void DBClientReplicaSet::checkResponse( const char* data, int nReturned, bool* retry, string* targetHost ){
// For now, do exactly as we did before, so as not to break things. In general though, we
// should fix this so checkResponse has a more consistent contract.
if( ! retry ){
if( _lazyState._lastClient )
return _lazyState._lastClient->checkResponse( data, nReturned );
else
return checkMaster()->checkResponse( data, nReturned );
}
*retry = false;
if( targetHost && _lazyState._lastClient ) *targetHost = _lazyState._lastClient->getServerAddress();
else *targetHost = "";
if( ! _lazyState._lastClient ) return;
if( nReturned != 1 && nReturned != -1 ) return;
BSONObj dataObj;
if( nReturned == 1 ) dataObj = BSONObj( data );
// Check if we should retry here
if( _lazyState._lastOp == dbQuery && _lazyState._slaveOk ){
// Check the error code for a slave not secondary error
if( nReturned == -1 ||
( hasErrField( dataObj ) && ! dataObj["code"].eoo() && dataObj["code"].Int() == 13436 ) ){
bool wasMaster = false;
if( _lazyState._lastClient == _slave.get() ){
isntSecondary();
}
else if( _lazyState._lastClient == _master.get() ){
wasMaster = true;
isntMaster();
}
else
warning() << "passed " << dataObj << " but last rs client " << _lazyState._lastClient->toString() << " is not master or secondary" << endl;
if( _lazyState._retries < 3 ){
_lazyState._retries++;
*retry = true;
}
else{
// assert( wasMaster );
// printStackTrace();
log() << "too many retries (" << _lazyState._retries << "), could not get data from replica set" << endl;
}
}
}
}
bool DBClientReplicaSet::call(Message &toSend,
Message &response,
bool assertOk,
string * actualServer) {
const char * ns = 0;
if (toSend.operation() == dbQuery) {
// TODO: might be possible to do this faster by changing api
DbMessage dm(toSend);
QueryMessage qm(dm);
ns = qm.ns;
shared_ptr<ReadPreferenceSetting> readPref( _extractReadPref( qm.query,
qm.queryOptions ) );
if ( _isSecondaryQuery( ns, qm.query, *readPref ) ) {
LOG( 3 ) << "dbclient_rs call using secondary or tagged node selection in "
<< _getMonitor()->getName() << ", read pref is "
<< readPref->toBSON() << " (primary : "
<< ( _master.get() != NULL ?
_master->getServerAddress() : "[not cached]" )
<< ", lastTagged : "
<< ( _lastSlaveOkConn.get() != NULL ?
_lastSlaveOkConn->getServerAddress() : "[not cached]" )
<< ")" << endl;
for (size_t retry = 0; retry < MAX_RETRY; retry++) {
try {
DBClientConnection* conn = selectNodeUsingTags(readPref);
if (conn == NULL) {
return false;
}
if (actualServer != NULL) {
*actualServer = conn->getServerAddress();
}
return conn->call(toSend, response, assertOk);
}
catch ( const DBException& dbExcep ) {
LOG(1) << "can't call replica set node " << _lastSlaveOkHost << ": "
<< causedBy( dbExcep ) << endl;
if ( actualServer ) *actualServer = "";
invalidateLastSlaveOkCache();
}
}
// Was not able to successfully send after max retries
return false;
}
}
LOG( 3 ) << "dbclient_rs call to primary node in " << _getMonitor()->getName() << endl;
DBClientConnection* m = checkMaster();
if ( actualServer )
*actualServer = m->getServerAddress();
if ( ! m->call( toSend , response , assertOk ) )
return false;
if ( ns ) {
QueryResult * res = (QueryResult*)response.singleData();
if ( res->nReturned == 1 ) {
BSONObj x(res->data() );
if ( str::contains( ns , "$cmd" ) ) {
if ( isNotMasterErrorString( x["errmsg"] ) )
isntMaster();
}
else {
if ( isNotMasterErrorString( getErrField( x ) ) )
isntMaster();
}
}
}
return true;
}
void DBClientReplicaSet::checkResponse( const char* data, int nReturned, bool* retry, string* targetHost ){
// For now, do exactly as we did before, so as not to break things. In general though, we
// should fix this so checkResponse has a more consistent contract.
if( ! retry ){
if( _lazyState._lastClient )
return _lazyState._lastClient->checkResponse( data, nReturned );
else
return checkMaster()->checkResponse( data, nReturned );
}
*retry = false;
if( targetHost && _lazyState._lastClient ) *targetHost = _lazyState._lastClient->getServerAddress();
else if (targetHost) *targetHost = "";
if( ! _lazyState._lastClient ) return;
// nReturned == 1 means that we got one result back, which might be an error
// nReturned == -1 is a sentinel value for "no data returned" aka (usually) network problem
// If neither, this must be a query result so our response is ok wrt the replica set
if( nReturned != 1 && nReturned != -1 ) return;
BSONObj dataObj;
if( nReturned == 1 ) dataObj = BSONObj( data );
// Check if we should retry here
if( _lazyState._lastOp == dbQuery && _lazyState._secondaryQueryOk ){
// query could potentially go to a secondary, so see if this is an error (or empty) and
// retry if we're not past our retry limit.
if( nReturned == -1 /* no result, maybe network problem */ ||
( hasErrField( dataObj ) && ! dataObj["code"].eoo()
&& dataObj["code"].Int() == NotMasterOrSecondaryCode ) ){
if( _lazyState._lastClient == _lastSlaveOkConn.get() ){
isntSecondary();
}
else if( _lazyState._lastClient == _master.get() ){
isntMaster();
}
else {
warning() << "passed " << dataObj << " but last rs client "
<< _lazyState._lastClient->toString() << " is not master or secondary"
<< endl;
}
if ( _lazyState._retries < static_cast<int>( MAX_RETRY ) ) {
_lazyState._retries++;
*retry = true;
}
else{
log() << "too many retries (" << _lazyState._retries
<< "), could not get data from replica set" << endl;
}
}
}
else if( _lazyState._lastOp == dbQuery ){
// if query could not potentially go to a secondary, just mark the master as bad
if( nReturned == -1 /* no result, maybe network problem */ ||
( hasErrField( dataObj ) && ! dataObj["code"].eoo()
&& dataObj["code"].Int() == NotMasterNoSlaveOkCode ) )
{
if( _lazyState._lastClient == _master.get() ){
isntMaster();
}
}
}
}
