static Cursor maybeLexGlobalValue(Cursor C, MIToken &Token) {
if (C.peek() != '@')
return None;
auto Range = C;
C.advance(); // Skip the '@'
// TODO: add support for quoted names.
if (!isdigit(C.peek())) {
while (isIdentifierChar(C.peek()))
C.advance();
Token = MIToken(MIToken::NamedGlobalValue, Range.upto(C),
/*StringOffset=*/1); // Drop the '@'
return C;
}
auto NumberRange = C;
while (isdigit(C.peek()))
C.advance();
Token =
MIToken(MIToken::GlobalValue, Range.upto(C), APSInt(NumberRange.upto(C)));
return C;
}
static Cursor maybeLexHexadecimalLiteral(Cursor C, MIToken &Token) {
if (C.peek() != '0' || (C.peek(1) != 'x' && C.peek(1) != 'X'))
return None;
Cursor Range = C;
C.advance(2);
unsigned PrefLen = 2;
if (isValidHexFloatingPointPrefix(C.peek())) {
C.advance();
PrefLen++;
}
while (isxdigit(C.peek()))
C.advance();
StringRef StrVal = Range.upto(C);
if (StrVal.size() <= PrefLen)
return None;
if (PrefLen == 2)
Token.reset(MIToken::HexLiteral, Range.upto(C));
else // It must be 3, which means that there was a floating-point prefix.
Token.reset(MIToken::FloatingPointLiteral, Range.upto(C));
return C;
}
static Cursor lexName(Cursor C, MIToken &Token, MIToken::TokenKind Type,
unsigned PrefixLength, ErrorCallbackType ErrorCallback) {
auto Range = C;
C.advance(PrefixLength);
if (C.peek() == '"') {
if (Cursor R = lexStringConstant(C, ErrorCallback)) {
StringRef String = Range.upto(R);
Token.reset(Type, String)
.setOwnedStringValue(
unescapeQuotedString(String.drop_front(PrefixLength)));
return R;
}
Token.reset(MIToken::Error, Range.remaining());
return Range;
}
while (isIdentifierChar(C.peek()))
C.advance();
Token.reset(Type, Range.upto(C))
.setStringValue(Range.upto(C).drop_front(PrefixLength));
return C;
}
static Cursor maybeLexEscapedIRValue(Cursor C, MIToken &Token,
ErrorCallbackType ErrorCallback) {
if (C.peek() != '`')
return None;
auto Range = C;
C.advance();
auto StrRange = C;
while (C.peek() != '`') {
if (C.isEOF() || isNewlineChar(C.peek())) {
ErrorCallback(
C.location(),
"end of machine instruction reached before the closing '`'");
Token.reset(MIToken::Error, Range.remaining());
return C;
}
C.advance();
}
StringRef Value = StrRange.upto(C);
C.advance();
Token.reset(MIToken::QuotedIRValue, Range.upto(C)).setStringValue(Value);
return C;
}
static Cursor maybeLexIndexAndName(Cursor C, MIToken &Token, StringRef Rule,
MIToken::TokenKind Kind) {
if (!C.remaining().startswith(Rule) || !isdigit(C.peek(Rule.size())))
return None;
auto Range = C;
C.advance(Rule.size());
auto NumberRange = C;
while (isdigit(C.peek()))
C.advance();
StringRef Number = NumberRange.upto(C);
unsigned StringOffset = Rule.size() + Number.size();
if (C.peek() == '.') {
C.advance();
++StringOffset;
while (isIdentifierChar(C.peek()))
C.advance();
}
Token.reset(Kind, Range.upto(C))
.setIntegerValue(APSInt(Number))
.setStringValue(Range.upto(C).drop_front(StringOffset));
return C;
}
static Cursor maybeLexSymbol(Cursor C, MIToken &Token) {
MIToken::TokenKind Kind;
unsigned Length = 1;
if (C.peek() == ':' && C.peek(1) == ':') {
Kind = MIToken::coloncolon;
Length = 2;
} else
Kind = symbolToken(C.peek());
if (Kind == MIToken::Error)
return None;
auto Range = C;
C.advance(Length);
Token.reset(Kind, Range.upto(C));
return C;
}
/* must call this on a delete so we clean up the cursors. */
void ClientCursor::aboutToDelete(const DiskLoc& dl) {
recursive_scoped_lock lock(ccmutex);
CCByLoc::iterator j = byLoc.lower_bound(dl);
CCByLoc::iterator stop = byLoc.upper_bound(dl);
if ( j == stop )
return;
vector<ClientCursor*> toAdvance;
while ( 1 ) {
toAdvance.push_back(j->second);
DEV assert( j->first == dl );
++j;
if ( j == stop )
break;
}
wassert( toAdvance.size() < 5000 );
for ( vector<ClientCursor*>::iterator i = toAdvance.begin(); i != toAdvance.end(); ++i ){
ClientCursor* cc = *i;
if ( cc->_doingDeletes ) continue;
Cursor *c = cc->c.get();
if ( c->capped() ){
delete cc;
continue;
}
c->checkLocation();
DiskLoc tmp1 = c->refLoc();
if ( tmp1 != dl ) {
/* this might indicate a failure to call ClientCursor::updateLocation() */
problem() << "warning: cursor loc " << tmp1 << " does not match byLoc position " << dl << " !" << endl;
}
c->advance();
if ( c->eof() ) {
// advanced to end -- delete cursor
delete cc;
}
else {
wassert( c->refLoc() != dl );
cc->updateLocation();
}
}
}
/* must call this on a delete so we clean up the cursors. */
void ClientCursor::aboutToDelete(const DiskLoc& dl) {
NoPageFaultsAllowed npfa;
recursive_scoped_lock lock(ccmutex);
Database *db = cc().database();
verify(db);
aboutToDeleteForSharding( db , dl );
CCByLoc& bl = db->ccByLoc;
CCByLoc::iterator j = bl.lower_bound(ByLocKey::min(dl));
CCByLoc::iterator stop = bl.upper_bound(ByLocKey::max(dl));
if ( j == stop )
return;
vector<ClientCursor*> toAdvance;
while ( 1 ) {
toAdvance.push_back(j->second);
DEV verify( j->first.loc == dl );
++j;
if ( j == stop )
break;
}
if( toAdvance.size() >= 3000 ) {
log() << "perf warning MPW101: " << toAdvance.size() << " cursors for one diskloc "
<< dl.toString()
<< ' ' << toAdvance[1000]->_ns
<< ' ' << toAdvance[2000]->_ns
<< ' ' << toAdvance[1000]->_pinValue
<< ' ' << toAdvance[2000]->_pinValue
<< ' ' << toAdvance[1000]->_pos
<< ' ' << toAdvance[2000]->_pos
<< ' ' << toAdvance[1000]->_idleAgeMillis
<< ' ' << toAdvance[2000]->_idleAgeMillis
<< ' ' << toAdvance[1000]->_doingDeletes
<< ' ' << toAdvance[2000]->_doingDeletes
<< endl;
//wassert( toAdvance.size() < 5000 );
}
for ( vector<ClientCursor*>::iterator i = toAdvance.begin(); i != toAdvance.end(); ++i ) {
ClientCursor* cc = *i;
wassert(cc->_db == db);
if ( cc->_doingDeletes ) continue;
Cursor *c = cc->_c.get();
if ( c->capped() ) {
/* note we cannot advance here. if this condition occurs, writes to the oplog
have "caught" the reader. skipping ahead, the reader would miss postentially
important data.
*/
delete cc;
continue;
}
c->recoverFromYield();
DiskLoc tmp1 = c->refLoc();
if ( tmp1 != dl ) {
// This might indicate a failure to call ClientCursor::prepareToYield() but it can
// also happen during correct operation, see SERVER-2009.
problem() << "warning: cursor loc " << tmp1 << " does not match byLoc position " << dl << " !" << endl;
}
else {
c->advance();
}
while (!c->eof() && c->refLoc() == dl) {
/* We don't delete at EOF because we want to return "no more results" rather than "no such cursor".
* The loop is to handle MultiKey indexes where the deleted record is pointed to by multiple adjacent keys.
* In that case we need to advance until we get to the next distinct record or EOF.
* SERVER-4154
* SERVER-5198
* But see SERVER-5725.
*/
c->advance();
}
cc->updateLocation();
}
}
/// Skip the leading whitespace characters and return the updated cursor.
static Cursor skipWhitespace(Cursor C) {
while (isspace(C.peek()))
C.advance();
return C;
}
QueryResult* processGetMore(const char* ns,
int ntoreturn,
long long cursorid,
CurOp& curop,
int pass,
bool& exhaust,
bool* isCursorAuthorized ) {
bool hasRunner = false;
// Scoped to kill the pin after seeing if the runner's there.
{
// See if there's a runner. We do this until agg. is behind a Runner instead of a CC.
ClientCursorPin p(cursorid);
ClientCursor *cc = p.c();
if (NULL != cc && NULL != cc->getRunner()) {
hasRunner = true;
}
}
if (hasRunner) {
return newGetMore(ns, ntoreturn, cursorid, curop, pass, exhaust,
isCursorAuthorized);
}
exhaust = false;
int bufSize = 512 + sizeof( QueryResult ) + MaxBytesToReturnToClientAtOnce;
BufBuilder b( bufSize );
b.skip(sizeof(QueryResult));
int resultFlags = ResultFlag_AwaitCapable;
int start = 0;
int n = 0;
scoped_ptr<Client::ReadContext> ctx(new Client::ReadContext(ns));
// call this readlocked so state can't change
replVerifyReadsOk();
ClientCursorPin p(cursorid);
ClientCursor *cc = p.c();
if ( unlikely(!cc) ) {
LOGSOME << "getMore: cursorid not found " << ns << " " << cursorid << endl;
cursorid = 0;
resultFlags = ResultFlag_CursorNotFound;
}
else {
// Some internal users create a ClientCursor with a Runner. Don't crash if this
// happens. Instead, hand them off to the new framework.
if (NULL != cc->getRunner()) {
p.release();
return newGetMore(ns, ntoreturn, cursorid, curop, pass, exhaust, isCursorAuthorized);
}
// check for spoofing of the ns such that it does not match the one originally there for the cursor
uassert(14833, "auth error", str::equals(ns, cc->ns().c_str()));
*isCursorAuthorized = true;
// This must be done after auth check to ensure proper cleanup.
uassert(16951, "failing getmore due to set failpoint",
!MONGO_FAIL_POINT(getMoreError));
// 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.
if ( pass == 0 )
cc->updateSlaveLocation( curop );
int queryOptions = cc->queryOptions();
curop.debug().query = cc->query();
curop.setQuery( cc->query() );
start = cc->pos();
Cursor *c = cc->c();
if (!c->requiresLock()) {
// make sure it won't be destroyed under us
fassert(16952, !c->shouldDestroyOnNSDeletion());
fassert(16953, !c->supportYields());
ctx.reset(); // unlocks
}
c->recoverFromYield();
DiskLoc last;
// This metadata may be stale, but it's the state of chunking when the cursor was
// created.
CollectionMetadataPtr metadata = cc->getCollMetadata();
KeyPattern keyPattern( metadata ? metadata->getKeyPattern() : BSONObj() );
while ( 1 ) {
if ( !c->ok() ) {
if ( c->tailable() ) {
// when a tailable cursor hits "EOF", ok() goes false, and current() is
// null. however advance() can still be retries as a reactivation attempt.
// when there is new data, it will return true. that's what we are doing
// here.
if ( c->advance() )
continue;
if( n == 0 && (queryOptions & QueryOption_AwaitData) && pass < 1000 ) {
return 0;
}
break;
}
p.release();
bool ok = ClientCursor::erase(cursorid);
verify(ok);
cursorid = 0;
cc = 0;
break;
}
MatchDetails details;
if ( cc->fields && cc->fields->getArrayOpType() == Projection::ARRAY_OP_POSITIONAL ) {
// field projection specified, and contains an array operator
details.requestElemMatchKey();
}
// in some cases (clone collection) there won't be a matcher
if ( !c->currentMatches( &details ) ) {
}
else if ( metadata && !metadata->keyBelongsToMe( extractKey(c, keyPattern ) ) ) {
LOG(2) << "cursor skipping document in un-owned chunk: " << c->current()
<< endl;
}
else {
if( c->getsetdup(c->currLoc()) ) {
//out() << " but it's a dup \n";
}
else {
last = c->currLoc();
n++;
// Fill out the fields requested by the query.
const Projection::KeyOnly *keyFieldsOnly = c->keyFieldsOnly();
if ( keyFieldsOnly ) {
fillQueryResultFromObj( b, 0, keyFieldsOnly->hydrate(
c->currKey() ), &details );
}
else {
DiskLoc loc = c->currLoc();
fillQueryResultFromObj( b, cc->fields.get(), c->current(), &details,
( ( cc->pq.get() && cc->pq->showDiskLoc() ) ? &loc : 0 ) );
}
if ( ( ntoreturn && n >= ntoreturn ) || b.len() > MaxBytesToReturnToClientAtOnce ) {
c->advance();
cc->incPos( n );
break;
}
}
}
c->advance();
if ( ! cc->yieldSometimes( ( c->ok() && c->keyFieldsOnly() ) ?
ClientCursor::DontNeed : ClientCursor::WillNeed ) ) {
ClientCursor::erase(cursorid);
cursorid = 0;
cc = 0;
break;
}
}
if ( cc ) {
if ( c->supportYields() ) {
ClientCursor::YieldData data;
verify( cc->prepareToYield( data ) );
}
else {
cc->c()->noteLocation();
}
cc->storeOpForSlave( last );
exhaust = cc->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 = (QueryResult *) b.buf();
qr->len = b.len();
qr->setOperation(opReply);
qr->_resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = start;
qr->nReturned = n;
b.decouple();
return qr;
}
/* must call this on a delete so we clean up the cursors. */
void ClientCursor::aboutToDelete(const DiskLoc& dl) {
recursive_scoped_lock lock(ccmutex);
Database *db = cc().database();
assert(db);
aboutToDeleteForSharding( db , dl );
CCByLoc& bl = db->ccByLoc;
CCByLoc::iterator j = bl.lower_bound(ByLocKey::min(dl));
CCByLoc::iterator stop = bl.upper_bound(ByLocKey::max(dl));
if ( j == stop )
return;
vector<ClientCursor*> toAdvance;
while ( 1 ) {
toAdvance.push_back(j->second);
DEV assert( j->first.loc == dl );
++j;
if ( j == stop )
break;
}
if( toAdvance.size() >= 3000 ) {
log() << "perf warning MPW101: " << toAdvance.size() << " cursors for one diskloc "
<< dl.toString()
<< ' ' << toAdvance[1000]->_ns
<< ' ' << toAdvance[2000]->_ns
<< ' ' << toAdvance[1000]->_pinValue
<< ' ' << toAdvance[2000]->_pinValue
<< ' ' << toAdvance[1000]->_pos
<< ' ' << toAdvance[2000]->_pos
<< ' ' << toAdvance[1000]->_idleAgeMillis
<< ' ' << toAdvance[2000]->_idleAgeMillis
<< ' ' << toAdvance[1000]->_doingDeletes
<< ' ' << toAdvance[2000]->_doingDeletes
<< endl;
//wassert( toAdvance.size() < 5000 );
}
for ( vector<ClientCursor*>::iterator i = toAdvance.begin(); i != toAdvance.end(); ++i ) {
ClientCursor* cc = *i;
wassert(cc->_db == db);
if ( cc->_doingDeletes ) continue;
Cursor *c = cc->_c.get();
if ( c->capped() ) {
/* note we cannot advance here. if this condition occurs, writes to the oplog
have "caught" the reader. skipping ahead, the reader would miss postentially
important data.
*/
delete cc;
continue;
}
c->checkLocation();
DiskLoc tmp1 = c->refLoc();
if ( tmp1 != dl ) {
// This might indicate a failure to call ClientCursor::updateLocation() but it can
// also happen during correct operation, see SERVER-2009.
problem() << "warning: cursor loc " << tmp1 << " does not match byLoc position " << dl << " !" << endl;
}
else {
c->advance();
}
if ( c->eof() ) {
// advanced to end
// leave ClientCursor in place so next getMore doesn't fail
// still need to mark new location though
cc->updateLocation();
}
else {
wassert( c->refLoc() != dl );
cc->updateLocation();
}
}
}
/* must call this on a delete so we clean up the cursors. */
void ClientCursor::aboutToDelete(const StringData& ns,
const NamespaceDetails* nsd,
const DiskLoc& dl) {
// Begin cursor-only
NoPageFaultsAllowed npfa;
// End cursor-only
recursive_scoped_lock lock(ccmutex);
Database *db = cc().database();
verify(db);
aboutToDeleteForSharding( ns, db, nsd, dl );
// Check our non-cached active runner list.
for (set<Runner*>::iterator it = nonCachedRunners.begin(); it != nonCachedRunners.end();
++it) {
Runner* runner = *it;
if (0 == ns.compare(runner->ns())) {
runner->invalidate(dl);
}
}
// TODO: This requires optimization. We walk through *all* CCs and send the delete to every
// CC open on the db we're deleting from. We could:
// 1. Map from ns to open runners,
// 2. Map from ns -> (a map of DiskLoc -> runners who care about that DL)
//
// We could also queue invalidations somehow and have them processed later in the runner's
// read locks.
for (CCById::const_iterator it = clientCursorsById.begin(); it != clientCursorsById.end();
++it) {
ClientCursor* cc = it->second;
// We're only interested in cursors over one db.
if (cc->_db != db) { continue; }
if (NULL == cc->_runner.get()) { continue; }
cc->_runner->invalidate(dl);
}
// Begin cursor-only. Only cursors that are in ccByLoc are processed here.
CCByLoc& bl = db->ccByLoc();
CCByLoc::iterator j = bl.lower_bound(ByLocKey::min(dl));
CCByLoc::iterator stop = bl.upper_bound(ByLocKey::max(dl));
if ( j == stop )
return;
vector<ClientCursor*> toAdvance;
while ( 1 ) {
toAdvance.push_back(j->second);
DEV verify( j->first.loc == dl );
++j;
if ( j == stop )
break;
}
if( toAdvance.size() >= 3000 ) {
log() << "perf warning MPW101: " << toAdvance.size() << " cursors for one diskloc "
<< dl.toString()
<< ' ' << toAdvance[1000]->_ns
<< ' ' << toAdvance[2000]->_ns
<< ' ' << toAdvance[1000]->_pinValue
<< ' ' << toAdvance[2000]->_pinValue
<< ' ' << toAdvance[1000]->_pos
<< ' ' << toAdvance[2000]->_pos
<< ' ' << toAdvance[1000]->_idleAgeMillis
<< ' ' << toAdvance[2000]->_idleAgeMillis
<< ' ' << toAdvance[1000]->_doingDeletes
<< ' ' << toAdvance[2000]->_doingDeletes
<< endl;
//wassert( toAdvance.size() < 5000 );
}
for ( vector<ClientCursor*>::iterator i = toAdvance.begin(); i != toAdvance.end(); ++i ) {
ClientCursor* cc = *i;
wassert(cc->_db == db);
if ( cc->_doingDeletes ) continue;
Cursor *c = cc->_c.get();
if ( c->capped() ) {
/* note we cannot advance here. if this condition occurs, writes to the oplog
have "caught" the reader. skipping ahead, the reader would miss postentially
important data.
*/
delete cc;
continue;
}
c->recoverFromYield();
DiskLoc tmp1 = c->refLoc();
if ( tmp1 != dl ) {
// This might indicate a failure to call ClientCursor::prepareToYield() but it can
// also happen during correct operation, see SERVER-2009.
problem() << "warning: cursor loc " << tmp1 << " does not match byLoc position " << dl << " !" << endl;
}
else {
c->advance();
}
while (!c->eof() && c->refLoc() == dl) {
/* We don't delete at EOF because we want to return "no more results" rather than "no such cursor".
* The loop is to handle MultiKey indexes where the deleted record is pointed to by multiple adjacent keys.
* In that case we need to advance until we get to the next distinct record or EOF.
* SERVER-4154
* SERVER-5198
* But see SERVER-5725.
*/
c->advance();
}
cc->updateLocation();
}
// End cursor-only
}
QueryResult* processGetMore(const char* ns,
int ntoreturn,
long long cursorid,
CurOp& curop,
int pass,
bool& exhaust,
bool* isCursorAuthorized ) {
exhaust = false;
ClientCursor::Pin p(cursorid);
ClientCursor *client_cursor = p.c();
int bufSize = 512 + sizeof( QueryResult ) + MaxBytesToReturnToClientAtOnce;
BufBuilder b( bufSize );
b.skip(sizeof(QueryResult));
int resultFlags = ResultFlag_AwaitCapable;
int start = 0;
int n = 0;
if ( unlikely(!client_cursor) ) {
LOGSOME << "getMore: cursorid not found " << ns << " " << cursorid << endl;
cursorid = 0;
resultFlags = ResultFlag_CursorNotFound;
}
else {
// check for spoofing of the ns such that it does not match the one originally there for the cursor
uassert(14833, "auth error", str::equals(ns, client_cursor->ns().c_str()));
uassert(16784, "oplog cursor reading data that is too old", !client_cursor->lastOpForSlaveTooOld());
int queryOptions = client_cursor->queryOptions();
OpSettings settings;
settings.setBulkFetch(true);
settings.setQueryCursorMode(DEFAULT_LOCK_CURSOR);
settings.setCappedAppendPK(queryOptions & QueryOption_AddHiddenPK);
cc().setOpSettings(settings);
// Check if the cursor is part of a multi-statement transaction. If it is
// and this is not the right client (meaning the current transaction stack
// does not match that in the cursor), it will uassert. If the cursor is
// not part of a multi-statement transaction, then we need to use the stack
// in the cursor for this scope.
const bool cursorPartOfMultiStatementTxn = client_cursor->checkMultiStatementTxn();
scoped_ptr<Client::WithTxnStack> wts;
if (!cursorPartOfMultiStatementTxn) {
// For simplicity, prevent multi-statement transactions from
// reading cursors it didn't create.
uassert(16813, "Cannot getMore() on a cursor not created by this multi-statement transaction",
!cc().hasTxn());
wts.reset(new Client::WithTxnStack(client_cursor->transactions));
}
*isCursorAuthorized = true;
if (pass == 0) {
client_cursor->updateSlaveLocation( curop );
}
curop.debug().query = client_cursor->query();
start = client_cursor->pos();
Cursor *c = client_cursor->c();
// This manager may be stale, but it's the state of chunking when the cursor was created.
ShardChunkManagerPtr manager = client_cursor->getChunkManager();
while ( 1 ) {
if ( !c->ok() ) {
if ( c->tailable() ) {
/* when a tailable cursor hits "EOF", ok() goes false, and current() is null. however
advance() can still be retries as a reactivation attempt. when there is new data, it will
return true. that's what we are doing here.
*/
if ( c->advance() )
continue;
if( n == 0 && (queryOptions & QueryOption_AwaitData) && pass < 1000 ) {
return 0;
}
break;
}
p.release();
// Done with this cursor, steal transaction stack back to commit or abort it here.
bool ok = ClientCursor::erase(cursorid);
verify(ok);
cursorid = 0;
client_cursor = 0;
break;
}
MatchDetails details;
if ( client_cursor->fields && client_cursor->fields->getArrayOpType() == Projection::ARRAY_OP_POSITIONAL ) {
// field projection specified, and contains an array operator
details.requestElemMatchKey();
}
// in some cases (clone collection) there won't be a matcher
if ( !c->currentMatches( &details ) ) {
}
else if ( manager && ! manager->belongsToMe( client_cursor ) ){
LOG(2) << "cursor skipping document in un-owned chunk: " << c->current() << endl;
}
else {
if( c->getsetdup(c->currPK()) ) {
//out() << " but it's a dup \n";
}
else {
// save this so that at the end of the loop,
// we can update the location for write concern
// in replication. Note that if this cursor is not
// doing replication, this is pointless
if ( client_cursor->queryOptions() & QueryOption_OplogReplay ) {
client_cursor->storeOpForSlave( c->current() );
}
n++;
client_cursor->fillQueryResultFromObj( b, &details );
if ( ( ntoreturn && n >= ntoreturn ) || b.len() > MaxBytesToReturnToClientAtOnce ) {
c->advance();
client_cursor->incPos( n );
break;
}
}
}
c->advance();
}
if ( client_cursor ) {
client_cursor->resetIdleAge();
exhaust = client_cursor->queryOptions() & QueryOption_Exhaust;
} else if (!cursorPartOfMultiStatementTxn) {
// This cursor is done and it wasn't part of a multi-statement
// transaction. We can commit the transaction now.
cc().commitTopTxn();
wts->release();
}
}
QueryResult *qr = (QueryResult *) b.buf();
qr->len = b.len();
qr->setOperation(opReply);
qr->_resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = start;
qr->nReturned = n;
b.decouple();
return qr;
}
QueryResult* getMore(const char *ns, int ntoreturn, long long cursorid , stringstream& ss) {
ClientCursor *cc = ClientCursor::find(cursorid);
int bufSize = 512;
if ( cc ){
bufSize += sizeof( QueryResult );
bufSize += ( ntoreturn ? 4 : 1 ) * 1024 * 1024;
}
BufBuilder b( bufSize );
b.skip(sizeof(QueryResult));
int resultFlags = 0;
int start = 0;
int n = 0;
if ( !cc ) {
log() << "getMore: cursorid not found " << ns << " " << cursorid << endl;
cursorid = 0;
resultFlags = QueryResult::ResultFlag_CursorNotFound;
}
else {
ss << " query: " << cc->query << " ";
start = cc->pos;
Cursor *c = cc->c.get();
c->checkLocation();
while ( 1 ) {
if ( !c->ok() ) {
if ( c->tailable() ) {
if ( c->advance() ) {
continue;
}
break;
}
bool ok = ClientCursor::erase(cursorid);
assert(ok);
cursorid = 0;
cc = 0;
break;
}
if ( !cc->matcher->matches(c->currKey(), c->currLoc() ) ) {
}
else {
//out() << "matches " << c->currLoc().toString() << '\n';
if( c->getsetdup(c->currLoc()) ) {
//out() << " but it's a dup \n";
}
else {
BSONObj js = c->current();
fillQueryResultFromObj(b, cc->filter.get(), js);
n++;
if ( (ntoreturn>0 && (n >= ntoreturn || b.len() > MaxBytesToReturnToClientAtOnce)) ||
(ntoreturn==0 && b.len()>1*1024*1024) ) {
c->advance();
cc->pos += n;
//cc->updateLocation();
break;
}
}
}
c->advance();
}
if ( cc ) {
cc->updateLocation();
cc->mayUpgradeStorage();
}
}
QueryResult *qr = (QueryResult *) b.buf();
qr->len = b.len();
qr->setOperation(opReply);
qr->resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = start;
qr->nReturned = n;
b.decouple();
return qr;
}
QueryResult* processGetMore(const char *ns, int ntoreturn, long long cursorid , CurOp& curop, int pass, bool& exhaust ) {
exhaust = false;
ClientCursor::Pointer p(cursorid);
ClientCursor *cc = p.c();
int bufSize = 512 + sizeof( QueryResult ) + MaxBytesToReturnToClientAtOnce;
BufBuilder b( bufSize );
b.skip(sizeof(QueryResult));
int resultFlags = ResultFlag_AwaitCapable;
int start = 0;
int n = 0;
if ( unlikely(!cc) ) {
LOGSOME << "getMore: cursorid not found " << ns << " " << cursorid << endl;
cursorid = 0;
resultFlags = ResultFlag_CursorNotFound;
}
else {
// check for spoofing of the ns such that it does not match the one originally there for the cursor
uassert(14833, "auth error", str::equals(ns, cc->ns().c_str()));
if ( pass == 0 )
cc->updateSlaveLocation( curop );
int queryOptions = cc->queryOptions();
curop.debug().query = cc->query();
start = cc->pos();
Cursor *c = cc->c();
c->recoverFromYield();
DiskLoc last;
scoped_ptr<Projection::KeyOnly> keyFieldsOnly;
if ( cc->modifiedKeys() == false && cc->isMultiKey() == false && cc->fields )
keyFieldsOnly.reset( cc->fields->checkKey( cc->indexKeyPattern() ) );
// This manager may be stale, but it's the state of chunking when the cursor was created.
ShardChunkManagerPtr manager = cc->getChunkManager();
while ( 1 ) {
if ( !c->ok() ) {
if ( c->tailable() ) {
/* when a tailable cursor hits "EOF", ok() goes false, and current() is null. however
advance() can still be retries as a reactivation attempt. when there is new data, it will
return true. that's what we are doing here.
*/
if ( c->advance() )
continue;
if( n == 0 && (queryOptions & QueryOption_AwaitData) && pass < 1000 ) {
return 0;
}
break;
}
p.release();
bool ok = ClientCursor::erase(cursorid);
verify(ok);
cursorid = 0;
cc = 0;
break;
}
// in some cases (clone collection) there won't be a matcher
if ( !c->currentMatches() ) {
}
else if ( manager && ! manager->belongsToMe( cc ) ){
LOG(2) << "cursor skipping document in un-owned chunk: " << c->current() << endl;
}
else {
if( c->getsetdup(c->currLoc()) ) {
//out() << " but it's a dup \n";
}
else {
last = c->currLoc();
n++;
if ( keyFieldsOnly ) {
fillQueryResultFromObj(b, 0, keyFieldsOnly->hydrate( c->currKey() ) );
}
else {
BSONObj js = c->current();
// show disk loc should be part of the main query, not in an $or clause, so this should be ok
fillQueryResultFromObj(b, cc->fields.get(), js, ( cc->pq.get() && cc->pq->showDiskLoc() ? &last : 0));
}
if ( ( ntoreturn && n >= ntoreturn ) || b.len() > MaxBytesToReturnToClientAtOnce ) {
c->advance();
cc->incPos( n );
break;
}
}
}
c->advance();
if ( ! cc->yieldSometimes( ClientCursor::MaybeCovered ) ) {
ClientCursor::erase(cursorid);
cursorid = 0;
cc = 0;
p.deleted();
break;
}
}
if ( cc ) {
if ( c->supportYields() ) {
ClientCursor::YieldData data;
verify( cc->prepareToYield( data ) );
}
else {
cc->c()->noteLocation();
}
cc->mayUpgradeStorage();
cc->storeOpForSlave( last );
exhaust = cc->queryOptions() & QueryOption_Exhaust;
}
}
QueryResult *qr = (QueryResult *) b.buf();
qr->len = b.len();
qr->setOperation(opReply);
qr->_resultFlags() = resultFlags;
qr->cursorId = cursorid;
qr->startingFrom = start;
qr->nReturned = n;
b.decouple();
return qr;
}
