void NetworkTestEnv::onFindWithMetadataCommand(OnFindCommandWithMetadataFunction func) { onCommandWithMetadata([&func](const RemoteCommandRequest& request) -> RemoteCommandResponse { const auto& resultStatus = func(request); if (!resultStatus.isOK()) { return resultStatus.getStatus(); } std::vector<BSONObj> result; BSONObj metadata; std::tie(result, metadata) = resultStatus.getValue(); BSONArrayBuilder arr; for (const auto& obj : result) { arr.append(obj); } const NamespaceString nss = NamespaceString(request.dbname, request.cmdObj.firstElement().String()); BSONObjBuilder resultBuilder; appendCursorResponseObject(0LL, nss.toString(), arr.arr(), &resultBuilder); return RemoteCommandResponse(resultBuilder.obj(), metadata, Milliseconds(1)); }); }
void NetworkTestEnv::onFindCommand(OnFindCommandFunction func) { onCommand([&func](const RemoteCommandRequest& request) -> StatusWith<BSONObj> { const auto& resultStatus = func(request); if (!resultStatus.isOK()) { return resultStatus.getStatus(); } BSONArrayBuilder arr; for (const auto& obj : resultStatus.getValue()) { arr.append(obj); } const NamespaceString nss = NamespaceString(request.dbname, request.cmdObj.firstElement().String()); BSONObjBuilder result; appendCursorResponseObject(0LL, nss.toString(), arr.arr(), &result); return result.obj(); }); }
bool appendEmptyResultSet(OperationContext* opCtx, BSONObjBuilder& result, Status status, const std::string& ns) { invariant(!status.isOK()); CurOp::get(opCtx)->debug().nreturned = 0; CurOp::get(opCtx)->debug().nShards = 0; if (status == ErrorCodes::NamespaceNotFound) { // Old style reply result << "result" << BSONArray(); // New (command) style reply appendCursorResponseObject(0LL, ns, BSONArray(), &result); return true; } uassertStatusOK(status); return true; }
RemoteCommandResponse upconvertLegacyQueryResponse(std::int32_t requestId, StringData cursorNamespace, const Message& response) { auto swBatch = getBatchFromReply(requestId, response); if (!swBatch.isOK()) { return {swBatch.getStatus()}; } BSONArray batch; CursorId cursorId; std::tie(cursorId, batch) = std::move(swBatch.getValue()); BSONObjBuilder result; appendCursorResponseObject(cursorId, cursorNamespace, std::move(batch), &result); // Using Command::appendCommandStatus would create a circular dep, so it's simpler to just do // this. result.append("ok", 1.0); RemoteCommandResponse upconvertedResponse; upconvertedResponse.data = result.obj(); return upconvertedResponse; }
bool run(OperationContext* txn, const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result) { BSONElement first = cmdObj.firstElement(); uassert(28528, str::stream() << "Argument to listIndexes must be of type String, not " << typeName(first.type()), first.type() == String); StringData collectionName = first.valueStringData(); uassert(28529, str::stream() << "Argument to listIndexes must be a collection name, " << "not the empty string", !collectionName.empty()); const NamespaceString ns(dbname, collectionName); const long long defaultBatchSize = std::numeric_limits<long long>::max(); long long batchSize; Status parseCursorStatus = parseCommandCursorOptions(cmdObj, defaultBatchSize, &batchSize); if (!parseCursorStatus.isOK()) { return appendCommandStatus(result, parseCursorStatus); } AutoGetCollectionForRead autoColl(txn, ns); if (!autoColl.getDb()) { return appendCommandStatus(result, Status(ErrorCodes::NamespaceNotFound, "no database")); } const Collection* collection = autoColl.getCollection(); if (!collection) { return appendCommandStatus(result, Status(ErrorCodes::NamespaceNotFound, "no collection")); } const CollectionCatalogEntry* cce = collection->getCatalogEntry(); invariant(cce); vector<string> indexNames; MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN { indexNames.clear(); cce->getAllIndexes(txn, &indexNames); } MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "listIndexes", ns.ns()); std::unique_ptr<WorkingSet> ws(new WorkingSet()); std::unique_ptr<QueuedDataStage> root(new QueuedDataStage(ws.get())); for (size_t i = 0; i < indexNames.size(); i++) { BSONObj indexSpec; MONGO_WRITE_CONFLICT_RETRY_LOOP_BEGIN { indexSpec = cce->getIndexSpec(txn, indexNames[i]); } MONGO_WRITE_CONFLICT_RETRY_LOOP_END(txn, "listIndexes", ns.ns()); WorkingSetID id = ws->allocate(); WorkingSetMember* member = ws->get(id); member->keyData.clear(); member->loc = RecordId(); member->obj = Snapshotted<BSONObj>(SnapshotId(), indexSpec.getOwned()); member->transitionToOwnedObj(); root->pushBack(id); } std::string cursorNamespace = str::stream() << dbname << ".$cmd." << name << "." << ns.coll(); dassert(NamespaceString(cursorNamespace).isValid()); dassert(NamespaceString(cursorNamespace).isListIndexesCursorNS()); dassert(ns == NamespaceString(cursorNamespace).getTargetNSForListIndexes()); auto statusWithPlanExecutor = PlanExecutor::make( txn, std::move(ws), std::move(root), cursorNamespace, PlanExecutor::YIELD_MANUAL); if (!statusWithPlanExecutor.isOK()) { return appendCommandStatus(result, statusWithPlanExecutor.getStatus()); } std::unique_ptr<PlanExecutor> exec = std::move(statusWithPlanExecutor.getValue()); BSONArrayBuilder firstBatch; const int byteLimit = MaxBytesToReturnToClientAtOnce; for (long long objCount = 0; objCount < batchSize && firstBatch.len() < byteLimit; objCount++) { BSONObj next; PlanExecutor::ExecState state = exec->getNext(&next, NULL); if (state == PlanExecutor::IS_EOF) { break; } invariant(state == PlanExecutor::ADVANCED); firstBatch.append(next); } CursorId cursorId = 0LL; if (!exec->isEOF()) { exec->saveState(); ClientCursor* cursor = new ClientCursor( CursorManager::getGlobalCursorManager(), exec.release(), cursorNamespace); cursorId = cursor->cursorid(); } appendCursorResponseObject(cursorId, cursorNamespace, firstBatch.arr(), &result); return true; }
/** * Runs a query using the following steps: * 1) Parsing. * 2) Acquire locks. * 3) Plan query, obtaining an executor that can run it. * 4) Setup a cursor for the query, which may be used on subsequent getMores. * 5) Generate the first batch. * 6) Save state for getMore. * 7) Generate response to send to the client. * * TODO: Rather than using the sharding version available in thread-local storage (i.e. the * call to ShardingState::needCollectionMetadata() below), shard version information * should be passed as part of the command parameter. */ bool run(OperationContext* txn, const std::string& dbname, BSONObj& cmdObj, int options, std::string& errmsg, BSONObjBuilder& result) override { const std::string fullns = parseNs(dbname, cmdObj); const NamespaceString nss(fullns); if (!nss.isValid()) { return appendCommandStatus(result, {ErrorCodes::InvalidNamespace, str::stream() << "Invalid collection name: " << nss.ns()}); } // Although it is a command, a find command gets counted as a query. globalOpCounters.gotQuery(); if (txn->getClient()->isInDirectClient()) { return appendCommandStatus( result, Status(ErrorCodes::IllegalOperation, "Cannot run find command from eval()")); } // 1a) Parse the command BSON to a LiteParsedQuery. const bool isExplain = false; auto lpqStatus = LiteParsedQuery::makeFromFindCommand(nss, cmdObj, isExplain); if (!lpqStatus.isOK()) { return appendCommandStatus(result, lpqStatus.getStatus()); } auto& lpq = lpqStatus.getValue(); // Validate term, if provided. if (auto term = lpq->getReplicationTerm()) { auto replCoord = repl::ReplicationCoordinator::get(txn); Status status = replCoord->updateTerm(*term); // Note: updateTerm returns ok if term stayed the same. if (!status.isOK()) { return appendCommandStatus(result, status); } } // Fill out curop information. long long ntoreturn = lpq->getBatchSize().value_or(0); beginQueryOp(txn, nss, cmdObj, ntoreturn, lpq->getSkip()); // 1b) Finish the parsing step by using the LiteParsedQuery to create a CanonicalQuery. WhereCallbackReal whereCallback(txn, nss.db()); auto statusWithCQ = CanonicalQuery::canonicalize(lpq.release(), whereCallback); if (!statusWithCQ.isOK()) { return appendCommandStatus(result, statusWithCQ.getStatus()); } std::unique_ptr<CanonicalQuery> cq = std::move(statusWithCQ.getValue()); // 2) Acquire locks. AutoGetCollectionForRead ctx(txn, nss); Collection* collection = ctx.getCollection(); const int dbProfilingLevel = ctx.getDb() ? ctx.getDb()->getProfilingLevel() : serverGlobalParams.defaultProfile; ShardingState* const shardingState = ShardingState::get(txn); // It is possible that the sharding version will change during yield while we are // retrieving a plan executor. If this happens we will throw an error and mongos will // retry. const ChunkVersion shardingVersionAtStart = shardingState->getVersion(nss.ns()); // 3) Get the execution plan for the query. auto statusWithPlanExecutor = getExecutorFind(txn, collection, nss, std::move(cq), PlanExecutor::YIELD_AUTO); if (!statusWithPlanExecutor.isOK()) { return appendCommandStatus(result, statusWithPlanExecutor.getStatus()); } std::unique_ptr<PlanExecutor> exec = std::move(statusWithPlanExecutor.getValue()); // 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)) { // Version changed while retrieving a PlanExecutor. Terminate the operation, // signaling that mongos should retry. throw SendStaleConfigException(nss.ns(), "version changed during find command", shardingVersionAtStart, shardingState->getVersion(nss.ns())); } if (!collection) { // No collection. Just fill out curop indicating that there were zero results and // there is no ClientCursor id, and then return. const long long numResults = 0; const CursorId cursorId = 0; endQueryOp(txn, *exec, dbProfilingLevel, numResults, cursorId); appendCursorResponseObject(cursorId, nss.ns(), BSONArray(), &result); return true; } const LiteParsedQuery& pq = exec->getCanonicalQuery()->getParsed(); // 4) If possible, register the execution plan inside a ClientCursor, and pin that // cursor. In this case, ownership of the PlanExecutor is transferred to the // ClientCursor, and 'exec' becomes null. // // First unregister the PlanExecutor so it can be re-registered with ClientCursor. exec->deregisterExec(); // Create a ClientCursor containing this plan executor. We don't have to worry // about leaking it as it's inserted into a global map by its ctor. ClientCursor* cursor = new ClientCursor(collection->getCursorManager(), exec.release(), nss.ns(), txn->recoveryUnit()->isReadingFromMajorityCommittedSnapshot(), pq.getOptions(), pq.getFilter()); CursorId cursorId = cursor->cursorid(); ClientCursorPin ccPin(collection->getCursorManager(), cursorId); // On early return, get rid of the the cursor. ScopeGuard cursorFreer = MakeGuard(&ClientCursorPin::deleteUnderlying, ccPin); invariant(!exec); PlanExecutor* cursorExec = cursor->getExecutor(); // 5) Stream query results, adding them to a BSONArray as we go. BSONArrayBuilder firstBatch; BSONObj obj; PlanExecutor::ExecState state; long long numResults = 0; while (!enoughForFirstBatch(pq, numResults, firstBatch.len()) && PlanExecutor::ADVANCED == (state = cursorExec->getNext(&obj, NULL))) { // If adding this object will cause us to exceed the BSON size limit, then we stash // it for later. if (firstBatch.len() + obj.objsize() > BSONObjMaxUserSize && numResults > 0) { cursorExec->enqueue(obj); break; } // Add result to output buffer. firstBatch.append(obj); numResults++; } // Throw an assertion if query execution fails for any reason. if (PlanExecutor::FAILURE == state || PlanExecutor::DEAD == state) { const std::unique_ptr<PlanStageStats> stats(cursorExec->getStats()); error() << "Plan executor error during find command: " << PlanExecutor::statestr(state) << ", stats: " << Explain::statsToBSON(*stats); return appendCommandStatus(result, Status(ErrorCodes::OperationFailed, str::stream() << "Executor error during find command: " << WorkingSetCommon::toStatusString(obj))); } // 6) Set up the cursor for getMore. if (shouldSaveCursor(txn, collection, state, cursorExec)) { // State will be restored on getMore. cursorExec->saveState(); cursorExec->detachFromOperationContext(); cursor->setLeftoverMaxTimeMicros(CurOp::get(txn)->getRemainingMaxTimeMicros()); cursor->setPos(numResults); } else { cursorId = 0; } // Fill out curop based on the results. endQueryOp(txn, *cursorExec, dbProfilingLevel, numResults, cursorId); // 7) Generate the response object to send to the client. appendCursorResponseObject(cursorId, nss.ns(), firstBatch.arr(), &result); if (cursorId) { cursorFreer.Dismiss(); } return true; }
/** * Runs a query using the following steps: * --Parsing. * --Acquire locks. * --Plan query, obtaining an executor that can run it. * --Generate the first batch. * --Save state for getMore, transferring ownership of the executor to a ClientCursor. * --Generate response to send to the client. */ bool run(OperationContext* txn, const std::string& dbname, BSONObj& cmdObj, int options, std::string& errmsg, BSONObjBuilder& result) override { const std::string fullns = parseNs(dbname, cmdObj); const NamespaceString nss(fullns); if (!nss.isValid() || nss.isCommand() || nss.isSpecialCommand()) { return appendCommandStatus(result, {ErrorCodes::InvalidNamespace, str::stream() << "Invalid collection name: " << nss.ns()}); } // Although it is a command, a find command gets counted as a query. globalOpCounters.gotQuery(); if (txn->getClient()->isInDirectClient()) { return appendCommandStatus( result, Status(ErrorCodes::IllegalOperation, "Cannot run find command from eval()")); } // Parse the command BSON to a LiteParsedQuery. const bool isExplain = false; auto lpqStatus = LiteParsedQuery::makeFromFindCommand(nss, cmdObj, isExplain); if (!lpqStatus.isOK()) { return appendCommandStatus(result, lpqStatus.getStatus()); } auto& lpq = lpqStatus.getValue(); // Validate term before acquiring locks, if provided. if (auto term = lpq->getReplicationTerm()) { auto replCoord = repl::ReplicationCoordinator::get(txn); Status status = replCoord->updateTerm(txn, *term); // Note: updateTerm returns ok if term stayed the same. if (!status.isOK()) { return appendCommandStatus(result, status); } } // Fill out curop information. // // We pass negative values for 'ntoreturn' and 'ntoskip' to indicate that these values // should be omitted from the log line. Limit and skip information is already present in the // find command parameters, so these fields are redundant. const int ntoreturn = -1; const int ntoskip = -1; beginQueryOp(txn, nss, cmdObj, ntoreturn, ntoskip); // Finish the parsing step by using the LiteParsedQuery to create a CanonicalQuery. ExtensionsCallbackReal extensionsCallback(txn, &nss); auto statusWithCQ = CanonicalQuery::canonicalize(lpq.release(), extensionsCallback); if (!statusWithCQ.isOK()) { return appendCommandStatus(result, statusWithCQ.getStatus()); } std::unique_ptr<CanonicalQuery> cq = std::move(statusWithCQ.getValue()); ShardingState* const shardingState = ShardingState::get(txn); if (OperationShardVersion::get(txn).hasShardVersion() && shardingState->enabled()) { ChunkVersion receivedVersion = OperationShardVersion::get(txn).getShardVersion(nss); ChunkVersion latestVersion; // Wait for migration completion to get the correct chunk version. const int maxTimeoutSec = 30; int timeoutSec = cq->getParsed().getMaxTimeMS() / 1000; if (!timeoutSec || timeoutSec > maxTimeoutSec) { timeoutSec = maxTimeoutSec; } if (!shardingState->waitTillNotInCriticalSection(timeoutSec)) { uasserted(ErrorCodes::LockTimeout, "Timeout while waiting for migration commit"); } // If the received version is newer than the version cached in 'shardingState', then we // have to refresh 'shardingState' from the config servers. We do this before acquiring // locks so that we don't hold locks while waiting on the network. uassertStatusOK(shardingState->refreshMetadataIfNeeded( txn, nss.ns(), receivedVersion, &latestVersion)); } // Acquire locks. AutoGetCollectionForRead ctx(txn, nss); Collection* collection = ctx.getCollection(); const int dbProfilingLevel = ctx.getDb() ? ctx.getDb()->getProfilingLevel() : serverGlobalParams.defaultProfile; // It is possible that the sharding version will change during yield while we are // retrieving a plan executor. If this happens we will throw an error and mongos will // retry. const ChunkVersion shardingVersionAtStart = shardingState->getVersion(nss.ns()); // Get the execution plan for the query. auto statusWithPlanExecutor = getExecutorFind(txn, collection, nss, std::move(cq), PlanExecutor::YIELD_AUTO); if (!statusWithPlanExecutor.isOK()) { return appendCommandStatus(result, statusWithPlanExecutor.getStatus()); } std::unique_ptr<PlanExecutor> exec = std::move(statusWithPlanExecutor.getValue()); if (!collection) { // No collection. Just fill out curop indicating that there were zero results and // there is no ClientCursor id, and then return. const long long numResults = 0; const CursorId cursorId = 0; endQueryOp(txn, collection, *exec, dbProfilingLevel, numResults, cursorId); appendCursorResponseObject(cursorId, nss.ns(), BSONArray(), &result); return true; } const LiteParsedQuery& pq = exec->getCanonicalQuery()->getParsed(); // Stream query results, adding them to a BSONArray as we go. BSONArrayBuilder firstBatch; BSONObj obj; PlanExecutor::ExecState state = PlanExecutor::ADVANCED; long long numResults = 0; while (!FindCommon::enoughForFirstBatch(pq, numResults, firstBatch.len()) && PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) { // If adding this object will cause us to exceed the BSON size limit, then we stash // it for later. if (firstBatch.len() + obj.objsize() > BSONObjMaxUserSize && numResults > 0) { exec->enqueue(obj); break; } // Add result to output buffer. firstBatch.append(obj); numResults++; } // Throw an assertion if query execution fails for any reason. if (PlanExecutor::FAILURE == state || PlanExecutor::DEAD == state) { const std::unique_ptr<PlanStageStats> stats(exec->getStats()); error() << "Plan executor error during find command: " << PlanExecutor::statestr(state) << ", stats: " << Explain::statsToBSON(*stats); return appendCommandStatus(result, Status(ErrorCodes::OperationFailed, str::stream() << "Executor error during find command: " << 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)) { // Version changed while retrieving a PlanExecutor. Terminate the operation, // signaling that mongos should retry. throw SendStaleConfigException(nss.ns(), "version changed during find command", shardingVersionAtStart, shardingState->getVersion(nss.ns())); } // Set up the cursor for getMore. CursorId cursorId = 0; if (shouldSaveCursor(txn, collection, state, exec.get())) { // Register the execution plan inside a ClientCursor. Ownership of the PlanExecutor is // transferred to the ClientCursor. // // First unregister the PlanExecutor so it can be re-registered with ClientCursor. exec->deregisterExec(); // Create a ClientCursor containing this plan executor. We don't have to worry about // leaking it as it's inserted into a global map by its ctor. ClientCursor* cursor = new ClientCursor(collection->getCursorManager(), exec.release(), nss.ns(), txn->recoveryUnit()->isReadingFromMajorityCommittedSnapshot(), pq.getOptions(), pq.getFilter()); cursorId = cursor->cursorid(); invariant(!exec); PlanExecutor* cursorExec = cursor->getExecutor(); // State will be restored on getMore. cursorExec->saveState(); cursorExec->detachFromOperationContext(); cursor->setLeftoverMaxTimeMicros(CurOp::get(txn)->getRemainingMaxTimeMicros()); cursor->setPos(numResults); // Fill out curop based on the results. endQueryOp(txn, collection, *cursorExec, dbProfilingLevel, numResults, cursorId); } else { endQueryOp(txn, collection, *exec, dbProfilingLevel, numResults, cursorId); } // Generate the response object to send to the client. appendCursorResponseObject(cursorId, nss.ns(), firstBatch.arr(), &result); return true; }
/** * Runs a query using the following steps: * --Parsing. * --Acquire locks. * --Plan query, obtaining an executor that can run it. * --Generate the first batch. * --Save state for getMore, transferring ownership of the executor to a ClientCursor. * --Generate response to send to the client. */ bool run(OperationContext* txn, const std::string& dbname, BSONObj& cmdObj, int options, std::string& errmsg, BSONObjBuilder& result) override { const NamespaceString nss(parseNs(dbname, cmdObj)); if (!nss.isValid() || nss.isCommand() || nss.isSpecialCommand()) { return appendCommandStatus(result, {ErrorCodes::InvalidNamespace, str::stream() << "Invalid collection name: " << nss.ns()}); } // Although it is a command, a find command gets counted as a query. globalOpCounters.gotQuery(); if (txn->getClient()->isInDirectClient()) { return appendCommandStatus( result, Status(ErrorCodes::IllegalOperation, "Cannot run find command from eval()")); } // Parse the command BSON to a QueryRequest. const bool isExplain = false; auto qrStatus = QueryRequest::makeFromFindCommand(nss, cmdObj, isExplain); if (!qrStatus.isOK()) { return appendCommandStatus(result, qrStatus.getStatus()); } auto& qr = qrStatus.getValue(); // Validate term before acquiring locks, if provided. if (auto term = qr->getReplicationTerm()) { auto replCoord = repl::ReplicationCoordinator::get(txn); Status status = replCoord->updateTerm(txn, *term); // Note: updateTerm returns ok if term stayed the same. if (!status.isOK()) { return appendCommandStatus(result, status); } } // Fill out curop information. // // We pass negative values for 'ntoreturn' and 'ntoskip' to indicate that these values // should be omitted from the log line. Limit and skip information is already present in the // find command parameters, so these fields are redundant. const int ntoreturn = -1; const int ntoskip = -1; beginQueryOp(txn, nss, cmdObj, ntoreturn, ntoskip); // Finish the parsing step by using the QueryRequest to create a CanonicalQuery. ExtensionsCallbackReal extensionsCallback(txn, &nss); auto statusWithCQ = CanonicalQuery::canonicalize(txn, std::move(qr), extensionsCallback); if (!statusWithCQ.isOK()) { return appendCommandStatus(result, statusWithCQ.getStatus()); } std::unique_ptr<CanonicalQuery> cq = std::move(statusWithCQ.getValue()); // Acquire locks. AutoGetCollectionForRead ctx(txn, nss); Collection* collection = ctx.getCollection(); // Get the execution plan for the query. auto statusWithPlanExecutor = getExecutorFind(txn, collection, nss, std::move(cq), PlanExecutor::YIELD_AUTO); if (!statusWithPlanExecutor.isOK()) { return appendCommandStatus(result, statusWithPlanExecutor.getStatus()); } std::unique_ptr<PlanExecutor> exec = std::move(statusWithPlanExecutor.getValue()); { stdx::lock_guard<Client>(*txn->getClient()); CurOp::get(txn)->setPlanSummary_inlock(Explain::getPlanSummary(exec.get())); } if (!collection) { // No collection. Just fill out curop indicating that there were zero results and // there is no ClientCursor id, and then return. const long long numResults = 0; const CursorId cursorId = 0; endQueryOp(txn, collection, *exec, numResults, cursorId); appendCursorResponseObject(cursorId, nss.ns(), BSONArray(), &result); return true; } const QueryRequest& originalQR = exec->getCanonicalQuery()->getQueryRequest(); // Stream query results, adding them to a BSONArray as we go. CursorResponseBuilder firstBatch(/*isInitialResponse*/ true, &result); BSONObj obj; PlanExecutor::ExecState state = PlanExecutor::ADVANCED; long long numResults = 0; while (!FindCommon::enoughForFirstBatch(originalQR, numResults) && 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, firstBatch.bytesUsed())) { exec->enqueue(obj); break; } // Add result to output buffer. firstBatch.append(obj); numResults++; } // Throw an assertion if query execution fails for any reason. if (PlanExecutor::FAILURE == state || PlanExecutor::DEAD == state) { firstBatch.abandon(); error() << "Plan executor error during find command: " << PlanExecutor::statestr(state) << ", stats: " << Explain::getWinningPlanStats(exec.get()); return appendCommandStatus(result, Status(ErrorCodes::OperationFailed, str::stream() << "Executor error during find command: " << 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); // Set up the cursor for getMore. CursorId cursorId = 0; if (shouldSaveCursor(txn, collection, state, exec.get())) { // Register the execution plan inside a ClientCursor. Ownership of the PlanExecutor is // transferred to the ClientCursor. // // First unregister the PlanExecutor so it can be re-registered with ClientCursor. exec->deregisterExec(); // Create a ClientCursor containing this plan executor. We don't have to worry about // leaking it as it's inserted into a global map by its ctor. ClientCursor* cursor = new ClientCursor(collection->getCursorManager(), exec.release(), nss.ns(), txn->recoveryUnit()->isReadingFromMajorityCommittedSnapshot(), originalQR.getOptions(), cmdObj.getOwned()); cursorId = cursor->cursorid(); invariant(!exec); PlanExecutor* cursorExec = cursor->getExecutor(); // State will be restored on getMore. cursorExec->saveState(); cursorExec->detachFromOperationContext(); cursor->setLeftoverMaxTimeMicros(txn->getRemainingMaxTimeMicros()); cursor->setPos(numResults); // Fill out curop based on the results. endQueryOp(txn, collection, *cursorExec, numResults, cursorId); } else { endQueryOp(txn, collection, *exec, numResults, cursorId); } // Generate the response object to send to the client. firstBatch.done(cursorId, nss.ns()); return true; }