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();
}
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() : "";
}
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);
}
/**
* Called by db/instance.cpp. This is the getMore entry point.
*
* pass - when QueryOption_AwaitData is in use, the caller will make repeated calls
* when this method returns an empty result, incrementing pass on each call.
* Thus, pass == 0 indicates this is the first "attempt" before any 'awaiting'.
*/
QueryResult::View newGetMore(OperationContext* txn,
const char* ns,
int ntoreturn,
long long cursorid,
CurOp& curop,
int pass,
bool& exhaust,
bool* isCursorAuthorized,
bool fromDBDirectClient) {
// For testing, we may want to fail if we receive a getmore.
if (MONGO_FAIL_POINT(failReceivedGetmore)) {
invariant(0);
}
exhaust = false;
// This is a read lock.
const NamespaceString nss(ns);
scoped_ptr<AutoGetCollectionForRead> ctx(new AutoGetCollectionForRead(txn, nss));
Collection* collection = ctx->getCollection();
uassert( 17356, "collection dropped between getMore calls", collection );
QLOG() << "Running getMore, 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.
Status status = repl::getGlobalReplicationCoordinator()->checkCanServeReadsFor(
txn,
nss,
true);
uassertStatusOK(status);
// 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(collection, cursorid);
ClientCursor* cc = ccPin.c();
// If we're not being called from DBDirectClient we want to associate the RecoveryUnit
// used to create the execution machinery inside the cursor with our OperationContext.
// If we throw or otherwise exit this method in a disorderly fashion, we must ensure
// that further calls to getMore won't fail, and that the provided OperationContext
// has a valid RecoveryUnit. As such, we use RAII to accomplish this.
//
// This must be destroyed before the ClientCursor is destroyed.
std::auto_ptr<ScopedRecoveryUnitSwapper> ruSwapper;
// These are set in the QueryResult msg we return.
int resultFlags = ResultFlag_AwaitCapable;
int numResults = 0;
int startingResult = 0;
const int InitialBufSize =
512 + sizeof(QueryResult::Value) + MaxBytesToReturnToClientAtOnce;
BufBuilder bb(InitialBufSize);
bb.skip(sizeof(QueryResult::Value));
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;
// Restore the RecoveryUnit if we need to.
if (fromDBDirectClient) {
if (cc->hasRecoveryUnit())
invariant(txn->recoveryUnit() == cc->getUnownedRecoveryUnit());
}
else {
if (!cc->hasRecoveryUnit()) {
// Start using a new RecoveryUnit
cc->setOwnedRecoveryUnit(
getGlobalEnvironment()->getGlobalStorageEngine()->newRecoveryUnit(txn));
}
// Swap RecoveryUnit(s) between the ClientCursor and OperationContext.
ruSwapper.reset(new ScopedRecoveryUnitSwapper(cc, txn));
}
// Reset timeout timer on the cursor since the cursor is still in use.
cc->setIdleTime(0);
// 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());
txn->checkForInterrupt(); // May trigger maxTimeAlwaysTimeOut fail point.
if (0 == pass) {
cc->updateSlaveLocation(txn, curop);
}
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;
// In the old system tailable capped cursors would be killed off at the
// cursorid level. If a tailable capped cursor is nuked the cursorid
// would vanish.
//
// In the new system they die and are cleaned up later (or time out).
// So this is where we get to remove the cursorid.
if (0 == numResults) {
resultFlags = ResultFlag_CursorNotFound;
}
}
else if (PlanExecutor::IS_EOF == state) {
// EOF is also end of the line unless it's tailable.
saveClientCursor = queryOptions & QueryOption_CursorTailable;
}
else {
verify(PlanExecutor::ADVANCED == state);
saveClientCursor = true;
}
if (!saveClientCursor) {
ruSwapper.reset();
ccPin.deleteUnderlying();
// cc is now invalid, as is the executor
cursorid = 0;
cc = NULL;
QLOG() << "getMore NOT saving client cursor, ended with state "
<< PlanExecutor::statestr(state)
<< endl;
}
else {
// Continue caching the ClientCursor.
cc->incPos(numResults);
exec->saveState();
QLOG() << "getMore saving client cursor ended with state "
<< PlanExecutor::statestr(state)
<< endl;
if (PlanExecutor::IS_EOF == state && (queryOptions & QueryOption_CursorTailable)) {
if (!fromDBDirectClient) {
// Don't stash the RU. Get a new one on the next getMore.
ruSwapper.reset();
delete cc->releaseOwnedRecoveryUnit();
}
if ((queryOptions & QueryOption_AwaitData)
&& (numResults == 0)
&& (pass < 1000)) {
// Bubble up to the AwaitData handling code in receivedGetMore which will
// try again.
return NULL;
}
}
// Possibly note slave's position in the oplog.
if ((queryOptions & QueryOption_OplogReplay) && !slaveReadTill.isNull()) {
cc->slaveReadTill(slaveReadTill);
}
exhaust = (queryOptions & QueryOption_Exhaust);
// If the getmore had a time limit, remaining time is "rolled over" back to the
// cursor (for use by future getmore ops).
cc->setLeftoverMaxTimeMicros( curop.getRemainingMaxTimeMicros() );
}
}
QueryResult::View qr = bb.buf();
qr.msgdata().setLen(bb.len());
qr.msgdata().setOperation(opReply);
qr.setResultFlags(resultFlags);
qr.setCursorId(cursorid);
qr.setStartingFrom(startingResult);
qr.setNReturned(numResults);
bb.decouple();
QLOG() << "getMore returned " << numResults << " results\n";
return qr;
}
/**
* This is called by db/ops/query.cpp. This is the entry point for answering a query.
*/
std::string newRunQuery(CanonicalQuery* cq, CurOp& curop, Message &result) {
QLOG() << "Running query on new system: " << cq->toString();
// This is a read lock.
Client::ReadContext ctx(cq->ns(), storageGlobalParams.dbpath);
// Parse, canonicalize, plan, transcribe, and get a runner.
Runner* rawRunner = NULL;
// We use this a lot below.
const LiteParsedQuery& pq = cq->getParsed();
// Need to call cq->toString() now, since upon error getRunner doesn't guarantee
// cq is in a consistent state.
string cqStr = cq->toString();
// 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 (ctx.ctx().db()->getCollection(cq->ns()) == NULL) {
rawRunner = new EOFRunner(cq, cq->ns());
}
else if (pq.hasOption(QueryOption_OplogReplay)) {
status = getOplogStartHack(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()) {
uasserted(17007, "Couldn't get runner for query because: " + status.reason() + " query is " + cqStr);
}
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.
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.
//
// 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:
// * in-memory sort using too much RAM.
if (Runner::RUNNER_ERROR == state) {
uasserted(17144, "Runner error, memory limit for sort probably exceeded");
}
// 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()));
}
// Append explain information to query results by asking the runner to produce them.
if (isExplain) {
TypeExplain* bareExplain;
Status res = runner->getExplainPlan(&bareExplain);
if (!res.isOK()) {
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.
}
else {
boost::scoped_ptr<TypeExplain> explain(bareExplain);
// Fill in the missing run-time fields in explain, starting with propeties of
// the process running the query.
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(curop.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(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;
curop.debug().ntoskip = pq.getSkip();
curop.debug().nreturned = numResults;
// curop.debug().exhaust is set above.
return curop.debug().exhaust ? pq.ns() : "";
}
/**
* 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
// data some # of times first.
return 0;
}
bool saveClientCursor = false;
if (Runner::RUNNER_DEAD == state || Runner::RUNNER_ERROR == state) {
// If we're dead there's no way to get more results.
saveClientCursor = false;
// In the old system tailable capped cursors would be killed off at the
// cursorid level. If a tailable capped cursor is nuked the cursorid
// would vanish.
//
// In the new system they die and are cleaned up later (or time out).
// So this is where we get to remove the cursorid.
if (0 == numResults) {
resultFlags = ResultFlag_CursorNotFound;
}
}
else if (Runner::RUNNER_EOF == state) {
// EOF is also end of the line unless it's tailable.
saveClientCursor = queryOptions & QueryOption_CursorTailable;
}
else {
verify(Runner::RUNNER_ADVANCED == state);
saveClientCursor = true;
}
if (!saveClientCursor) {
ccPin.deleteUnderlying();
// cc is now invalid, as is the runner
cursorid = 0;
cc = NULL;
QLOG() << "getMore NOT saving client cursor, ended w/state "
<< Runner::statestr(state)
<< endl;
}
else {
// Continue caching the ClientCursor.
cc->incPos(numResults);
runner->saveState();
QLOG() << "getMore saving client cursor ended w/state "
<< Runner::statestr(state)
<< endl;
// Possibly note slave's position in the oplog.
if ((queryOptions & QueryOption_OplogReplay) && !slaveReadTill.isNull()) {
cc->slaveReadTill(slaveReadTill);
}
exhaust = (queryOptions & QueryOption_Exhaust);
// If the getmore had a time limit, remaining time is "rolled over" back to the
// cursor (for use by future getmore ops).
cc->setLeftoverMaxTimeMicros( curop.getRemainingMaxTimeMicros() );
}
}
QueryResult* qr = reinterpret_cast<QueryResult*>(bb.buf());
qr->len = bb.len();
qr->setOperation(opReply);
qr->_resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = startingResult;
qr->nReturned = numResults;
bb.decouple();
QLOG() << "getMore returned " << numResults << " results\n";
return qr;
}
/**
* 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();
}
// If the tailing request succeeded.
if ( cursor->tailable() ) {
saveClientCursor = true;
}
}
if ( ! shardingState.getVersion( 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(ns, "version changed during initial query",
shardingVersionAtStart,
shardingState.getVersion(ns));
}
parentPageFaultSection.reset(0);
noPageFault.reset( new NoPageFaultsAllowed() );
int nReturned = queryResponseBuilder->handoff( result );
ccPointer.reset();
long long cursorid = 0;
if ( saveClientCursor ) {
// Create a new ClientCursor, with a default timeout.
ccPointer.reset( new ClientCursor( queryOptions, cursor, ns,
jsobj.getOwned() ) );
cursorid = ccPointer->cursorid();
DEV { MONGO_TLOG(2) << "query has more, cursorid: " << cursorid << endl; }
if ( cursor->supportYields() ) {
ClientCursor::YieldData data;
ccPointer->prepareToYield( data );
}
else {
ccPointer->c()->noteLocation();
}
// Save slave's position in the oplog.
if ( pq.hasOption( QueryOption_OplogReplay ) && !slaveReadTill.isNull() ) {
ccPointer->slaveReadTill( slaveReadTill );
}
if ( !ccPointer->ok() && ccPointer->c()->tailable() ) {
DEV {
MONGO_TLOG(0) << "query has no more but tailable, cursorid: " << cursorid <<
endl;
}
}
if( queryOptions & QueryOption_Exhaust ) {
curop.debug().exhaust = true;
}
// Set attributes for getMore.
ccPointer->setCollMetadata( queryResponseBuilder->collMetadata() );
ccPointer->setPos( nReturned );
ccPointer->pq = pq_shared;
ccPointer->fields = pq.getFieldPtr();
// If the query had a time limit, remaining time is "rolled over" to the cursor (for
// use by future getmore ops).
ccPointer->setLeftoverMaxTimeMicros( curop.getRemainingMaxTimeMicros() );
ccPointer.release();
}
/**
* 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;
}
}
if (Runner::RUNNER_DEAD == state || Runner::RUNNER_EOF == state) {
log() << "getMore(): runner with id " << cursorid << " EOF/DEAD, state = "
<< static_cast<int>(state) << endl;
// TODO: If the cursor is tailable we don't kill it if it's eof.
ccPin.free();
// cc is now invalid, as is the runner
cursorid = 0;
cc = NULL;
}
else {
// Continue caching the ClientCursor.
cc->incPos(numResults);
runner->saveState();
// Possibly note slave's position in the oplog.
if ((queryOptions & QueryOption_OplogReplay) && !slaveReadTill.isNull()) {
cc->slaveReadTill(slaveReadTill);
}
exhaust = (queryOptions & QueryOption_Exhaust);
// If the getmore had a time limit, remaining time is "rolled over" back to the
// cursor (for use by future getmore ops).
cc->setLeftoverMaxTimeMicros( curop.getRemainingMaxTimeMicros() );
}
}
QueryResult* qr = reinterpret_cast<QueryResult*>(bb.buf());
qr->len = bb.len();
qr->setOperation(opReply);
qr->_resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = startingResult;
qr->nReturned = numResults;
bb.decouple();
return qr;
}
/**
* 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())) {
// If only one result requested assume it's a findOne() and don't save the cursor.
if (pq.wantMore() && 1 != pq.getNumToReturn()) {
saveClientCursor = true;
}
break;
}
}
// 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.
ClientCursor::deregisterRunner(runner.get());
// Why save a dead runner?
if (Runner::RUNNER_DEAD == state) { saveClientCursor = false; }
// TODO: Stage creation can set tailable depending on what's in the parsed query. We have
// the full parsed query available during planning...set it there.
//
// TODO: If we're tailable we want to save the client cursor. Make sure we do this later.
//if (pq.hasOption(QueryOption_CursorTailable) && pq.getNumToReturn() != 1) { ... }
// 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()));
}
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(runner.get(), cq->getParsed().getOptions(),
cq->getParsed().getFilter());
ccId = cc->cursorid();
log() << "caching runner with cursorid " << ccId << 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());
}
// 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;
// TODO: nscanned is bogus.
// curop.debug().nscanned = ( cursor ? cursor->nscanned() : 0LL );
curop.debug().ntoskip = pq.getSkip();
curop.debug().nreturned = 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() : "";
}
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) {
return Status(ErrorCodes::TypeMismatch,
str::stream()
<< "Expected \"" << kMemberStateFieldName
<< "\" field in response to replSetHeartbeat "
"command to have type NumberInt or NumberLong, but found type "
<< typeName(memberStateElement.type()));
} else {
long long stateInt = memberStateElement.numberLong();
if (stateInt < 0 || stateInt > MemberState::RS_MAX) {
return Status(ErrorCodes::BadValue,
str::stream()
<< "Value for \"" << kMemberStateFieldName
<< "\" in response to replSetHeartbeat is "
"out of range; legal values are non-negative and no more than "
<< MemberState::RS_MAX);
}
_stateSet = true;
_state = MemberState(static_cast<int>(stateInt));
}
_stateDisagreement = doc[kHasStateDisagreementFieldName].trueValue();
// Not required for the case of uninitialized members -- they have no config
const BSONElement versionElement = doc[kConfigVersionFieldName];
// If we have an optime then we must have a version
if (_opTimeSet && versionElement.eoo()) {
return Status(ErrorCodes::NoSuchKey,
str::stream() << "Response to replSetHeartbeat missing required \""
<< kConfigVersionFieldName
<< "\" field even though initialized");
}
// If there is a "v" (config version) then it must be an int.
if (!versionElement.eoo() && versionElement.type() != NumberInt) {
return Status(ErrorCodes::TypeMismatch,
str::stream() << "Expected \"" << kConfigVersionFieldName
<< "\" field in response to replSetHeartbeat to have "
"type NumberInt, but found "
<< typeName(versionElement.type()));
}
_version = versionElement.numberInt();
const BSONElement hbMsgElement = doc[kHbMessageFieldName];
if (hbMsgElement.eoo()) {
_hbmsg.clear();
} else if (hbMsgElement.type() != String) {
return Status(ErrorCodes::TypeMismatch,
str::stream() << "Expected \"" << kHbMessageFieldName
<< "\" field in response to replSetHeartbeat to have "
"type String, but found " << typeName(hbMsgElement.type()));
} else {
_hbmsg = hbMsgElement.String();
}
const BSONElement syncingToElement = doc[kSyncSourceFieldName];
if (syncingToElement.eoo()) {
_syncingTo.clear();
} else if (syncingToElement.type() != String) {
return Status(ErrorCodes::TypeMismatch,
str::stream() << "Expected \"" << kSyncSourceFieldName
<< "\" field in response to replSetHeartbeat to "
"have type String, but found "
<< typeName(syncingToElement.type()));
} else {
_syncingTo = syncingToElement.String();
}
const BSONElement rsConfigElement = doc[kConfigFieldName];
if (rsConfigElement.eoo()) {
_configSet = false;
_config = ReplicaSetConfig();
return Status::OK();
} else if (rsConfigElement.type() != Object) {
return Status(ErrorCodes::TypeMismatch,
str::stream() << "Expected \"" << kConfigFieldName
<< "\" in response to replSetHeartbeat to have type "
"Object, but found " << typeName(rsConfigElement.type()));
}
_configSet = true;
return _config.initialize(rsConfigElement.Obj());
}
/**
* 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);
// If the getmore had a time limit, remaining time is "rolled over" back to the
// cursor (for use by future getmore ops).
cc->setLeftoverMaxTimeMicros( curop.getRemainingMaxTimeMicros() );
}
QueryResult* qr = reinterpret_cast<QueryResult*>(bb.buf());
qr->len = bb.len();
qr->setOperation(opReply);
qr->_resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = startingResult;
qr->nReturned = numResults;
bb.decouple();
return qr;
}
/**
* 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;
}
}
if ( shardingState.getVersion( ns ) != 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( ns , "version changed during initial query", shardingVersionAtStart, shardingState.getVersion( ns ) );
}
int nReturned = queryResponseBuilder.handoff( result );
ccPointer.reset();
long long cursorid = 0;
if ( saveClientCursor ) {
// Create a new ClientCursor, with a default timeout.
ccPointer.reset( new ClientCursor( queryOptions, cursor, ns,
jsobj.getOwned() ) );
cursorid = ccPointer->cursorid();
DEV tlog(2) << "query has more, cursorid: " << cursorid << endl;
if ( cursor->supportYields() ) {
ClientCursor::YieldData data;
ccPointer->prepareToYield( data );
}
else {
ccPointer->c()->noteLocation();
}
// !!! Save the original message buffer, so it can be referenced in getMore.
ccPointer->originalMessage = m;
// Save slave's position in the oplog.
if ( pq.hasOption( QueryOption_OplogReplay ) && !slaveReadTill.isNull() ) {
ccPointer->slaveReadTill( slaveReadTill );
}
if ( !ccPointer->ok() && ccPointer->c()->tailable() ) {
DEV tlog() << "query has no more but tailable, cursorid: " << cursorid << endl;
}
if( queryOptions & QueryOption_Exhaust ) {
exhaust = ns;
curop.debug().exhaust = true;
}
// Set attributes for getMore.
ccPointer->setChunkManager( queryResponseBuilder.chunkManager() );
ccPointer->setPos( nReturned );
ccPointer->pq = pq_shared;
ccPointer->fields = pq.getFieldPtr();
ccPointer.release();
}
QueryResult *qr = (QueryResult *) result.header();
qr->cursorId = cursorid;
curop.debug().cursorid = ( cursorid == 0 ? -1 : qr->cursorId );
qr->setResultFlagsToOk();
// qr->len is updated automatically by appendData()
curop.debug().responseLength = qr->len;
qr->setOperation(opReply);
qr->startingFrom = 0;
qr->nReturned = nReturned;
int duration = curop.elapsedMillis();
bool dbprofile = curop.shouldDBProfile( duration );
if ( dbprofile || duration >= cmdLine.slowMS ) {
curop.debug().nscanned = (int)( cursor ? cursor->nscanned() : 0 );
curop.debug().ntoskip = pq.getSkip();
}
curop.debug().nreturned = nReturned;
return exhaust;
}
Status ReplSetHeartbeatResponse::initialize(const BSONObj& doc) {
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) {
_electionTime = 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 reponse 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();
}
_mismatch = doc[kMismatchFieldName].trueValue();
_isReplSet = doc[kIsReplSetFieldName].trueValue();
const BSONElement memberStateElement = doc[kMemberStateFieldName];
if (memberStateElement.eoo()) {
_stateSet = false;
}
else if (memberStateElement.type() != NumberInt &&
memberStateElement.type() != NumberLong) {
return Status(ErrorCodes::TypeMismatch, str::stream() << "Expected \"" <<
kMemberStateFieldName << "\" field in response to replSetHeartbeat "
" command to have type NumberInt or NumberLong, but found type " <<
typeName(memberStateElement.type()));
}
else {
long long stateInt = memberStateElement.numberLong();
if (stateInt < 0 || stateInt > MemberState::RS_MAX) {
return Status(ErrorCodes::BadValue, str::stream() << "Value for \"" <<
kMemberStateFieldName << "\" in response to replSetHeartbeat is "
" out of range; legal values are non-negative and no more than " <<
MemberState::RS_MAX);
}
_state = MemberState(static_cast<int>(stateInt));
}
_stateDisagreement = doc[kHasStateDisagreementFieldName].trueValue();
const BSONElement versionElement = doc[kConfigVersionFieldName];
if (versionElement.eoo()) {
return Status(ErrorCodes::NoSuchKey, str::stream() <<
"Response to replSetHeartbeat missing required \"" <<
kConfigVersionFieldName << " field");
}
if (versionElement.type() != NumberInt) {
return Status(ErrorCodes::TypeMismatch, str::stream() << "Expected \"" <<
kConfigVersionFieldName <<
"\" field in response to replSetHeartbeat to have "
"type NumberInt, but found " << typeName(versionElement.type()));
}
_version = versionElement.numberInt();
const BSONElement replSetNameElement = doc[kReplSetFieldName];
if (replSetNameElement.eoo()) {
return Status(ErrorCodes::NoSuchKey, str::stream() <<
"Response to replSetHeartbeat missing required \"" <<
kReplSetFieldName << "\" field");
}
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()));
}
_setName = replSetNameElement.String();
const BSONElement hbMsgElement = doc[kHbMessageFieldName];
if (hbMsgElement.eoo()) {
_hbmsg.clear();
}
else if (hbMsgElement.type() != String) {
return Status(ErrorCodes::TypeMismatch, str::stream() << "Expected \"" <<
kHbMessageFieldName << "\" field in response to replSetHeartbeat to have "
"type String, but found " << typeName(hbMsgElement.type()));
}
_hbmsg = hbMsgElement.String();
const BSONElement syncingToElement = doc[kSyncSourceFieldName];
if (syncingToElement.eoo()) {
_syncingTo.clear();
}
else if (syncingToElement.type() != String) {
return Status(ErrorCodes::TypeMismatch, str::stream() << "Expected \"" <<
kSyncSourceFieldName << "\" field in response to replSetHeartbeat to "
"have type String, but found " << typeName(syncingToElement.type()));
}
_syncingTo = syncingToElement.String();
const BSONElement rsConfigElement = doc[kConfigFieldName];
if (rsConfigElement.eoo()) {
_configSet = false;
_config = ReplicaSetConfig();
}
else if (rsConfigElement.type() != Object) {
return Status(ErrorCodes::TypeMismatch, str::stream() << "Expected \"" <<
kConfigFieldName << "\" in response to replSetHeartbeat to have type "
"Object, but found " << typeName(rsConfigElement.type()));
}
_configSet = true;
return _config.initialize(rsConfigElement.Obj());
}
