void arithmeticExpression_1() { cout << "\narithmetic expression: " << endl; string exp("substr(a)+ 100.00 * sum(tpch.part.p_type) / sum(tpch.lineitem.l_extendedprice *(1-tpch.lineitem.l_discount))"); cout << exp << endl; ArithmeticColumn a(exp, 0); ParseTree* pt = const_cast<ParseTree*>(a.expression()); if (pt != NULL) { pt->walk(walkfnString); pt->drawTree("arithmeticExpression_1.dot"); } cout << " --- end of test 6 ---" << endl; }
void derivedTableOptimization(SCSEP& csep) { // @bug5634. replace the unused column with ConstantColumn from derived table column list, // ExeMgr will not project ConstantColumn. Only count for local derived column. // subquery may carry main query derived table list for column reference, those // derived tables are not checked for optimization in this scope. erydbSelectExecutionPlan::SelectList derivedTbList = csep->derivedTableList(); // @bug6156. Skip horizontal optimization for no table union. bool horizontalOptimization = true; for (uint i = 0; i < derivedTbList.size(); i++) { erydbSelectExecutionPlan *plan = dynamic_cast<erydbSelectExecutionPlan*>(derivedTbList[i].get()); erydbSelectExecutionPlan::ReturnedColumnList cols = plan->returnedCols(); vector<erydbSelectExecutionPlan::ReturnedColumnList> unionColVec; // only do vertical optimization for union all // @bug6134. Also skip the vertical optimization for select distinct // because all columns need to be projected to check the distinctness. bool verticalOptimization = false; if (plan->distinctUnionNum() == 0 && !plan->distinct()) { verticalOptimization = true; for (uint j = 0; j < plan->unionVec().size(); j++) { unionColVec.push_back( dynamic_cast<erydbSelectExecutionPlan*>(plan->unionVec()[j].get())->returnedCols()); } } if (plan->tableList().empty()) horizontalOptimization = false; for (uint j = 0; j < plan->unionVec().size(); j++) { if (dynamic_cast<erydbSelectExecutionPlan*>(plan->unionVec()[j].get())->tableList().empty()) { horizontalOptimization = false; break; } } if (verticalOptimization) { int64_t val = 1; for (uint i = 0; i < cols.size(); i++) { //if (cols[i]->derivedTable().empty()) if (cols[i]->refCount() == 0) { if (cols[i]->derivedRefCol()) cols[i]->derivedRefCol()->decRefCount(); cols[i].reset(new ConstantColumn(val)); for (uint j = 0; j < unionColVec.size(); j++) unionColVec[j][i].reset(new ConstantColumn(val)); } } // set back plan->returnedCols(cols); for (uint j = 0; j < unionColVec.size(); j++) dynamic_cast<erydbSelectExecutionPlan*>(plan->unionVec()[j].get())->returnedCols(unionColVec[j]); } } /* * @bug5635. Move filters that only belongs to a derived table to inside the derived table. * 1. parse tree walk to populate derivedTableFilterMap and set null candidate on the tree. * 2. remove the null filters * 3. and the filters of derivedTableFilterMap and append to the WHERE filter of the derived table * * Note: * 1. Subquery filters is ignored because derived table can not be in subquery * 2. While walking tree, whenever a single derive simplefilter is encountered, * this filter is pushed to the corresponding stack * 2. Whenever an OR operator is encountered, all the filter stack of * that OR involving derived table are emptied and null candidate of each * stacked filter needs to be reset (not null) */ ParseTree* pt = csep->filters(); map<string, ParseTree*> derivedTbFilterMap; if (horizontalOptimization && pt) { pt->walk(setDerivedTable); setDerivedFilter(pt, derivedTbFilterMap, derivedTbList); csep->filters(pt); } // AND the filters of individual stack to the derived table filter tree // @todo union filters. // @todo outer join complication map<string, ParseTree*>::iterator mapIt; for (uint i = 0; i < derivedTbList.size(); i++) { erydbSelectExecutionPlan *plan = dynamic_cast<erydbSelectExecutionPlan*>(derivedTbList[i].get()); erydbSelectExecutionPlan::ReturnedColumnList derivedColList = plan->returnedCols(); mapIt = derivedTbFilterMap.find(plan->derivedTbAlias()); if (mapIt != derivedTbFilterMap.end()) { // replace all derived column of this filter with real column from // derived table projection list. ParseTree *mainFilter = new ParseTree(); mainFilter->copyTree(*(mapIt->second)); replaceRefCol(mainFilter, derivedColList); ParseTree* derivedFilter = plan->filters(); if (derivedFilter) { LogicOperator *op = new LogicOperator("and"); ParseTree *filter = new ParseTree(op); filter->left(derivedFilter); filter->right(mainFilter); plan->filters(filter); } else { plan->filters(mainFilter); } // union filter handling for (uint j = 0; j < plan->unionVec().size(); j++) { erydbSelectExecutionPlan *unionPlan = dynamic_cast<erydbSelectExecutionPlan*>(plan->unionVec()[j].get()); erydbSelectExecutionPlan::ReturnedColumnList unionColList = unionPlan->returnedCols(); ParseTree* mainFilterForUnion = new ParseTree(); mainFilterForUnion->copyTree(*(mapIt->second)); replaceRefCol(mainFilterForUnion, unionColList); ParseTree *unionFilter = unionPlan->filters(); if (unionFilter) { LogicOperator *op = new LogicOperator("and"); ParseTree *filter = new ParseTree(op); filter->left(unionFilter); filter->right(mainFilterForUnion); unionPlan->filters(filter); } else { unionPlan->filters(mainFilterForUnion); } } } } // clean derivedTbFilterMap because all the filters are copied for( mapIt = derivedTbFilterMap.begin(); mapIt != derivedTbFilterMap.end(); ++mapIt ) delete (*mapIt).second; // recursively process the nested derived table for (uint i = 0; i < csep->subSelectList().size(); i++) { SCSEP subselect(boost::dynamic_pointer_cast<erydbSelectExecutionPlan>(csep->subSelectList()[i])); derivedTableOptimization(subselect); } }
void selectExecutionPlan_1() { cout << "SQL: select r_regionkey from region, nation where n_regionkey = r_regionkey and n_regionkey = 2;" << endl; CalpontSelectExecutionPlan csep; CalpontSelectExecutionPlan::ReturnedColumnList colList; ParseTree* filterList; // returned columns SimpleColumn r_regionkey("tpch.region.r_regionkey"); SimpleColumn n_regionkey("tpch.nation.n_regionkey"); SimpleColumn p_partkey("tpch.part.p_partkey"); SimpleColumn n_name("tpch.nation.n_name"); SimpleColumn c_custkey("tpch.customer.c_custkey"); colList.push_back(new SimpleColumn(r_regionkey)); // filters CalpontSelectExecutionPlan::Parser parser; std::vector<Token> tokens; //tokens.push_back(Token(new Operator("("))); //tokens.push_back(Token(new Operator(")"))); tokens.push_back(Token(new SimpleFilter(new Operator("="), new SimpleColumn(r_regionkey), new SimpleColumn(n_regionkey)))); tokens.push_back(Token(new Operator("and"))); tokens.push_back(Token(new SimpleFilter(new Operator("="), new SimpleColumn(r_regionkey), new SimpleColumn(c_custkey)))); tokens.push_back(Token(new Operator("and"))); tokens.push_back(Token(new Operator("("))); tokens.push_back(Token(new SimpleFilter(new Operator("="), new SimpleColumn(n_regionkey), new ConstantColumn("779")))); tokens.push_back(Token(new Operator("or"))); tokens.push_back(Token(new SimpleFilter(new Operator("!="), new SimpleColumn(n_name), new ConstantColumn("'ASIA'")))); tokens.push_back(Token(new Operator(")"))); tokens.push_back(Token(new Operator ("and"))); tokens.push_back(Token(new Operator("("))); tokens.push_back(Token(new SimpleFilter(new Operator("<"), new SimpleColumn(n_regionkey), new ConstantColumn("77")))); tokens.push_back(Token(new Operator("or"))); tokens.push_back(Token(new SimpleFilter(new Operator(">"), new SimpleColumn(p_partkey), new ConstantColumn("7007")))); tokens.push_back(Token(new Operator(")"))); filterList = parser.parse(tokens.begin(), tokens.end()); // draw filterList tree filterList->drawTree("selectExecutionPlan_1.dot"); // calpont execution plan csep.returnedCols (colList); csep.filters (filterList); cout << "\nCalpont Execution Plan:" << endl; cout << csep << endl; cout << " --- end of test 1 ---" << endl; filterList->walk(walkfnString); }