feat(rbo,rules): rule matcher has been replaced with csep filter for a rule to relax rule filter predicates and endless loop has been fixed in pushdown predicates.

dbcon/mysql/CMakeLists.txt

@@ -44,6 +44,7 @@ set(libcalmysql_SRCS
     columnstore_dataload.cpp
     rulebased_optimizer.cpp
     rbo_apply_parallel_ces.cpp
+    rbo_predicate_pushdown.cpp
 )
 
 set_source_files_properties(ha_mcs.cpp PROPERTIES COMPILE_FLAGS "-fno-implicit-templates")

dbcon/mysql/rbo_apply_parallel_ces.cpp

@@ -54,7 +54,7 @@ bool tableIsInUnion(const execplan::CalpontSystemCatalog::TableAliasName& table,
                      });
 }
 
-bool matchParallelCES(execplan::CalpontSelectExecutionPlan& csep)
+bool parallelCESFilter(execplan::CalpontSelectExecutionPlan& csep)
 {
   auto tables = csep.tableList();
   // This is leaf and there are no other tables at this level in neither UNION, nor derived table.
@@ -198,12 +198,13 @@ execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable(
 
   return unionVec;
 }
-void applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerContext& ctx)
+bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerContext& ctx)
 {
   auto tables = csep.tableList();
   execplan::CalpontSelectExecutionPlan::TableList newTableList;
   execplan::CalpontSelectExecutionPlan::SelectList newDerivedTableList;
   execplan::CalpontSelectExecutionPlan::ReturnedColumnList newReturnedColumns;
+  bool ruleHasBeenApplied = false;
 
   // ATM Must be only 1 table
   for (auto& table : tables)
@@ -246,6 +247,7 @@ void applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerCon
       // Remove the filters as they were pushed down to union units
       // This is inappropriate for EXISTS filter and join conditions
       derivedSCEP->filters(nullptr);
+      ruleHasBeenApplied = true;
     }
   }
   // Remove the filters if necessary using csep.filters(nullptr) as they were pushed down to union units
@@ -255,6 +257,7 @@ void applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerCon
   // Replace table list with new table list populated with union units
   csep.tableList(newTableList);
   csep.returnedCols(newReturnedColumns);
+  return ruleHasBeenApplied;
 }
 
 }  // namespace optimizer

dbcon/mysql/rbo_apply_parallel_ces.h

@@ -25,6 +25,6 @@
 #include "rulebased_optimizer.h"
 
 namespace optimizer {
-  bool matchParallelCES(execplan::CalpontSelectExecutionPlan& csep);
-  void applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx);
+  bool parallelCESFilter(execplan::CalpontSelectExecutionPlan& csep);
+  bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx);
 }

dbcon/mysql/rbo_predicate_pushdown.cpp

@@ -29,13 +29,16 @@
 namespace optimizer
 {
 
-bool matchPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep)
+   using DerivedToFiltersMap = std::map<std::string, execplan::ParseTree*>;
+
+bool predicatePushdownFilter(execplan::CalpontSelectExecutionPlan& csep)
 {
   // The original rule match contains questionable decision to filter out
   // queries that contains any UNION UNIT with only derived tables.
   // See ha_from_sub.cpp before MCS 23.10.7 for more details and @bug6156.
   // All tables are derived thus nothing to optimize.
-  return !csep.tableList().empty();
+  return !
+  csep.tableList().empty();
 }
 
 
@@ -73,7 +76,7 @@ void setDerivedTable(execplan::ParseTree* n)
   }
 }
 
-execplan::ParseTree* setDerivedFilter(cal_impl_if::gp_walk_info* gwip, execplan::ParseTree*& n, map<string, execplan::ParseTree*>& filterMap,
+execplan::ParseTree* setDerivedFilter(cal_impl_if::gp_walk_info* gwip, execplan::ParseTree*& n, DerivedToFiltersMap& filterMap,
                             const execplan::CalpontSelectExecutionPlan::SelectList& derivedTbList)
 {
   if (!(n->derivedTable().empty()))
@@ -99,11 +102,11 @@ execplan::ParseTree* setDerivedFilter(cal_impl_if::gp_walk_info* gwip, execplan:
 
     // 2. push the filter to the derived table filter stack, or 'and' with
     // the filters in the stack
-    map<string, execplan::ParseTree*>::iterator mapIter = filterMap.find(n->derivedTable());
+    auto mapIter = filterMap.find(n->derivedTable());
 
     if (mapIter == filterMap.end())
     {
-      filterMap.insert(pair<string, execplan::ParseTree*>(n->derivedTable(), n));
+      filterMap.insert({n->derivedTable(), n});
     }
     else
     {
@@ -141,7 +144,7 @@ execplan::ParseTree* setDerivedFilter(cal_impl_if::gp_walk_info* gwip, execplan:
   return n;
 }
 
-void applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerContext& ctx)
+bool applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerContext& ctx)
 {
   /*
    * @bug5635. Move filters that only belongs to a derived table to inside the derived table.
@@ -158,8 +161,9 @@ void applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, RBOptimi
    * stacked filter needs to be reset (not null)
    */
    execplan::ParseTree* pt = csep.filters();
-   map<string, execplan::ParseTree*> derivedTbFilterMap;
+   DerivedToFiltersMap derivedTbFilterMap;
    auto& derivedTbList = csep.derivedTableList();
+   bool hasBeenApplied = false;
 
    if (pt)
    {
@@ -171,13 +175,11 @@ void applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, RBOptimi
    // AND the filters of individual stack to the derived table filter tree
    // @todo union filters.
    // @todo outer join complication
-   map<string, execplan::ParseTree*>::iterator mapIt;
- 
    for (uint i = 0; i < derivedTbList.size(); i++)
    {
      execplan::CalpontSelectExecutionPlan* plan = dynamic_cast<execplan::CalpontSelectExecutionPlan*>(derivedTbList[i].get());
      execplan::CalpontSelectExecutionPlan::ReturnedColumnList derivedColList = plan->returnedCols();
-     mapIt = derivedTbFilterMap.find(plan->derivedTbAlias());
+     auto mapIt = derivedTbFilterMap.find(plan->derivedTbAlias());
  
      if (mapIt != derivedTbFilterMap.end())
      {
@@ -225,12 +227,17 @@ void applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, RBOptimi
            unionPlan->filters(mainFilterForUnion);
          }
        }
+       hasBeenApplied = true;
      }
    }
  
    // clean derivedTbFilterMap because all the filters are copied
-   for (mapIt = derivedTbFilterMap.begin(); mapIt != derivedTbFilterMap.end(); ++mapIt)
-     delete (*mapIt).second;
+   for (auto mapIt = derivedTbFilterMap.begin(); mapIt != derivedTbFilterMap.end(); ++mapIt)
+   {
+    delete (*mapIt).second;
+   }
+
+  return hasBeenApplied;
 }
 
 }  // namespace optimizer

dbcon/mysql/rbo_predicate_pushdown.h

@@ -25,6 +25,6 @@
 #include "rulebased_optimizer.h"
 
 namespace optimizer {
-  bool matchPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep);
-  void applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx);
+  bool predicatePushdownFilter(execplan::CalpontSelectExecutionPlan& csep);
+  bool applyPredicatePushdown(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx);
 }

dbcon/mysql/rulebased_optimizer.cpp

@@ -31,6 +31,7 @@
 #include "predicateoperator.h"
 #include "simplefilter.h"
 #include "rbo_apply_parallel_ces.h"
+#include "rbo_predicate_pushdown.h"
 
 namespace optimizer
 {
@@ -54,12 +55,12 @@ bool optimizeCSEP(execplan::CalpontSelectExecutionPlan& root, optimizer::RBOptim
 
   if (get_unstable_optimizer(&ctx.thd))
   {
-    optimizer::Rule parallelCES{"parallelCES", optimizer::matchParallelCES, optimizer::applyParallelCES};
+    optimizer::Rule parallelCES{"parallelCES", optimizer::parallelCESFilter, optimizer::applyParallelCES};
     rules.push_back(parallelCES);
   }
 
-  optimizer::Rule predicatePushdown{"predicatePushdown", optimizer::matchParallelCES,
-                                    optimizer::applyParallelCES};
+  optimizer::Rule predicatePushdown{"predicatePushdown", optimizer::predicatePushdownFilter,
+                                    optimizer::applyPredicatePushdown};
   rules.push_back(predicatePushdown);
 
   return optimizeCSEPWithRules(root, rules, ctx);
@@ -114,7 +115,7 @@ bool Rule::walk(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimiz
     }
 
     // Walk nested subselect in filters, e.g. SEMI-JOIN
-    for (auto& subselect : csep.subSelectList())
+    for (auto& subselect : current->subSelectList())
     {
       auto* subselectPtr = dynamic_cast<execplan::CalpontSelectExecutionPlan*>(subselect.get());
       if (subselectPtr)
@@ -125,11 +126,10 @@ bool Rule::walk(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimiz
 
     // TODO add walking nested subselect in projection. See CSEP::fSelectSubList
 
-    if (matchRule(*current))
+    if (mayApply(*current))
     {
-      applyRule(*current, ctx);
+      rewrite = applyRule(*current, ctx);
       ++ctx.uniqueId;
-      rewrite = true;
     }
   }
 

dbcon/mysql/rulebased_optimizer.h

@@ -42,14 +42,16 @@ public:
 
 struct Rule
 {
-  using RuleMatcher = bool (*)(execplan::CalpontSelectExecutionPlan&);
-  using RuleApplier = void (*)(execplan::CalpontSelectExecutionPlan&, RBOptimizerContext&);
+  // returns true if rule may be applied
+  using RuleApplierFilter = bool (*)(execplan::CalpontSelectExecutionPlan&);
+  // returns true if rule was applied
+  using RuleApplier = bool (*)(execplan::CalpontSelectExecutionPlan&, RBOptimizerContext&);
 
-  Rule(std::string&& name, RuleMatcher matchRule, RuleApplier applyRule)
-   : name(name), matchRule(matchRule), applyRule(applyRule) {};
+  Rule(std::string&& name, RuleApplierFilter mayApply, RuleApplier applyRule)
+   : name(name), mayApply(mayApply), applyRule(applyRule) {};
 
   std::string name;
-  RuleMatcher matchRule;
+  RuleApplierFilter mayApply;
   RuleApplier applyRule;
   // TODO Wrap CSEP into Nodes to be able to navigate up and down the tree and remove this flag
   bool applyOnlyOnce = true;