResult *qr = msgdata;
qr->resultFlags() = QueryResult::ResultFlag_ErrSet;
qr->len = b.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
}
Message *resp = new Message();
resp->setData(msgdata, true); // transport will free
dbresponse.response = resp;
dbresponse.responseTo = responseTo;
Database *database = cc().database();
if ( database ) {
if ( database->profile )
ss << " bytes:" << resp->data->dataLen();
}
else {
if ( strstr(q.ns, "$cmd") == 0 ) // (this condition is normal for $cmd dropDatabase)
log() << "ERROR: receiveQuery: database is null; ns=" << q.ns << endl;
}
}
std::string newRunQuery(OperationContext* txn,
Message& m,
QueryMessage& q,
CurOp& curop,
Message &result,
bool fromDBDirectClient) {
// Validate the namespace.
const char *ns = q.ns;
uassert(16332, "can't have an empty ns", ns[0]);
const NamespaceString nsString(ns);
uassert(16256, str::stream() << "Invalid ns [" << ns << "]", nsString.isValid());
// Set curop information.
curop.debug().ns = ns;
curop.debug().ntoreturn = q.ntoreturn;
curop.debug().query = q.query;
curop.setQuery(q.query);
// If the query is really a command, run it.
if (nsString.isCommand()) {
int nToReturn = q.ntoreturn;
uassert(16979, str::stream() << "bad numberToReturn (" << nToReturn
<< ") for $cmd type ns - can only be 1 or -1",
nToReturn == 1 || nToReturn == -1);
curop.markCommand();
BufBuilder bb;
bb.skip(sizeof(QueryResult::Value));
BSONObjBuilder cmdResBuf;
if (!runCommands(txn, ns, q.query, curop, bb, cmdResBuf, false, q.queryOptions)) {
uasserted(13530, "bad or malformed command request?");
}
curop.debug().iscommand = true;
// TODO: Does this get overwritten/do we really need to set this twice?
curop.debug().query = q.query;
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 "";
}
const NamespaceString nss(q.ns);
// Parse the qm into a CanonicalQuery.
CanonicalQuery* cq;
Status canonStatus = CanonicalQuery::canonicalize(
q, &cq, WhereCallbackReal(txn, StringData(nss.db())));
if (!canonStatus.isOK()) {
uasserted(17287, str::stream() << "Can't canonicalize query: " << canonStatus.toString());
}
QLOG() << "Running query:\n" << cq->toString();
LOG(2) << "Running query: " << cq->toStringShort();
// Parse, canonicalize, plan, transcribe, and get a plan executor.
PlanExecutor* rawExec = NULL;
// We use this a lot below.
const LiteParsedQuery& pq = cq->getParsed();
AutoGetCollectionForRead ctx(txn, nss);
const int dbProfilingLevel = (ctx.getDb() != NULL) ? ctx.getDb()->getProfilingLevel() :
serverGlobalParams.defaultProfile;
Collection* collection = ctx.getCollection();
// We'll now try to get the query executor that will execute this query for us. There
// are a few cases in which we know upfront which executor we should get and, therefore,
// we shortcut the selection process here.
//
// (a) If the query is over a collection that doesn't exist, we use an EOFStage.
//
// (b) if the query is a replication's initial sync one, we use a specifically designed
// stage that skips extents faster (see details in exec/oplogstart.h).
//
// Otherwise we go through the selection of which executor is most suited to the
// query + run-time context at hand.
Status status = Status::OK();
if (NULL != collection && pq.getOptions().oplogReplay) {
// Takes ownership of 'cq'.
status = getOplogStartHack(txn, collection, cq, &rawExec);
}
else {
size_t options = QueryPlannerParams::DEFAULT;
if (shardingState.needCollectionMetadata(pq.ns())) {
options |= QueryPlannerParams::INCLUDE_SHARD_FILTER;
}
// Takes ownership of 'cq'.
status = getExecutor(txn, collection, cq, PlanExecutor::YIELD_AUTO, &rawExec, options);
}
if (!status.isOK()) {
// NOTE: Do not access cq as getExecutor has deleted it.
uasserted(17007, "Unable to execute query: " + status.reason());
}
verify(NULL != rawExec);
auto_ptr<PlanExecutor> exec(rawExec);
// 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());
curop.debug().iscommand = true;
// 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 "";
}
// We freak out later if this changes before we're done with the query.
const ChunkVersion shardingVersionAtStart = shardingState.getVersion(cq->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.getOptions().slaveOk || pq.hasReadPref();
status = repl::getGlobalReplicationCoordinator()->checkCanServeReadsFor(
txn,
NamespaceString(cq->ns()),
slaveOK);
uassertStatusOK(status);
// If this exists, the collection is sharded.
// If it doesn't exist, we can assume we're not sharded.
// If we're sharded, we might encounter data that is not consistent with our sharding state.
// We must ignore this data.
CollectionMetadataPtr collMetadata;
if (!shardingState.needCollectionMetadata(pq.ns())) {
collMetadata = CollectionMetadataPtr();
}
else {
collMetadata = shardingState.getCollectionMetadata(pq.ns());
}
// 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.
OpTime slaveReadTill;
// Do we save the PlanExecutor in a ClientCursor for getMore calls later?
bool saveClientCursor = false;
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.getOptions().oplogReplay) {
BSONElement e = obj["ts"];
if (Date == e.type() || Timestamp == e.type()) {
slaveReadTill = e._opTime();
}
}
// TODO: only one type of 2d search doesn't support this. We need a way to pull it out
// of CanonicalQuery. :(
const bool supportsGetMore = true;
if (!supportsGetMore && (enough(pq, numResults)
|| bb.len() >= MaxBytesToReturnToClientAtOnce)) {
break;
}
else if (enoughForFirstBatch(pq, numResults, bb.len())) {
QLOG() << "Enough for first batch, wantMore=" << pq.wantMore()
<< " numToReturn=" << pq.getNumToReturn()
<< " numResults=" << numResults
<< endl;
// If only one result requested assume it's a findOne() and don't save the cursor.
if (pq.wantMore() && 1 != pq.getNumToReturn()) {
QLOG() << " executor EOF=" << exec->isEOF() << endl;
saveClientCursor = !exec->isEOF();
}
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::EXEC_ERROR == state) {
scoped_ptr<PlanStageStats> stats(exec->getStats());
error() << "Plan executor error, stats: "
<< Explain::statsToBSON(*stats);
uasserted(17144, "Executor error: " + WorkingSetCommon::toStatusString(obj));
}
// Why save a dead executor?
if (PlanExecutor::DEAD == state) {
saveClientCursor = false;
}
else if (pq.getOptions().tailable) {
// If we're tailing a capped collection, we don't bother saving the cursor if the
// collection is empty. Otherwise, the semantics of the tailable cursor is that the
// client will keep trying to read from it. So we'll keep it around.
if (collection && collection->numRecords(txn) != 0 && pq.getNumToReturn() != 1) {
saveClientCursor = true;
}
}
// TODO(greg): This will go away soon.
if (!shardingState.getVersion(pq.ns()).isWriteCompatibleWith(shardingVersionAtStart)) {
// if the version changed during the query we might be missing some data and its safe to
// send this as mongos can resend at this point
throw SendStaleConfigException(pq.ns(), "version changed during initial query",
shardingVersionAtStart,
shardingState.getVersion(pq.ns()));
}
const logger::LogComponent queryLogComponent = logger::LogComponent::kQuery;
const logger::LogSeverity logLevelOne = logger::LogSeverity::Debug(1);
PlanSummaryStats summaryStats;
Explain::getSummaryStats(exec.get(), &summaryStats);
curop.debug().ntoskip = pq.getSkip();
curop.debug().nreturned = numResults;
curop.debug().scanAndOrder = summaryStats.hasSortStage;
curop.debug().nscanned = summaryStats.totalKeysExamined;
curop.debug().nscannedObjects = summaryStats.totalDocsExamined;
curop.debug().idhack = summaryStats.isIdhack;
// Set debug information for consumption by the profiler.
if (dbProfilingLevel > 0 ||
curop.elapsedMillis() > serverGlobalParams.slowMS ||
logger::globalLogDomain()->shouldLog(queryLogComponent, logLevelOne)) {
// Get BSON stats.
scoped_ptr<PlanStageStats> execStats(exec->getStats());
BSONObjBuilder statsBob;
Explain::statsToBSON(*execStats, &statsBob);
curop.debug().execStats.set(statsBob.obj());
// Replace exec stats with plan summary if stats cannot fit into CachedBSONObj.
if (curop.debug().execStats.tooBig() && !curop.debug().planSummary.empty()) {
BSONObjBuilder bob;
bob.append("summary", curop.debug().planSummary.toString());
curop.debug().execStats.set(bob.done());
}
}
long long ccId = 0;
if (saveClientCursor) {
// We won't use the executor until it's getMore'd.
exec->saveState();
// 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, exec.get(),
cq->getParsed().getOptions().toInt(),
cq->getParsed().getFilter());
ccId = cc->cursorid();
if (fromDBDirectClient) {
cc->setUnownedRecoveryUnit(txn->recoveryUnit());
}
else if (state == PlanExecutor::IS_EOF && pq.getOptions().tailable) {
// Don't stash the RU for tailable cursors at EOF, let them get a new RU on their
// next getMore.
}
else {
// We stash away the RecoveryUnit in the ClientCursor. It's used for subsequent
// getMore requests. The calling OpCtx gets a fresh RecoveryUnit.
cc->setOwnedRecoveryUnit(txn->releaseRecoveryUnit());
StorageEngine* storageEngine = getGlobalEnvironment()->getGlobalStorageEngine();
txn->setRecoveryUnit(storageEngine->newRecoveryUnit(txn));
}
QLOG() << "caching executor with cursorid " << ccId
<< " after returning " << numResults << " results" << endl;
// ClientCursor takes ownership of executor. Release to make sure it's not deleted.
exec.release();
// TODO document
if (pq.getOptions().oplogReplay && !slaveReadTill.isNull()) {
cc->slaveReadTill(slaveReadTill);
}
// TODO document
if (pq.getOptions().exhaust) {
curop.debug().exhaust = true;
}
// Set attributes for getMore.
cc->setCollMetadata(collMetadata);
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());
}
else {
QLOG() << "Not caching executor but returning " << numResults << " results.\n";
}
// 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);
curop.debug().cursorid = (0 == ccId ? -1 : ccId);
qr.setResultFlagsToOk();
qr.msgdata().setOperation(opReply);
qr.setStartingFrom(0);
qr.setNReturned(numResults);
// curop.debug().exhaust is set above.
return curop.debug().exhaust ? pq.ns() : "";
}
std::string newRunQuery(Message& m, QueryMessage& q, CurOp& curop, Message &result) {
// Validate the namespace.
const char *ns = q.ns;
uassert(16332, "can't have an empty ns", ns[0]);
const NamespaceString nsString(ns);
uassert(16256, str::stream() << "Invalid ns [" << ns << "]", nsString.isValid());
// Set curop information.
curop.debug().ns = ns;
curop.debug().ntoreturn = q.ntoreturn;
curop.debug().query = q.query;
curop.setQuery(q.query);
// If the query is really a command, run it.
if (nsString.isCommand()) {
int nToReturn = q.ntoreturn;
uassert(16979, str::stream() << "bad numberToReturn (" << nToReturn
<< ") for $cmd type ns - can only be 1 or -1",
nToReturn == 1 || nToReturn == -1);
curop.markCommand();
BufBuilder bb;
bb.skip(sizeof(QueryResult));
BSONObjBuilder cmdResBuf;
if (!runCommands(ns, q.query, curop, bb, cmdResBuf, false, q.queryOptions)) {
uasserted(13530, "bad or malformed command request?");
}
curop.debug().iscommand = true;
// TODO: Does this get overwritten/do we really need to set this twice?
curop.debug().query = q.query;
QueryResult* qr = reinterpret_cast<QueryResult*>(bb.buf());
bb.decouple();
qr->setResultFlagsToOk();
qr->len = bb.len();
curop.debug().responseLength = bb.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
result.setData(qr, true);
return "";
}
// This is a read lock. We require this because if we're parsing a $where, the
// where-specific parsing code assumes we have a lock and creates execution machinery that
// requires it.
Client::ReadContext ctx(q.ns);
Collection* collection = ctx.ctx().db()->getCollection( ns );
// Parse the qm into a CanonicalQuery.
CanonicalQuery* cq;
Status canonStatus = CanonicalQuery::canonicalize(q, &cq);
if (!canonStatus.isOK()) {
uasserted(17287, str::stream() << "Can't canonicalize query: " << canonStatus.toString());
}
verify(cq);
QLOG() << "Running query:\n" << cq->toString();
LOG(2) << "Running query: " << cq->toStringShort();
// Parse, canonicalize, plan, transcribe, and get a runner.
Runner* rawRunner = NULL;
// We use this a lot below.
const LiteParsedQuery& pq = cq->getParsed();
// We'll now try to get the query runner that will execute this query for us. There
// are a few cases in which we know upfront which runner we should get and, therefore,
// we shortcut the selection process here.
//
// (a) If the query is over a collection that doesn't exist, we get a special runner
// that's is so (a runner) which doesn't return results, the EOFRunner.
//
// (b) if the query is a replication's initial sync one, we get a SingleSolutinRunner
// that uses a specifically designed stage that skips extents faster (see details in
// exec/oplogstart.h)
//
// Otherwise we go through the selection of which runner is most suited to the
// query + run-time context at hand.
Status status = Status::OK();
if (collection == NULL) {
rawRunner = new EOFRunner(cq, cq->ns());
}
else if (pq.hasOption(QueryOption_OplogReplay)) {
status = getOplogStartHack(collection, cq, &rawRunner);
}
else {
// Takes ownership of cq.
size_t options = QueryPlannerParams::DEFAULT;
if (shardingState.needCollectionMetadata(pq.ns())) {
options |= QueryPlannerParams::INCLUDE_SHARD_FILTER;
}
status = getRunner(cq, &rawRunner, options);
}
if (!status.isOK()) {
// NOTE: Do not access cq as getRunner has deleted it.
uasserted(17007, "Unable to execute query: " + status.reason());
}
verify(NULL != rawRunner);
auto_ptr<Runner> runner(rawRunner);
// We freak out later if this changes before we're done with the query.
const ChunkVersion shardingVersionAtStart = shardingState.getVersion(cq->ns());
// Handle query option $maxTimeMS (not used with commands).
curop.setMaxTimeMicros(static_cast<unsigned long long>(pq.getMaxTimeMS()) * 1000);
killCurrentOp.checkForInterrupt(); // May trigger maxTimeAlwaysTimeOut fail point.
// uassert if we are not on a primary, and not a secondary with SlaveOk query parameter set.
replVerifyReadsOk(&pq);
// If this exists, the collection is sharded.
// If it doesn't exist, we can assume we're not sharded.
// If we're sharded, we might encounter data that is not consistent with our sharding state.
// We must ignore this data.
CollectionMetadataPtr collMetadata;
if (!shardingState.needCollectionMetadata(pq.ns())) {
collMetadata = CollectionMetadataPtr();
}
else {
collMetadata = shardingState.getCollectionMetadata(pq.ns());
}
// 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));
// How many results have we obtained from the runner?
int numResults = 0;
// If we're replaying the oplog, we save the last time that we read.
OpTime slaveReadTill;
// Do we save the Runner in a ClientCursor for getMore calls later?
bool saveClientCursor = false;
// We turn on auto-yielding for the runner here. The runner registers itself with the
// active runners list in ClientCursor.
auto_ptr<ScopedRunnerRegistration> safety(new ScopedRunnerRegistration(runner.get()));
runner->setYieldPolicy(Runner::YIELD_AUTO);
BSONObj obj;
Runner::RunnerState state;
// uint64_t numMisplacedDocs = 0;
// set this outside loop. we will need to use this both within loop and when deciding
// to fill in explain information
const bool isExplain = pq.isExplain();
// Have we retrieved info about which plan the runner will
// use to execute the query yet?
bool gotPlanInfo = false;
PlanInfo* rawInfo;
boost::scoped_ptr<PlanInfo> planInfo;
while (Runner::RUNNER_ADVANCED == (state = runner->getNext(&obj, NULL))) {
// Add result to output buffer. This is unnecessary if explain info is requested
if (!isExplain) {
bb.appendBuf((void*)obj.objdata(), obj.objsize());
}
// Count the result.
++numResults;
// In the case of the multi plan runner, we may not be able to
// successfully retrieve plan info until after the query starts
// to run. This is because the multi plan runner doesn't know what
// plan it will end up using until it runs candidates and selects
// the best.
//
// TODO: Do we ever want to output what the MPR is comparing?
if (!gotPlanInfo) {
Status infoStatus = runner->getInfo(NULL, &rawInfo);
if (infoStatus.isOK()) {
gotPlanInfo = true;
planInfo.reset(rawInfo);
// planSummary is really a ThreadSafeString which copies the data from
// the provided pointer.
curop.debug().planSummary = planInfo->planSummary.c_str();
}
}
// Possibly note slave's position in the oplog.
if (pq.hasOption(QueryOption_OplogReplay)) {
BSONElement e = obj["ts"];
if (Date == e.type() || Timestamp == e.type()) {
slaveReadTill = e._opTime();
}
}
// TODO: only one type of 2d search doesn't support this. We need a way to pull it out
// of CanonicalQuery. :(
const bool supportsGetMore = true;
if (isExplain) {
if (enoughForExplain(pq, numResults)) {
break;
}
}
else if (!supportsGetMore && (enough(pq, numResults)
|| bb.len() >= MaxBytesToReturnToClientAtOnce)) {
break;
}
else if (enoughForFirstBatch(pq, numResults, bb.len())) {
QLOG() << "Enough for first batch, wantMore=" << pq.wantMore()
<< " numToReturn=" << pq.getNumToReturn()
<< " numResults=" << numResults
<< endl;
// If only one result requested assume it's a findOne() and don't save the cursor.
if (pq.wantMore() && 1 != pq.getNumToReturn()) {
QLOG() << " runner EOF=" << runner->isEOF() << endl;
saveClientCursor = !runner->isEOF();
}
break;
}
}
// Try to get information about the plan which the runner
// will use to execute the query, it we don't have it already.
if (!gotPlanInfo) {
Status infoStatus = runner->getInfo(NULL, &rawInfo);
if (infoStatus.isOK()) {
gotPlanInfo = true;
planInfo.reset(rawInfo);
// planSummary is really a ThreadSafeString which copies the data from
// the provided pointer.
curop.debug().planSummary = planInfo->planSummary.c_str();
}
}
// If we cache the runner later, we want to deregister it as it receives notifications
// anyway by virtue of being cached.
//
// If we don't cache the runner later, we are deleting it, so it must be deregistered.
//
// So, no matter what, deregister the runner.
safety.reset();
// Caller expects exceptions thrown in certain cases.
if (Runner::RUNNER_ERROR == state) {
TypeExplain* bareExplain;
Status res = runner->getInfo(&bareExplain, NULL);
if (res.isOK()) {
boost::scoped_ptr<TypeExplain> errorExplain(bareExplain);
error() << "Runner error, stats:\n"
<< errorExplain->stats.jsonString(Strict, true);
}
uasserted(17144, "Runner error: " + WorkingSetCommon::toStatusString(obj));
}
// Why save a dead runner?
if (Runner::RUNNER_DEAD == state) {
saveClientCursor = false;
}
else if (pq.hasOption(QueryOption_CursorTailable)) {
// If we're tailing a capped collection, we don't bother saving the cursor if the
// collection is empty. Otherwise, the semantics of the tailable cursor is that the
// client will keep trying to read from it. So we'll keep it around.
Collection* collection = ctx.ctx().db()->getCollection(cq->ns());
if (collection && collection->numRecords() != 0 && pq.getNumToReturn() != 1) {
saveClientCursor = true;
}
}
// TODO(greg): This will go away soon.
if (!shardingState.getVersion(pq.ns()).isWriteCompatibleWith(shardingVersionAtStart)) {
// if the version changed during the query we might be missing some data and its safe to
// send this as mongos can resend at this point
throw SendStaleConfigException(pq.ns(), "version changed during initial query",
shardingVersionAtStart,
shardingState.getVersion(pq.ns()));
}
// Used to fill in explain and to determine if the query is slow enough to be logged.
int elapsedMillis = curop.elapsedMillis();
// Get explain information if:
// 1) it is needed by an explain query;
// 2) profiling is enabled; or
// 3) profiling is disabled but we still need explain details to log a "slow" query.
// Producing explain information is expensive and should be done only if we are certain
// the information will be used.
boost::scoped_ptr<TypeExplain> explain(NULL);
if (isExplain ||
ctx.ctx().db()->getProfilingLevel() > 0 ||
elapsedMillis > serverGlobalParams.slowMS) {
// Ask the runner to produce explain information.
TypeExplain* bareExplain;
Status res = runner->getInfo(&bareExplain, NULL);
if (res.isOK()) {
explain.reset(bareExplain);
}
else if (isExplain) {
error() << "could not produce explain of query '" << pq.getFilter()
<< "', error: " << res.reason();
// If numResults and the data in bb don't correspond, we'll crash later when rooting
// through the reply msg.
BSONObj emptyObj;
bb.appendBuf((void*)emptyObj.objdata(), emptyObj.objsize());
// The explain output is actually a result.
numResults = 1;
// TODO: we can fill out millis etc. here just fine even if the plan screwed up.
}
}
// Fill in the missing run-time fields in explain, starting with propeties of
// the process running the query.
if (isExplain && NULL != explain.get()) {
std::string server = mongoutils::str::stream()
<< getHostNameCached() << ":" << serverGlobalParams.port;
explain->setServer(server);
// We might have skipped some results due to chunk migration etc. so our count is
// correct.
explain->setN(numResults);
// Clock the whole operation.
explain->setMillis(elapsedMillis);
BSONObj explainObj = explain->toBSON();
bb.appendBuf((void*)explainObj.objdata(), explainObj.objsize());
// The explain output is actually a result.
numResults = 1;
}
long long ccId = 0;
if (saveClientCursor) {
// We won't use the runner until it's getMore'd.
runner->saveState();
// 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, runner.get(),
cq->getParsed().getOptions(),
cq->getParsed().getFilter());
ccId = cc->cursorid();
QLOG() << "caching runner with cursorid " << ccId
<< " after returning " << numResults << " results" << endl;
// ClientCursor takes ownership of runner. Release to make sure it's not deleted.
runner.release();
// TODO document
if (pq.hasOption(QueryOption_OplogReplay) && !slaveReadTill.isNull()) {
cc->slaveReadTill(slaveReadTill);
}
// TODO document
if (pq.hasOption(QueryOption_Exhaust)) {
curop.debug().exhaust = true;
}
// Set attributes for getMore.
cc->setCollMetadata(collMetadata);
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());
}
else {
QLOG() << "Not caching runner but returning " << numResults << " results.\n";
}
// 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* qr = static_cast<QueryResult*>(result.header());
qr->cursorId = ccId;
curop.debug().cursorid = (0 == ccId ? -1 : ccId);
qr->setResultFlagsToOk();
qr->setOperation(opReply);
qr->startingFrom = 0;
qr->nReturned = numResults;
// Set debug information for consumption by the profiler.
curop.debug().ntoskip = pq.getSkip();
curop.debug().nreturned = numResults;
if (NULL != explain.get()) {
if (explain->isScanAndOrderSet()) {
curop.debug().scanAndOrder = explain->getScanAndOrder();
}
else {
curop.debug().scanAndOrder = false;
}
if (explain->isNScannedSet()) {
curop.debug().nscanned = explain->getNScanned();
}
if (explain->isNScannedObjectsSet()) {
curop.debug().nscannedObjects = explain->getNScannedObjects();
}
if (explain->isIDHackSet()) {
curop.debug().idhack = explain->getIDHack();
}
if (!explain->stats.isEmpty()) {
// execStats is a CachedBSONObj because it lives in the race-prone
// curop.
curop.debug().execStats.set(explain->stats);
// Replace exec stats with plan summary if stats cannot fit into CachedBSONObj.
if (curop.debug().execStats.tooBig() && !curop.debug().planSummary.empty()) {
BSONObjBuilder bob;
bob.append("summary", curop.debug().planSummary.toString());
curop.debug().execStats.set(bob.done());
}
}
}
// curop.debug().exhaust is set above.
return curop.debug().exhaust ? pq.ns() : "";
}
/**
* Run a query -- includes checking for and running a Command.
* @return points to ns if exhaust mode. 0=normal mode
* @locks the db mutex for reading (and potentially for writing temporarily to create a new db).
* @asserts on scan and order memory exhaustion and other cases.
*/
string runQuery(Message& m, QueryMessage& q, CurOp& curop, Message &result) {
shared_ptr<ParsedQuery> pq_shared( new ParsedQuery(q) );
ParsedQuery& pq( *pq_shared );
BSONObj jsobj = q.query;
int queryOptions = q.queryOptions;
const char *ns = q.ns;
uassert( 16332 , "can't have an empty ns" , ns[0] );
if( logLevel >= 2 )
log() << "runQuery called " << ns << " " << jsobj << endl;
curop.debug().ns = ns;
curop.debug().ntoreturn = pq.getNumToReturn();
curop.debug().query = jsobj;
curop.setQuery(jsobj);
uassert( 16256, str::stream() << "Invalid ns [" << ns << "]", NamespaceString::isValid(ns) );
// Run a command.
if ( pq.couldBeCommand() ) {
curop.markCommand();
BufBuilder bb;
bb.skip(sizeof(QueryResult));
BSONObjBuilder cmdResBuf;
if ( runCommands(ns, jsobj, curop, bb, cmdResBuf, false, queryOptions) ) {
curop.debug().iscommand = true;
curop.debug().query = jsobj;
auto_ptr< QueryResult > qr;
qr.reset( (QueryResult *) bb.buf() );
bb.decouple();
qr->setResultFlagsToOk();
qr->len = bb.len();
curop.debug().responseLength = bb.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
result.setData( qr.release(), true );
}
else {
uasserted(13530, "bad or malformed command request?");
}
return "";
}
const bool explain = pq.isExplain();
const bool tailable = pq.hasOption(QueryOption_CursorTailable);
BSONObj order = pq.getOrder();
BSONObj query = pq.getFilter();
/* The ElemIter will not be happy if this isn't really an object. So throw exception
here when that is true.
(Which may indicate bad data from client.)
*/
if ( query.objsize() == 0 ) {
out() << "Bad query object?\n jsobj:";
out() << jsobj.toString() << "\n query:";
out() << query.toString() << endl;
uassert( 10110 , "bad query object", false);
}
// Tailable cursors need to read newly written entries from the tail
// of the collection. They manually arbitrate with the collection over
// what data is readable and when, so we choose read uncommited isolation.
OpSettings settings;
settings.setQueryCursorMode(DEFAULT_LOCK_CURSOR);
settings.setBulkFetch(true);
settings.setCappedAppendPK(pq.hasOption(QueryOption_AddHiddenPK));
cc().setOpSettings(settings);
// If our caller has a transaction, it's multi-statement.
const bool inMultiStatementTxn = cc().hasTxn();
if (tailable) {
// Because it's easier to disable this. It shouldn't be happening in a normal system.
uassert(16812, "May not perform a tailable query in a multi-statement transaction.",
!inMultiStatementTxn);
}
// Begin a read-only, snapshot transaction under normal circumstances.
// If the cursor is tailable, we need to be able to read uncommitted data.
const int txnFlags = (tailable ? DB_READ_UNCOMMITTED : DB_TXN_SNAPSHOT) | DB_TXN_READ_ONLY;
LOCK_REASON(lockReason, "query");
Client::ReadContext ctx(ns, lockReason);
scoped_ptr<Client::Transaction> transaction(!inMultiStatementTxn ?
new Client::Transaction(txnFlags) : NULL);
bool hasRetried = false;
while ( 1 ) {
try {
replVerifyReadsOk(&pq);
// Fast-path for primary key queries.
if (!explain && !tailable) {
replVerifyReadsOk(&pq);
if (_tryQueryByPKHack(ns, query, pq, curop, result)) {
if (transaction) {
transaction->commit();
}
return "";
}
}
// sanity check the query and projection
if (pq.getFields() != NULL) {
pq.getFields()->validateQuery( query );
}
if (tailable) {
Collection *cl = getCollection( ns );
if (cl != NULL && !(cl->isCapped() || str::equals(ns, rsoplog))) {
uasserted( 13051, "tailable cursor requested on non-capped, non-oplog collection" );
}
const BSONObj nat1 = BSON( "$natural" << 1 );
if ( order.isEmpty() ) {
order = nat1;
} else {
uassert( 13052, "only {$natural:1} order allowed for tailable cursor", order == nat1 );
}
}
// Run a regular query.
// these now may stored in a ClientCursor or somewhere else,
// so make sure we use a real copy
jsobj = jsobj.getOwned();
query = query.getOwned();
order = order.getOwned();
const ConfigVersion shardingVersionAtStart = shardingState.getVersion( ns );
const bool getCachedExplainPlan = ! hasRetried && explain && ! pq.hasIndexSpecifier();
const bool savedCursor = queryWithQueryOptimizer( queryOptions, ns, jsobj, curop, query,
order, pq_shared, shardingVersionAtStart,
getCachedExplainPlan, inMultiStatementTxn,
result );
// Did not save the cursor, so we can commit the transaction now if it exists.
if (transaction && !savedCursor) {
transaction->commit();
}
return curop.debug().exhaust ? ns : "";
}
catch ( const QueryRetryException & ) {
// In some cases the query may be retried if there is an in memory sort size assertion.
verify( ! hasRetried );
hasRetried = true;
}
}
}
std::string runQuery(OperationContext* txn,
QueryMessage& q,
const NamespaceString& nss,
CurOp& curop,
Message &result) {
// Validate the namespace.
uassert(16256, str::stream() << "Invalid ns [" << nss.ns() << "]", nss.isValid());
invariant(!nss.isCommand());
// Set curop information.
beginQueryOp(nss, q.query, q.ntoreturn, q.ntoskip, &curop);
// Parse the qm into a CanonicalQuery.
std::auto_ptr<CanonicalQuery> cq;
{
CanonicalQuery* cqRaw;
Status canonStatus = CanonicalQuery::canonicalize(q,
&cqRaw,
WhereCallbackReal(txn, nss.db()));
if (!canonStatus.isOK()) {
uasserted(17287, str::stream() << "Can't canonicalize query: "
<< canonStatus.toString());
}
cq.reset(cqRaw);
}
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;
{
PlanExecutor* rawExec;
Status execStatus = getExecutorFind(txn,
collection,
nss,
cq.release(),
PlanExecutor::YIELD_AUTO,
&rawExec);
uassertStatusOK(execStatus);
exec.reset(rawExec);
}
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 "";
}
// 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 (enoughForFirstBatch(pq, numResults, bb.len())) {
LOG(5) << "Enough for first batch, wantMore=" << pq.wantMore()
<< " numToReturn=" << pq.getNumToReturn()
<< " 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) {
scoped_ptr<PlanStageStats> stats(exec->getStats());
error() << "Plan executor error, 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 mongos 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();
// 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(),
pq.getOptions(),
pq.getFilter());
ccId = cc->cursorid();
if (txn->getClient()->isInDirectClient()) {
cc->setUnownedRecoveryUnit(txn->recoveryUnit());
}
else if (state == PlanExecutor::IS_EOF && pq.isTailable()) {
// Don't stash the RU for tailable cursors at EOF, let them get a new RU on their
// next getMore.
}
else {
// We stash away the RecoveryUnit in the ClientCursor. It's used for subsequent
// getMore requests. The calling OpCtx gets a fresh RecoveryUnit.
txn->recoveryUnit()->abandonSnapshot();
cc->setOwnedRecoveryUnit(txn->releaseRecoveryUnit());
StorageEngine* storageEngine = getGlobalServiceContext()->getGlobalStorageEngine();
invariant(txn->setRecoveryUnit(storageEngine->newRecoveryUnit(),
OperationContext::kNotInUnitOfWork)
== OperationContext::kNotInUnitOfWork);
}
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(cc->getExecutor(), dbProfilingLevel, numResults, ccId, &curop);
}
else {
LOG(5) << "Not caching executor but returning " << numResults << " results.\n";
endQueryOp(exec.get(), dbProfilingLevel, numResults, ccId, &curop);
}
// 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() : "";
}
static bool receivedQuery(Client& c, DbResponse& dbresponse, Message& m ) {
bool ok = true;
MSGID responseTo = m.header()->id;
DbMessage d(m);
QueryMessage q(d);
auto_ptr< Message > resp( new Message() );
CurOp& op = *(c.curop());
shared_ptr<AssertionException> ex;
try {
if (!NamespaceString(d.getns()).isCommand()) {
// Auth checking for Commands happens later.
Status status = cc().getAuthorizationManager()->checkAuthForQuery(d.getns());
uassert(16550, status.reason(), status.isOK());
}
dbresponse.exhaustNS = runQuery(m, q, op, *resp);
verify( !resp->empty() );
}
catch ( SendStaleConfigException& e ){
ex.reset( new SendStaleConfigException( e.getns(), e.getInfo().msg, e.getVersionReceived(), e.getVersionWanted() ) );
ok = false;
}
catch ( AssertionException& e ) {
ex.reset( new AssertionException( e.getInfo().msg, e.getCode() ) );
ok = false;
}
if( ex ){
op.debug().exceptionInfo = ex->getInfo();
LOGWITHRATELIMIT {
log() << "assertion " << ex->toString() << " ns:" << q.ns << " query:" <<
(q.query.valid() ? q.query.toString() : "query object is corrupt") << endl;
if( q.ntoskip || q.ntoreturn )
log() << " ntoskip:" << q.ntoskip << " ntoreturn:" << q.ntoreturn << endl;
}
SendStaleConfigException* scex = NULL;
if ( ex->getCode() == SendStaleConfigCode ) scex = static_cast<SendStaleConfigException*>( ex.get() );
BSONObjBuilder err;
ex->getInfo().append( err );
if( scex ){
err.append( "ns", scex->getns() );
scex->getVersionReceived().addToBSON( err, "vReceived" );
scex->getVersionWanted().addToBSON( err, "vWanted" );
}
BSONObj errObj = err.done();
if( scex ){
log() << "stale version detected during query over "
<< q.ns << " : " << errObj << endl;
}
else{
log() << "problem detected during query over "
<< q.ns << " : " << errObj << endl;
}
BufBuilder b;
b.skip(sizeof(QueryResult));
b.appendBuf((void*) errObj.objdata(), errObj.objsize());
// todo: call replyToQuery() from here instead of this!!! see dbmessage.h
QueryResult * msgdata = (QueryResult *) b.buf();
b.decouple();
QueryResult *qr = msgdata;
qr->_resultFlags() = ResultFlag_ErrSet;
if( scex ) qr->_resultFlags() |= ResultFlag_ShardConfigStale;
qr->len = b.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
resp.reset( new Message() );
resp->setData( msgdata, true );
}
op.debug().responseLength = resp->header()->dataLen();
dbresponse.response = resp.release();
dbresponse.responseTo = responseTo;
return ok;
}
/**
* Run a query -- includes checking for and running a Command.
* @return points to ns if exhaust mode. 0=normal mode
* @locks the db mutex for reading (and potentially for writing temporarily to create a new db).
* @yields the db mutex periodically after acquiring it.
* @asserts on scan and order memory exhaustion and other cases.
*/
const char *runQuery(Message& m, QueryMessage& q, CurOp& curop, Message &result) {
shared_ptr<ParsedQuery> pq_shared( new ParsedQuery(q) );
ParsedQuery& pq( *pq_shared );
BSONObj jsobj = q.query;
int queryOptions = q.queryOptions;
const char *ns = q.ns;
if( logLevel >= 2 )
log() << "runQuery called " << ns << " " << jsobj << endl;
curop.debug().ns = ns;
curop.debug().ntoreturn = pq.getNumToReturn();
curop.debug().query = jsobj;
curop.setQuery(jsobj);
// Run a command.
if ( pq.couldBeCommand() ) {
BufBuilder bb;
bb.skip(sizeof(QueryResult));
BSONObjBuilder cmdResBuf;
if ( runCommands(ns, jsobj, curop, bb, cmdResBuf, false, queryOptions) ) {
curop.debug().iscommand = true;
curop.debug().query = jsobj;
curop.markCommand();
auto_ptr< QueryResult > qr;
qr.reset( (QueryResult *) bb.buf() );
bb.decouple();
qr->setResultFlagsToOk();
qr->len = bb.len();
curop.debug().responseLength = bb.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = 1;
result.setData( qr.release(), true );
}
else {
uasserted(13530, "bad or malformed command request?");
}
return 0;
}
bool explain = pq.isExplain();
BSONObj order = pq.getOrder();
BSONObj query = pq.getFilter();
/* The ElemIter will not be happy if this isn't really an object. So throw exception
here when that is true.
(Which may indicate bad data from client.)
*/
if ( query.objsize() == 0 ) {
out() << "Bad query object?\n jsobj:";
out() << jsobj.toString() << "\n query:";
out() << query.toString() << endl;
uassert( 10110 , "bad query object", false);
}
Client::ReadContext ctx( ns , dbpath ); // read locks
const ConfigVersion shardingVersionAtStart = shardingState.getVersion( ns );
replVerifyReadsOk(&pq);
if ( pq.hasOption( QueryOption_CursorTailable ) ) {
NamespaceDetails *d = nsdetails( ns );
uassert( 13051, "tailable cursor requested on non capped collection", d && d->isCapped() );
const BSONObj nat1 = BSON( "$natural" << 1 );
if ( order.isEmpty() ) {
order = nat1;
}
else {
uassert( 13052, "only {$natural:1} order allowed for tailable cursor", order == nat1 );
}
}
// Run a simple id query.
if ( ! (explain || pq.showDiskLoc()) && isSimpleIdQuery( query ) && !pq.hasOption( QueryOption_CursorTailable ) ) {
int n = 0;
bool nsFound = false;
bool indexFound = false;
BSONObj resObject;
Client& c = cc();
bool found = Helpers::findById( c, ns , query , resObject , &nsFound , &indexFound );
if ( nsFound == false || indexFound == true ) {
if ( shardingState.needShardChunkManager( ns ) ) {
ShardChunkManagerPtr m = shardingState.getShardChunkManager( ns );
if ( m && ! m->belongsToMe( resObject ) ) {
// I have something this _id
// but it doesn't belong to me
// so return nothing
resObject = BSONObj();
found = false;
}
}
BufBuilder bb(sizeof(QueryResult)+resObject.objsize()+32);
bb.skip(sizeof(QueryResult));
curop.debug().idhack = true;
if ( found ) {
n = 1;
fillQueryResultFromObj( bb , pq.getFields() , resObject );
}
auto_ptr< QueryResult > qr;
qr.reset( (QueryResult *) bb.buf() );
bb.decouple();
qr->setResultFlagsToOk();
qr->len = bb.len();
curop.debug().responseLength = bb.len();
qr->setOperation(opReply);
qr->cursorId = 0;
qr->startingFrom = 0;
qr->nReturned = n;
result.setData( qr.release(), true );
return NULL;
}
}
// Run a regular query.
BSONObj oldPlan;
if ( explain && ! pq.hasIndexSpecifier() ) {
MultiPlanScanner mps( ns, query, order );
if ( mps.usingCachedPlan() ) {
oldPlan =
mps.oldExplain().firstElement().embeddedObject()
.firstElement().embeddedObject().getOwned();
}
}
// In some cases the query may be retried if there is an in memory sort size assertion.
for( int retry = 0; retry < 2; ++retry ) {
try {
return queryWithQueryOptimizer( m, queryOptions, ns, jsobj, curop, query, order,
pq_shared, oldPlan, shardingVersionAtStart, result );
} catch ( const QueryRetryException & ) {
verify( retry == 0 );
}
}
verify( false );
return 0;
}
std::string runQuery(OperationContext* txn,
QueryMessage& q,
const NamespaceString& nss,
Message& result) {
CurOp& curop = *CurOp::get(txn);
uassert(ErrorCodes::InvalidNamespace,
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, ExtensionsCallbackReal(txn, &nss));
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 "";
}
// 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(FindCommon::kInitReplyBufferSize);
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;
// Get summary info about which plan the executor is using.
{
stdx::lock_guard<Client> lk(*txn->getClient());
curop.setPlanSummary_inlock(Explain::getPlanSummary(exec.get()));
}
while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) {
// If we can't fit this result inside the current batch, then we stash it for later.
if (!FindCommon::haveSpaceForNext(obj, numResults, bb.len())) {
exec->enqueue(obj);
break;
}
// 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)) {
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) {
error() << "Plan executor error during find: " << PlanExecutor::statestr(state)
<< ", stats: " << Explain::getWinningPlanStats(exec.get());
uasserted(17144, "Executor error: " + WorkingSetCommon::toStatusString(obj));
}
// Before saving the cursor, ensure that whatever plan we established happened with the expected
// collection version
auto css = CollectionShardingState::get(txn, nss);
css->checkShardVersionOrThrow(txn);
// 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() : "";
}
std::string runQuery(OperationContext* opCtx,
QueryMessage& q,
const NamespaceString& nss,
Message& result) {
CurOp& curOp = *CurOp::get(opCtx);
curOp.ensureStarted();
uassert(ErrorCodes::InvalidNamespace,
str::stream() << "Invalid ns [" << nss.ns() << "]",
nss.isValid());
invariant(!nss.isCommand());
// Set CurOp information.
const auto upconvertedQuery = upconvertQueryEntry(q.query, nss, q.ntoreturn, q.ntoskip);
beginQueryOp(opCtx, nss, upconvertedQuery, q.ntoreturn, q.ntoskip);
// Parse the qm into a CanonicalQuery.
const boost::intrusive_ptr<ExpressionContext> expCtx;
auto cq = uassertStatusOKWithContext(
CanonicalQuery::canonicalize(opCtx,
q,
expCtx,
ExtensionsCallbackReal(opCtx, &nss),
MatchExpressionParser::kAllowAllSpecialFeatures),
"Can't canonicalize query");
invariant(cq.get());
LOG(5) << "Running query:\n" << redact(cq->toString());
LOG(2) << "Running query: " << redact(cq->toStringShort());
// Parse, canonicalize, plan, transcribe, and get a plan executor.
AutoGetCollectionForReadCommand ctx(opCtx, nss, AutoGetCollection::ViewMode::kViewsForbidden);
Collection* const collection = ctx.getCollection();
{
const QueryRequest& qr = cq->getQueryRequest();
// Allow the query to run on secondaries if the read preference permits it. If no read
// preference was specified, allow the query to run iff slaveOk has been set.
const bool slaveOK = qr.hasReadPref()
? uassertStatusOK(ReadPreferenceSetting::fromContainingBSON(q.query))
.canRunOnSecondary()
: qr.isSlaveOk();
uassertStatusOK(
repl::ReplicationCoordinator::get(opCtx)->checkCanServeReadsFor(opCtx, nss, slaveOK));
}
// We have a parsed query. Time to get the execution plan for it.
auto exec = uassertStatusOK(getExecutorLegacyFind(opCtx, collection, nss, std::move(cq)));
const QueryRequest& qr = exec->getCanonicalQuery()->getQueryRequest();
// If it's actually an explain, do the explain and return rather than falling through
// to the normal query execution loop.
if (qr.isExplain()) {
BufBuilder bb;
bb.skip(sizeof(QueryResult::Value));
BSONObjBuilder explainBob;
Explain::explainStages(
exec.get(), collection, ExplainOptions::Verbosity::kExecAllPlans, &explainBob);
// Add the resulting object to the return buffer.
BSONObj explainObj = explainBob.obj();
bb.appendBuf((void*)explainObj.objdata(), explainObj.objsize());
// Set query result fields.
QueryResult::View qr = bb.buf();
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(bb.release());
return "";
}
// Handle query option $maxTimeMS (not used with commands).
if (qr.getMaxTimeMS() > 0) {
uassert(40116,
"Illegal attempt to set operation deadline within DBDirectClient",
!opCtx->getClient()->isInDirectClient());
opCtx->setDeadlineAfterNowBy(Milliseconds{qr.getMaxTimeMS()});
}
opCtx->checkForInterrupt(); // May trigger maxTimeAlwaysTimeOut fail point.
// 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(FindCommon::kInitReplyBufferSize);
bb.skip(sizeof(QueryResult::Value));
// How many results have we obtained from the executor?
int numResults = 0;
BSONObj obj;
PlanExecutor::ExecState state;
// Get summary info about which plan the executor is using.
{
stdx::lock_guard<Client> lk(*opCtx->getClient());
curOp.setPlanSummary_inlock(Explain::getPlanSummary(exec.get()));
}
while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) {
// If we can't fit this result inside the current batch, then we stash it for later.
if (!FindCommon::haveSpaceForNext(obj, numResults, bb.len())) {
exec->enqueue(obj);
break;
}
// Add result to output buffer.
bb.appendBuf((void*)obj.objdata(), obj.objsize());
// Count the result.
++numResults;
if (FindCommon::enoughForFirstBatch(qr, numResults)) {
LOG(5) << "Enough for first batch, wantMore=" << qr.wantMore()
<< " ntoreturn=" << qr.getNToReturn().value_or(0)
<< " numResults=" << numResults;
break;
}
}
// Caller expects exceptions thrown in certain cases.
if (PlanExecutor::FAILURE == state || PlanExecutor::DEAD == state) {
error() << "Plan executor error during find: " << PlanExecutor::statestr(state)
<< ", stats: " << redact(Explain::getWinningPlanStats(exec.get()));
uassertStatusOKWithContext(WorkingSetCommon::getMemberObjectStatus(obj),
"Executor error during OP_QUERY find");
MONGO_UNREACHABLE;
}
// Before saving the cursor, ensure that whatever plan we established happened with the expected
// collection version
auto css = CollectionShardingState::get(opCtx, nss);
css->checkShardVersionOrThrow(opCtx);
// 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(opCtx, collection, state, exec.get())) {
// We won't use the executor until it's getMore'd.
exec->saveState();
exec->detachFromOperationContext();
// Allocate a new ClientCursor and register it with the cursor manager.
ClientCursorPin pinnedCursor = collection->getCursorManager()->registerCursor(
opCtx,
{std::move(exec),
nss,
AuthorizationSession::get(opCtx->getClient())->getAuthenticatedUserNames(),
opCtx->recoveryUnit()->getReadConcernLevel(),
upconvertedQuery});
ccId = pinnedCursor.getCursor()->cursorid();
LOG(5) << "caching executor with cursorid " << ccId << " after returning " << numResults
<< " results";
// TODO document
if (qr.isExhaust()) {
curOp.debug().exhaust = true;
}
pinnedCursor.getCursor()->setPos(numResults);
// We assume that cursors created through a DBDirectClient are always used from their
// original OperationContext, so we do not need to move time to and from the cursor.
if (!opCtx->getClient()->isInDirectClient()) {
// If the query had a time limit, remaining time is "rolled over" to the cursor (for
// use by future getmore ops).
pinnedCursor.getCursor()->setLeftoverMaxTimeMicros(opCtx->getRemainingMaxTimeMicros());
}
endQueryOp(opCtx, collection, *pinnedCursor.getCursor()->getExecutor(), numResults, ccId);
} else {
LOG(5) << "Not caching executor but returning " << numResults << " results.";
endQueryOp(opCtx, collection, *exec, numResults, ccId);
}
// Fill out the output buffer's header.
QueryResult::View queryResultView = bb.buf();
queryResultView.setCursorId(ccId);
queryResultView.setResultFlagsToOk();
queryResultView.msgdata().setLen(bb.len());
queryResultView.msgdata().setOperation(opReply);
queryResultView.setStartingFrom(0);
queryResultView.setNReturned(numResults);
// Add the results from the query into the output buffer.
result.setData(bb.release());
// curOp.debug().exhaust is set above.
return curOp.debug().exhaust ? nss.ns() : "";
}
