void WriteBackListener::run(){ OID lastID; lastID.clear(); int secsToSleep = 0; while ( ! inShutdown() && Shard::isMember( _addr ) ){ if ( lastID.isSet() ){ scoped_lock lk( _seenWritebacksLock ); _seenWritebacks.insert( lastID ); lastID.clear(); } try { ScopedDbConnection conn( _addr ); BSONObj result; { BSONObjBuilder cmd; cmd.appendOID( "writebacklisten" , &serverID ); // Command will block for data if ( ! conn->runCommand( "admin" , cmd.obj() , result ) ){ log() << "writebacklisten command failed! " << result << endl; conn.done(); continue; } } log(1) << "writebacklisten result: " << result << endl; BSONObj data = result.getObjectField( "data" ); if ( data.getBoolField( "writeBack" ) ){ string ns = data["ns"].valuestrsafe(); { BSONElement e = data["id"]; if ( e.type() == jstOID ) lastID = e.OID(); } int len; Message m( (void*)data["msg"].binData( len ) , false ); massert( 10427 , "invalid writeback message" , m.header()->valid() ); DBConfigPtr db = grid.getDBConfig( ns ); ShardChunkVersion needVersion( data["version"] ); log(1) << "writeback id: " << lastID << " needVersion : " << needVersion.toString() << " mine : " << db->getChunkManager( ns )->getVersion().toString() << endl;// TODO change to log(3) if ( logLevel ) log(1) << debugString( m ) << endl; if ( needVersion.isSet() && needVersion <= db->getChunkManager( ns )->getVersion() ){ // this means when the write went originally, the version was old // if we're here, it means we've already updated the config, so don't need to do again //db->getChunkManager( ns , true ); // SERVER-1349 } else { // we received a writeback object that was sent to a previous version of a shard // the actual shard may not have the object the writeback operation is for // we need to reload the chunk manager and get the new shard versions db->getChunkManager( ns , true ); } Request r( m , 0 ); r.init(); r.process(); } else if ( result["noop"].trueValue() ){ // no-op } else { log() << "unknown writeBack result: " << result << endl; } conn.done(); secsToSleep = 0; continue; } catch ( std::exception e ){ if ( inShutdown() ){ // we're shutting down, so just clean up return; } log() << "WriteBackListener exception : " << e.what() << endl; // It's possible this shard was removed Shard::reloadShardInfo(); } catch ( ... ){ log() << "WriteBackListener uncaught exception!" << endl; } secsToSleep++; sleepsecs(secsToSleep); if ( secsToSleep > 10 ) secsToSleep = 0; } log() << "WriteBackListener exiting : address no longer in cluster " << _addr; }