C++ BSONElement::_opTime方法代码示例

    StatusWith<ReplicationCoordinatorExternalState::OpTimeAndHash>
    ReplicationCoordinatorExternalStateImpl::loadLastOpTimeAndHash(
            OperationContext* txn) {

        try {
            Lock::DBRead lk(txn->lockState(), rsoplog);
            BSONObj oplogEntry;
            if (!Helpers::getLast(txn, rsoplog, oplogEntry)) {
                return StatusWith<OpTimeAndHash>(
                        ErrorCodes::NoMatchingDocument,
                        str::stream() << "Did not find any entries in " << rsoplog);
            }
            BSONElement tsElement = oplogEntry[tsFieldName];
            if (tsElement.eoo()) {
                return StatusWith<OpTimeAndHash>(
                        ErrorCodes::NoSuchKey,
                        str::stream() << "Most recent entry in " << rsoplog << " missing \"" <<
                        tsFieldName << "\" field");
            }
            if (tsElement.type() != Timestamp) {
                return StatusWith<OpTimeAndHash>(
                        ErrorCodes::TypeMismatch,
                        str::stream() << "Expected type of \"" << tsFieldName <<
                        "\" in most recent " << rsoplog <<
                        " entry to have type Timestamp, but found " << typeName(tsElement.type()));
            }
            return StatusWith<OpTimeAndHash>(
                    OpTimeAndHash(tsElement._opTime(), oplogEntry[hashFieldName].safeNumberLong()));
        }
        catch (const DBException& ex) {
            return StatusWith<OpTimeAndHash>(ex.toStatus());
        }
    }

 Status bsonExtractOpTimeField(const BSONObj& object,
                               const StringData& fieldName,
                               OpTime* out) {
     BSONElement element;
     Status status = bsonExtractTypedField(object, fieldName, Timestamp, &element);
     if (!status.isOK())
         return status;
     *out = element._opTime();
     return Status::OK();
 }

    void ClientCursor::storeOpForSlave( DiskLoc last ) {
        if ( ! ( _queryOptions & QueryOption_OplogReplay ))
            return;

        if ( last.isNull() )
            return;

        BSONElement e = last.obj()["ts"];
        if ( e.type() == Date || e.type() == Timestamp )
            _slaveReadTill = e._opTime();
    }

//.........这里部分代码省略.........
        // 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();

    Status getOplogStartHack(OperationContext* txn,
                             Collection* collection,
                             CanonicalQuery* cq,
                             PlanExecutor** execOut) {
        invariant(cq);
        auto_ptr<CanonicalQuery> autoCq(cq);

        if ( collection == NULL )
            return Status(ErrorCodes::InternalError,
                          "getOplogStartHack called with a NULL collection" );

        // A query can only do oplog start finding if it has a top-level $gt or $gte predicate over
        // the "ts" field (the operation's timestamp). Find that predicate and pass it to
        // the OplogStart stage.
        MatchExpression* tsExpr = NULL;
        if (MatchExpression::AND == cq->root()->matchType()) {
            // The query has an AND at the top-level. See if any of the children
            // of the AND are $gt or $gte predicates over 'ts'.
            for (size_t i = 0; i < cq->root()->numChildren(); ++i) {
                MatchExpression* me = cq->root()->getChild(i);
                if (isOplogTsPred(me)) {
                    tsExpr = me;
                    break;
                }
            }
        }
        else if (isOplogTsPred(cq->root())) {
            // The root of the tree is a $gt or $gte predicate over 'ts'.
            tsExpr = cq->root();
        }

        if (NULL == tsExpr) {
            return Status(ErrorCodes::OplogOperationUnsupported,
                          "OplogReplay query does not contain top-level "
                          "$gt or $gte over the 'ts' field.");
        }

        DiskLoc startLoc = DiskLoc().setInvalid();

        // See if the RecordStore supports the oplogStartHack
        const BSONElement tsElem = extractOplogTsOptime(tsExpr);
        if (tsElem.type() == Timestamp) {
            StatusWith<DiskLoc> goal = oploghack::keyForOptime(tsElem._opTime());
            if (goal.isOK()) {
                startLoc = collection->getRecordStore()->oplogStartHack(txn, goal.getValue());
            }
        }

        if (startLoc.isValid()) {
            LOG(3) << "Using direct oplog seek";
        }
        else {
            LOG(3) << "Using OplogStart stage";

            // Fallback to trying the OplogStart stage.
            WorkingSet* oplogws = new WorkingSet();
            OplogStart* stage = new OplogStart(txn, collection, tsExpr, oplogws);
            PlanExecutor* rawExec;

            // Takes ownership of oplogws and stage.
            Status execStatus = PlanExecutor::make(txn, oplogws, stage, collection,
                                                   PlanExecutor::YIELD_AUTO, &rawExec);
            invariant(execStatus.isOK());
            scoped_ptr<PlanExecutor> exec(rawExec);

            // The stage returns a DiskLoc of where to start.
            PlanExecutor::ExecState state = exec->getNext(NULL, &startLoc);

            // This is normal.  The start of the oplog is the beginning of the collection.
            if (PlanExecutor::IS_EOF == state) {
                return getExecutor(txn, collection, autoCq.release(), PlanExecutor::YIELD_AUTO,
                                   execOut);
            }

            // This is not normal.  An error was encountered.
            if (PlanExecutor::ADVANCED != state) {
                return Status(ErrorCodes::InternalError,
                              "quick oplog start location had error...?");
            }
        }

        // cout << "diskloc is " << startLoc.toString() << endl;

        // Build our collection scan...
        CollectionScanParams params;
        params.collection = collection;
        params.start = startLoc;
        params.direction = CollectionScanParams::FORWARD;
        params.tailable = cq->getParsed().getOptions().tailable;

        WorkingSet* ws = new WorkingSet();
        CollectionScan* cs = new CollectionScan(txn, params, ws, cq->root());
        // Takes ownership of 'ws', 'cs', and 'cq'.
        return PlanExecutor::make(txn, ws, cs, autoCq.release(), collection,
                                  PlanExecutor::YIELD_AUTO, execOut);
    }

//.........这里部分代码省略.........
            if (cc->isAggCursor) {
                // Agg cursors handle their own locking internally.
                ctx.reset(); // unlocks
            }

            CollectionMetadataPtr collMetadata = cc->getCollMetadata();

            // If we're replaying the oplog, we save the last time that we read.
            OpTime slaveReadTill;

            // What number result are we starting at?  Used to fill out the reply.
            startingResult = cc->pos();

            // What gives us results.
            PlanExecutor* exec = cc->getExecutor();
            const int queryOptions = cc->queryOptions();

            // Get results out of the executor.
            exec->restoreState(txn);

            BSONObj obj;
            PlanExecutor::ExecState state;
            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 (queryOptions & QueryOption_OplogReplay) {
                    BSONElement e = obj["ts"];
                    if (Date == e.type() || Timestamp == e.type()) {
                        slaveReadTill = e._opTime();
                    }
                }

                if ((ntoreturn && numResults >= ntoreturn)
                    || bb.len() > MaxBytesToReturnToClientAtOnce) {
                    break;
                }
            }

            // We save the client cursor when there might be more results, and hence we may receive
            // another getmore. If we receive a EOF or an error, or 'exec' is dead, then we know
            // that we will not be producing more results. We indicate that the cursor is closed by
            // sending a cursorId of 0 back to the client.
            //
            // On the other hand, if we retrieve all results necessary for this batch, then
            // 'saveClientCursor' is true and we send a valid cursorId back to the client. In
            // this case, there may or may not actually be more results (for example, the next call
            // to getNext(...) might just return EOF).
            bool saveClientCursor = false;

            if (PlanExecutor::DEAD == state || PlanExecutor::EXEC_ERROR == state) {
                // Propagate this error to caller.
                if (PlanExecutor::EXEC_ERROR == state) {
                    scoped_ptr<PlanStageStats> stats(exec->getStats());
                    error() << "Plan executor error, stats: "
                            << Explain::statsToBSON(*stats);
                    uasserted(17406, "getMore executor error: " +
                              WorkingSetCommon::toStatusString(obj));
                }

                // If we're dead there's no way to get more results.
                saveClientCursor = false;

//.........这里部分代码省略.........
        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.
        ClientCursor::registerRunner(runner.get());
        runner->setYieldPolicy(Runner::YIELD_AUTO);
        auto_ptr<DeregisterEvenIfUnderlyingCodeThrows> safety(
            new DeregisterEvenIfUnderlyingCodeThrows(runner.get()));

        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();

        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;

            // 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;
            }
        }

        // If we cache the runner later, we want to deregister it as it receives notifications
        // anyway by virtue of being cached.
        //

    /**
     * Also called by db/ops/query.cpp.  This is the new getMore entry point.
     */
    QueryResult* newGetMore(const char* ns, int ntoreturn, long long cursorid, CurOp& curop,
                            int pass, bool& exhaust, bool* isCursorAuthorized) {
        exhaust = false;
        int bufSize = 512 + sizeof(QueryResult) + MaxBytesToReturnToClientAtOnce;

        BufBuilder bb(bufSize);
        bb.skip(sizeof(QueryResult));

        // This is a read lock.  TODO: There is a cursor flag for not needing this.  Do we care?
        Client::ReadContext ctx(ns);

        QLOG() << "running getMore in new system, cursorid " << cursorid << endl;

        // This checks to make sure the operation is allowed on a replicated node.  Since we are not
        // passing in a query object (necessary to check SlaveOK query option), the only state where
        // reads are allowed is PRIMARY (or master in master/slave).  This function uasserts if
        // reads are not okay.
        replVerifyReadsOk();

        // A pin performs a CC lookup and if there is a CC, increments the CC's pin value so it
        // doesn't time out.  Also informs ClientCursor that there is somebody actively holding the
        // CC, so don't delete it.
        ClientCursorPin ccPin(cursorid);
        ClientCursor* cc = ccPin.c();

        // These are set in the QueryResult msg we return.
        int resultFlags = ResultFlag_AwaitCapable;

        int numResults = 0;
        int startingResult = 0;

        if (NULL == cc) {
            cursorid = 0;
            resultFlags = ResultFlag_CursorNotFound;
        }
        else {
            // Quote: check for spoofing of the ns such that it does not match the one originally
            // there for the cursor
            uassert(17011, "auth error", str::equals(ns, cc->ns().c_str()));
            *isCursorAuthorized = true;

            // TODO: fail point?

            // If the operation that spawned this cursor had a time limit set, apply leftover
            // time to this getmore.
            curop.setMaxTimeMicros(cc->getLeftoverMaxTimeMicros());
            killCurrentOp.checkForInterrupt(); // May trigger maxTimeAlwaysTimeOut fail point.

            // TODO:
            // curop.debug().query = BSONForQuery
            // curop.setQuery(curop.debug().query);

            // TODO: What is pass?
            if (0 == pass) { cc->updateSlaveLocation(curop); }

            CollectionMetadataPtr collMetadata = cc->getCollMetadata();

            // If we're replaying the oplog, we save the last time that we read.
            OpTime slaveReadTill;

            // What number result are we starting at?  Used to fill out the reply.
            startingResult = cc->pos();

            // What gives us results.
            Runner* runner = cc->getRunner();
            const int queryOptions = cc->queryOptions();

            // Get results out of the runner.
            runner->restoreState();

            BSONObj obj;
            Runner::RunnerState state;
            while (Runner::RUNNER_ADVANCED == (state = runner->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 (queryOptions & QueryOption_OplogReplay) {
                    BSONElement e = obj["ts"];
                    if (Date == e.type() || Timestamp == e.type()) {
                        slaveReadTill = e._opTime();
                    }
                }

                if ((ntoreturn && numResults >= ntoreturn)
                    || bb.len() > MaxBytesToReturnToClientAtOnce) {
                    break;
                }
            }

            if (Runner::RUNNER_EOF == state && 0 == numResults
                && (queryOptions & QueryOption_CursorTailable)
                && (queryOptions & QueryOption_AwaitData) && (pass < 1000)) {
                // If the cursor is tailable we don't kill it if it's eof.  We let it try to get
//.........这里部分代码省略.........

    /**
     * Run a query with a cursor provided by the query optimizer, or FindingStartCursor.
     * @yields the db lock.
     */
    string queryWithQueryOptimizer( int queryOptions, const string& ns,
                                    const BSONObj &jsobj, CurOp& curop,
                                    const BSONObj &query, const BSONObj &order,
                                    const shared_ptr<ParsedQuery> &pq_shared,
                                    const BSONObj &oldPlan,
                                    const ChunkVersion &shardingVersionAtStart,
                                    scoped_ptr<PageFaultRetryableSection>& parentPageFaultSection,
                                    scoped_ptr<NoPageFaultsAllowed>& noPageFault,
                                    Message &result ) {

        const ParsedQuery &pq( *pq_shared );
        shared_ptr<Cursor> cursor;
        QueryPlanSummary queryPlan;
        
        if ( pq.hasOption( QueryOption_OplogReplay ) ) {
            cursor = FindingStartCursor::getCursor( ns.c_str(), query, order );
        }
        else {
            cursor = getOptimizedCursor( ns.c_str(),
                                         query,
                                         order,
                                         QueryPlanSelectionPolicy::any(),
                                         pq_shared,
                                         false,
                                         &queryPlan );
        }
        verify( cursor );
        
        scoped_ptr<QueryResponseBuilder> queryResponseBuilder
                ( QueryResponseBuilder::make( pq, cursor, queryPlan, oldPlan ) );
        bool saveClientCursor = false;
        OpTime slaveReadTill;
        ClientCursorHolder ccPointer( new ClientCursor( QueryOption_NoCursorTimeout, cursor,
                                                         ns ) );
        
        for( ; cursor->ok(); cursor->advance() ) {

            bool yielded = false;
            if ( !ccPointer->yieldSometimes( ClientCursor::MaybeCovered, &yielded ) ||
                !cursor->ok() ) {
                cursor.reset();
                queryResponseBuilder->noteYield();
                // !!! TODO The queryResponseBuilder still holds cursor.  Currently it will not do
                // anything unsafe with the cursor in handoff(), but this is very fragile.
                //
                // We don't fail the query since we're fine with returning partial data if the
                // collection was dropped.
                // NOTE see SERVER-2454.
                // TODO This is wrong.  The cursor could be gone if the closeAllDatabases command
                // just ran.
                break;
            }

            if ( yielded ) {
                queryResponseBuilder->noteYield();
            }
            
            if ( pq.getMaxScan() && cursor->nscanned() > pq.getMaxScan() ) {
                break;
            }
            
            if ( !queryResponseBuilder->addMatch() ) {
                continue;
            }
            
            // Note slave's position in the oplog.
            if ( pq.hasOption( QueryOption_OplogReplay ) ) {
                BSONObj current = cursor->current();
                BSONElement e = current["ts"];
                if ( e.type() == Date || e.type() == Timestamp ) {
                    slaveReadTill = e._opTime();
                }
            }
            
            if ( !cursor->supportGetMore() || pq.isExplain() ) {
                if ( queryResponseBuilder->enoughTotalResults() ) {
                    break;
                }
            }
            else if ( queryResponseBuilder->enoughForFirstBatch() ) {
                // if only 1 requested, no cursor saved for efficiency...we assume it is findOne()
                if ( pq.wantMore() && pq.getNumToReturn() != 1 ) {
                    queryResponseBuilder->finishedFirstBatch();
                    if ( cursor->advance() ) {
                        saveClientCursor = true;
                    }
                }
                break;
            }
        }
        
        if ( cursor ) {
            if ( pq.hasOption( QueryOption_CursorTailable ) && pq.getNumToReturn() != 1 ) {
                cursor->setTailable();
            }
            
//.........这里部分代码省略.........

    /**
     * Also called by db/ops/query.cpp.  This is the new getMore entry point.
     */
    QueryResult* newGetMore(const char* ns, int ntoreturn, long long cursorid, CurOp& curop,
                            int pass, bool& exhaust, bool* isCursorAuthorized) {
        exhaust = false;
        int bufSize = 512 + sizeof(QueryResult) + MaxBytesToReturnToClientAtOnce;

        BufBuilder bb(bufSize);
        bb.skip(sizeof(QueryResult));

        // This is a read lock.  TODO: There is a cursor flag for not needing this.  Do we care?
        Client::ReadContext ctx(ns);

        log() << "running getMore in new system, cursorid " << cursorid << endl;

        // TODO: Document.
        // TODO: do this when we can pass in our own parsed query
        //replVerifyReadsOk();

        // A pin performs a CC lookup and if there is a CC, increments the CC's pin value so it
        // doesn't time out.  Also informs ClientCursor that there is somebody actively holding the
        // CC, so don't delete it.
        ClientCursorPin ccPin(cursorid);
        ClientCursor* cc = ccPin.c();

        // These are set in the QueryResult msg we return.
        int resultFlags = ResultFlag_AwaitCapable;

        int numResults = 0;
        int startingResult = 0;

        if (NULL == cc) {
            cursorid = 0;
            resultFlags = ResultFlag_CursorNotFound;
        }
        else {
            // Quote: check for spoofing of the ns such that it does not match the one originally
            // there for the cursor
            uassert(17011, "auth error", str::equals(ns, cc->ns().c_str()));
            *isCursorAuthorized = true;

            // TODO: fail point?

            // If the operation that spawned this cursor had a time limit set, apply leftover
            // time to this getmore.
            curop.setMaxTimeMicros(cc->getLeftoverMaxTimeMicros());
            // TODO:
            // curop.debug().query = BSONForQuery
            // curop.setQuery(curop.debug().query);

            // TODO: What is pass?
            if (0 == pass) { cc->updateSlaveLocation(curop); }

            CollectionMetadataPtr collMetadata = cc->getCollMetadata();

            // If we're replaying the oplog, we save the last time that we read.
            OpTime slaveReadTill;

            // What number result are we starting at?  Used to fill out the reply.
            startingResult = cc->pos();

            // What gives us results.
            Runner* runner = cc->getRunner();
            const int queryOptions = cc->queryOptions();

            // Get results out of the runner.
            // TODO: There may be special handling required for tailable cursors?
            runner->restoreState();

            BSONObj obj;
            Runner::RunnerState state;
            while (Runner::RUNNER_ADVANCED == (state = runner->getNext(&obj, NULL))) {
                // If we're sharded make sure that we don't return any data that hasn't been
                // migrated off of our shard yet.
                if (collMetadata) {
                    KeyPattern kp(collMetadata->getKeyPattern());
                    if (!collMetadata->keyBelongsToMe(kp.extractSingleKey(obj))) { continue; }
                }

                // 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 (queryOptions & QueryOption_OplogReplay) {
                    BSONElement e = obj["ts"];
                    if (Date == e.type() || Timestamp == e.type()) {
                        slaveReadTill = e._opTime();
                    }
                }

                if ((numResults && numResults >= ntoreturn)
                    || bb.len() > MaxBytesToReturnToClientAtOnce) {
                    break;
                }
            }

//.........这里部分代码省略.........

    /**
     * This is called by db/ops/query.cpp.  This is the entry point for answering a query.
     */
    string newRunQuery(Message& m, QueryMessage& q, CurOp& curop, Message &result) {
        log() << "Running query on new system: " << q.query.toString() << endl;

        // This is a read lock.
        Client::ReadContext ctx(q.ns, dbpath);

        // Parse, canonicalize, plan, transcribe, and get a runner.
        Runner* rawRunner;
        CanonicalQuery* cq;
        Status status = getRunner(q, &rawRunner, &cq);
        if (!status.isOK()) {
            uasserted(17007, "Couldn't process query " + q.query.toString()
                         + " why: " + 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(q.ns);

        // We use this a lot below.
        const LiteParsedQuery& pq = cq->getParsed();

        // TODO: Document why we do this.
        // TODO: do this when we can pass in our own parsed query
        //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.
        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, so we must register it with the active
        // runners list in ClientCursor.
        ClientCursor::registerRunner(runner.get());
        runner->setYieldPolicy(Runner::YIELD_AUTO);

        BSONObj obj;
        Runner::RunnerState state;

        while (Runner::RUNNER_ADVANCED == (state = runner->getNext(&obj, NULL))) {
            // If we're sharded make sure that we don't return any data that hasn't been migrated
            // off of our shared yet.
            if (collMetadata) {
                // This information can change if we yield and as such we must make sure to re-fetch
                // it if we yield.
                KeyPattern kp(collMetadata->getKeyPattern());
                // This performs excessive BSONObj creation but that's OK for now.
                if (!collMetadata->keyBelongsToMe(kp.extractSingleKey(obj))) { continue; }
            }

            // 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.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;
            const bool isExplain = pq.isExplain();
            if (isExplain && enoughForExplain(pq, numResults)) {
                break;
            }
            else if (!supportsGetMore && (enough(pq, numResults)
                                          || bb.len() >= MaxBytesToReturnToClientAtOnce)) {
                break;
            }
            else if (enoughForFirstBatch(pq, numResults, bb.len())) {
//.........这里部分代码省略.........

//.........这里部分代码省略.........
        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 ReplSetHeartbeatResponse::initialize(const BSONObj& doc) {
    // Old versions set this even though they returned not "ok"
    _mismatch = doc[kMismatchFieldName].trueValue();
    if (_mismatch)
        return Status(ErrorCodes::InconsistentReplicaSetNames, "replica set name doesn't match.");

    // Old versions sometimes set the replica set name ("set") but ok:0
    const BSONElement replSetNameElement = doc[kReplSetFieldName];
    if (replSetNameElement.eoo()) {
        _setName.clear();
    } else if (replSetNameElement.type() != String) {
        return Status(ErrorCodes::TypeMismatch,
                      str::stream() << "Expected \"" << kReplSetFieldName
                                    << "\" field in response to replSetHeartbeat to have "
                                       "type String, but found "
                                    << typeName(replSetNameElement.type()));
    } else {
        _setName = replSetNameElement.String();
    }

    if (_setName.empty() && !doc[kOkFieldName].trueValue()) {
        std::string errMsg = doc[kErrMsgFieldName].str();

        BSONElement errCodeElem = doc[kErrorCodeFieldName];
        if (errCodeElem.ok()) {
            if (!errCodeElem.isNumber())
                return Status(ErrorCodes::BadValue, "Error code is not a number!");

            int errorCode = errCodeElem.numberInt();
            return Status(ErrorCodes::Error(errorCode), errMsg);
        }
        return Status(ErrorCodes::UnknownError, errMsg);
    }

    const BSONElement hasDataElement = doc[kHasDataFieldName];
    _hasDataSet = !hasDataElement.eoo();
    _hasData = hasDataElement.trueValue();

    const BSONElement electionTimeElement = doc[kElectionTimeFieldName];
    if (electionTimeElement.eoo()) {
        _electionTimeSet = false;
    } else if (electionTimeElement.type() == Timestamp) {
        _electionTimeSet = true;
        _electionTime = electionTimeElement._opTime();
    } else if (electionTimeElement.type() == Date) {
        _electionTimeSet = true;
        _electionTime = OpTime(electionTimeElement.date());
    } else {
        return Status(ErrorCodes::TypeMismatch,
                      str::stream() << "Expected \"" << kElectionTimeFieldName
                                    << "\" field in response to replSetHeartbeat "
                                       "command to have type Date or Timestamp, but found type "
                                    << typeName(electionTimeElement.type()));
    }

    const BSONElement timeElement = doc[kTimeFieldName];
    if (timeElement.eoo()) {
        _timeSet = false;
    } else if (timeElement.isNumber()) {
        _timeSet = true;
        _time = Seconds(timeElement.numberLong());
    } else {
        return Status(ErrorCodes::TypeMismatch,
                      str::stream() << "Expected \"" << kTimeFieldName
                                    << "\" field in response to replSetHeartbeat "
                                       "command to have a numeric type, but found type "
                                    << typeName(timeElement.type()));
    }

    const BSONElement opTimeElement = doc[kOpTimeFieldName];
    if (opTimeElement.eoo()) {
        _opTimeSet = false;
    } else if (opTimeElement.type() == Timestamp) {
        _opTimeSet = true;
        _opTime = opTimeElement._opTime();
    } else if (opTimeElement.type() == Date) {
        _opTimeSet = true;
        _opTime = OpTime(opTimeElement.date());
    } else {
        return Status(ErrorCodes::TypeMismatch,
                      str::stream() << "Expected \"" << kOpTimeFieldName
                                    << "\" field in response to replSetHeartbeat "
                                       "command to have type Date or Timestamp, but found type "
                                    << typeName(opTimeElement.type()));
    }

    const BSONElement electableElement = doc[kIsElectableFieldName];
    if (electableElement.eoo()) {
        _electableSet = false;
    } else {
        _electableSet = true;
        _electable = electableElement.trueValue();
    }

    _isReplSet = doc[kIsReplSetFieldName].trueValue();

    const BSONElement memberStateElement = doc[kMemberStateFieldName];
    if (memberStateElement.eoo()) {
        _stateSet = false;
    } else if (memberStateElement.type() != NumberInt && memberStateElement.type() != NumberLong) {
//.........这里部分代码省略.........

    /**
     * Also called by db/ops/query.cpp.  This is the new getMore entry point.
     */
    QueryResult* newGetMore(const char* ns, int ntoreturn, long long cursorid, CurOp& curop,
                            int pass, bool& exhaust, bool* isCursorAuthorized) {
        exhaust = false;
        int bufSize = 512 + sizeof(QueryResult) + MaxBytesToReturnToClientAtOnce;

        BufBuilder bb(bufSize);
        bb.skip(sizeof(QueryResult));

        // This is a read lock.  TODO: There is a cursor flag for not needing this.  Do we care?
        Client::ReadContext ctx(ns);

        // TODO: Document.
        replVerifyReadsOk();

        ClientCursorPin ccPin(cursorid);
        ClientCursor* cc = ccPin.c();

        // These are set in the QueryResult msg we return.
        int resultFlags = ResultFlag_AwaitCapable;

        int numResults = 0;
        int startingResult = 0;

        if (NULL == cc) {
            cursorid = 0;
            resultFlags = ResultFlag_CursorNotFound;
        }
        else {
            // Quote: check for spoofing of the ns such that it does not match the one originally
            // there for the cursor
            uassert(17011, "auth error", str::equals(ns, cc->ns().c_str()));
            *isCursorAuthorized = true;

            // TODO: fail point?

            // If the operation that spawned this cursor had a time limit set, apply leftover
            // time to this getmore.
            curop.setMaxTimeMicros(cc->getLeftoverMaxTimeMicros());
            // TODO:
            // curop.debug().query = BSONForQuery
            // curop.setQuery(curop.debug().query);

            // TODO: What is pass?
            if (0 == pass) { cc->updateSlaveLocation(curop); }

            CollectionMetadataPtr collMetadata = cc->getCollMetadata();

            // If we're replaying the oplog, we save the last time that we read.
            OpTime slaveReadTill;

            startingResult = cc->pos();

            Runner* runner = cc->getRunner();
            const ParsedQuery& pq = runner->getQuery().getParsed();

            // Get results out of the runner.
            // TODO: There may be special handling required for tailable cursors?
            runner->restoreState();
            BSONObj obj;
            // TODO: Differentiate EOF from error.
            while (runner->getNext(&obj)) {
                // If we're sharded make sure that we don't return any data that hasn't been
                // migrated off of our shard yet.
                if (collMetadata) {
                    KeyPattern kp(collMetadata->getKeyPattern());
                    if (!collMetadata->keyBelongsToMe(kp.extractSingleKey(obj))) { continue; }
                }

                // 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.hasOption(QueryOption_OplogReplay)) {
                    BSONElement e = obj["ts"];
                    if (Date == e.type() || Timestamp == e.type()) {
                        slaveReadTill = e._opTime();
                    }
                }

                if ((numResults && numResults >= ntoreturn)
                    || bb.len() > MaxBytesToReturnToClientAtOnce) {
                    break;
                }
            }

            cc->incPos(numResults);
            runner->saveState();

            // Possibly note slave's position in the oplog.
            if (pq.hasOption(QueryOption_OplogReplay) && !slaveReadTill.isNull()) {
                cc->slaveReadTill(slaveReadTill);
            }

            exhaust = pq.hasOption(QueryOption_Exhaust);
//.........这里部分代码省略.........

    /**
     * Run a query with a cursor provided by the query optimizer, or FindingStartCursor.
     * @yields the db lock.
     */
    const char *queryWithQueryOptimizer( Message &m, int queryOptions, const char *ns,
                                        const BSONObj &jsobj, CurOp& curop,
                                        const BSONObj &query, const BSONObj &order,
                                        const shared_ptr<ParsedQuery> &pq_shared,
                                        const BSONObj &oldPlan,
                                        const ConfigVersion &shardingVersionAtStart,
                                        Message &result ) {

        const ParsedQuery &pq( *pq_shared );
        shared_ptr<Cursor> cursor;
        QueryPlanSummary queryPlan;
        
        if ( pq.hasOption( QueryOption_OplogReplay ) ) {
            cursor = FindingStartCursor::getCursor( ns, query, order );
        }
        else {
            cursor =
            NamespaceDetailsTransient::getCursor( ns, query, order, QueryPlanSelectionPolicy::any(),
                                                 0, pq_shared, &queryPlan );
        }
        verify( cursor );
        
        QueryResponseBuilder queryResponseBuilder( pq, cursor, queryPlan, oldPlan );
        bool saveClientCursor = false;
        const char *exhaust = 0;
        OpTime slaveReadTill;
        ClientCursor::CleanupPointer ccPointer;
        ccPointer.reset( new ClientCursor( QueryOption_NoCursorTimeout, cursor, ns ) );
        
        for( ; cursor->ok(); cursor->advance() ) {

            bool yielded = false;
            if ( !ccPointer->yieldSometimes( ClientCursor::MaybeCovered, &yielded ) ||
                !cursor->ok() ) {
                cursor.reset();
                queryResponseBuilder.noteYield();
                // !!! TODO The queryResponseBuilder still holds cursor.  Currently it will not do
                // anything unsafe with the cursor in handoff(), but this is very fragile.
                //
                // We don't fail the query since we're fine with returning partial data if the
                // collection was dropped.
                // NOTE see SERVER-2454.
                // TODO This is wrong.  The cursor could be gone if the closeAllDatabases command
                // just ran.
                break;
            }

            if ( yielded ) {
                queryResponseBuilder.noteYield();
            }
            
            if ( pq.getMaxScan() && cursor->nscanned() > pq.getMaxScan() ) {
                break;
            }
            
            if ( !queryResponseBuilder.addMatch() ) {
                continue;
            }
            
            // Note slave's position in the oplog.
            if ( pq.hasOption( QueryOption_OplogReplay ) ) {
                BSONObj current = cursor->current();
                BSONElement e = current["ts"];
                if ( e.type() == Date || e.type() == Timestamp ) {
                    slaveReadTill = e._opTime();
                }
            }
            
            if ( !cursor->supportGetMore() || pq.isExplain() ) {
                if ( queryResponseBuilder.enoughTotalResults() ) {
                    break;
                }
            }
            else if ( queryResponseBuilder.enoughForFirstBatch() ) {
                // if only 1 requested, no cursor saved for efficiency...we assume it is findOne()
                if ( pq.wantMore() && pq.getNumToReturn() != 1 ) {
                    queryResponseBuilder.finishedFirstBatch();
                    if ( cursor->advance() ) {
                        saveClientCursor = true;
                    }
                }
                break;
            }
        }
        
        if ( cursor ) {
            if ( pq.hasOption( QueryOption_CursorTailable ) && pq.getNumToReturn() != 1 ) {
                cursor->setTailable();
            }
            
            // If the tailing request succeeded.
            if ( cursor->tailable() ) {
                saveClientCursor = true;
            }
        }
        
//.........这里部分代码省略.........