Class GroupByNode

All Implemented Interfaces:
Optimizable, Visitable

class GroupByNode extends SingleChildResultSetNode
A GroupByNode represents a result set for a grouping operation on a select. Note that this includes a SELECT with aggregates and no grouping columns (in which case the select list is null) It has the same description as its input result set.

For the most part, it simply delegates operations to its bottomPRSet, which is currently expected to be a ProjectRestrictResultSet generated for a SelectNode.

NOTE: A GroupByNode extends FromTable since it can exist in a FromList.

There is a lot of room for optimizations here:

  • agg(distinct x) group by x => agg(x) group by x (for min and max)
  • min()/max() use index scans if possible, no sort may be needed.
  • Field Details

    • groupingList

      GroupByList groupingList
      The GROUP BY list
    • aggregates

      private List<AggregateNode> aggregates
      The list of all aggregates in the query block that contains this group by.
    • aggInfo

      private AggregatorInfoList aggInfo
      Information that is used at execution time to process aggregates.
    • parent

      FromTable parent
      The parent to the GroupByNode. If we need to generate a ProjectRestrict over the group by then this is set to that node. Otherwise it is null.
    • addDistinctAggregate

      private boolean addDistinctAggregate
    • singleInputRowOptimization

      private boolean singleInputRowOptimization
    • addDistinctAggregateColumnNum

      private int addDistinctAggregateColumnNum
    • isInSortedOrder

      private final boolean isInSortedOrder
    • havingClause

      private ValueNode havingClause
    • havingSubquerys

      private SubqueryList havingSubquerys
  • Constructor Details

    • GroupByNode

      GroupByNode(ResultSetNode bottomPR, GroupByList groupingList, List<AggregateNode> aggregates, ValueNode havingClause, SubqueryList havingSubquerys, int nestingLevel, ContextManager cm) throws StandardException
      Constructor for a GroupByNode.
      Parameters:
      bottomPR - The child FromTable
      groupingList - The groupingList
      aggregates - The list of aggregates from the query block. Since aggregation is done at the same time as grouping, we need them here.
      havingClause - The having clause.
      havingSubquerys - subqueries in the having clause.
      nestingLevel - NestingLevel of this group by node. This is used for error checking of group by queries with having clause.
      cm - The context manager
      Throws:
      StandardException - Thrown on error
  • Method Details

    • getIsInSortedOrder

      boolean getIsInSortedOrder()
      Get whether or not the source is in sorted order.
      Returns:
      Whether or not the source is in sorted order.
    • addAggregates

      private void addAggregates() throws StandardException
      Add the extra result columns required by the aggregates to the result list.
    • addDistinctAggregatesToOrderBy

      private void addDistinctAggregatesToOrderBy()
      Add any distinct aggregates to the order by list. Asserts that there are 0 or more distincts.
    • addNewPRNode

      private void addNewPRNode() throws StandardException
      Add a new PR node for aggregation. Put the new PR under the sort.
    • addUnAggColumns

      private ArrayList<SubstituteExpressionVisitor> addUnAggColumns() throws StandardException
      In the query rewrite for group by, add the columns on which we are doing the group by.
      Returns:
      havingRefsToSubstitute visitors array. Return any havingRefsToSubstitute visitors since it is too early to apply them yet; we need the AggregateNodes unmodified until after we add the new columns for aggregation (DERBY-4071).
      Throws:
      StandardException
      See Also:
    • addNewColumnsForAggregation

      private void addNewColumnsForAggregation() throws StandardException
      Add a whole slew of columns needed for aggregation. Basically, for each aggregate we add 3 columns: the aggregate input expression and the aggregator column and a column where the aggregate result is stored. The input expression is taken directly from the aggregator node. The aggregator is the run time aggregator. We add it to the RC list as a new object coming into the sort node.

      At this point this is invoked, we have the following tree:

        PR - (PARENT): RCL is the original select list | PR - GROUP BY: RCL is empty | PR - FROM TABLE: RCL is empty

      For each ColumnReference in PR RCL

      • clone the ref
      • create a new RC in the bottom RCL and set it to the col ref
      • create a new RC in the GROUPBY RCL and set it to point to the bottom RC
      • reset the top PR ref to point to the new GROUPBY RC
      For each aggregate in aggregates
      • create RC in FROM TABLE. Fill it with aggs Operator.
      • create RC in FROM TABLE for agg result
      • create RC in FROM TABLE for aggregator
      • create RC in GROUPBY for agg input, set it to point to FROM TABLE RC
      • create RC in GROUPBY for agg result
      • create RC in GROUPBY for aggregator
      • replace Agg with reference to RC for agg result
      .

      For a query like,

                select c1, sum(c2), max(c3)
                from t1 
                group by c1;
                
      the query tree ends up looking like this:
                  ProjectRestrictNode RCL -> (ptr to GBN(column[0]), ptr to GBN(column[1]), ptr to GBN(column[4]))
                            |
                  GroupByNode RCL->(C1, SUM(C2), <agg-input>, , MAX(C3), <agg-input>, <aggregator>)
                            |
                  ProjectRestrict RCL->(C1, C2, C3)
                            |
                  FromBaseTable
                  
      The RCL of the GroupByNode contains all the unagg (or grouping columns) followed by 3 RC's for each aggregate in this order: the final computed aggregate value, the aggregate input and the aggregator function.

      The Aggregator function puts the results in the first of the 3 RC's and the PR resultset in turn picks up the value from there.

      The notation (ptr to GBN(column[0])) basically means that it is a pointer to the 0th RC in the RCL of the GroupByNode.

      The addition of these unagg and agg columns to the GroupByNode and to the PRN is performed in addUnAggColumns and addAggregateColumns.

      Note that that addition of the GroupByNode is done after the query is optimized (in SelectNode#modifyAccessPaths) which means a fair amount of patching up is needed to account for generated group by columns.

    • addAggregateColumns

      private void addAggregateColumns() throws StandardException
      In the query rewrite involving aggregates, add the columns for aggregation.
      Throws:
      StandardException
      See Also:
    • getParent

      final FromTable getParent()
      Return the parent node to this one, if there is one. It will return 'this' if there is no generated node above this one.
      Returns:
      the parent node
    • optimizeIt

      public CostEstimate optimizeIt(Optimizer optimizer, OptimizablePredicateList predList, CostEstimate outerCost, RowOrdering rowOrdering) throws StandardException
      Description copied from interface: Optimizable
      Choose the best access path for this Optimizable.
      Specified by:
      optimizeIt in interface Optimizable
      Overrides:
      optimizeIt in class FromTable
      Parameters:
      optimizer - Optimizer to use.
      predList - The predicate list to optimize against
      outerCost - The CostEstimate for the outer tables in the join order, telling how many times this Optimizable will be scanned.
      rowOrdering - The row ordering for all the tables in the join order, including this one.
      Returns:
      The optimizer's estimated cost of the best access path.
      Throws:
      StandardException - Thrown on error
      See Also:
    • estimateCost

      public CostEstimate estimateCost(OptimizablePredicateList predList, ConglomerateDescriptor cd, CostEstimate outerCost, Optimizer optimizer, RowOrdering rowOrdering) throws StandardException
      Description copied from interface: Optimizable
      Estimate the cost of scanning this Optimizable using the given predicate list with the given conglomerate. It is assumed that the predicate list has already been classified. This cost estimate is just for one scan, not for the life of the query.
      Specified by:
      estimateCost in interface Optimizable
      Overrides:
      estimateCost in class FromTable
      Parameters:
      predList - The predicate list to optimize against
      cd - The conglomerate descriptor to get the cost of
      outerCost - The estimated cost of the part of the plan outer to this optimizable.
      optimizer - The optimizer to use to help estimate the cost
      rowOrdering - The row ordering for all the tables in the join order, including this one.
      Returns:
      The estimated cost of doing the scan
      Throws:
      StandardException - Thrown on error
      See Also:
    • pushOptPredicate

      public boolean pushOptPredicate(OptimizablePredicate optimizablePredicate) throws StandardException
      Description copied from interface: Optimizable
      Push an OptimizablePredicate down, if this node accepts it.
      Specified by:
      pushOptPredicate in interface Optimizable
      Overrides:
      pushOptPredicate in class FromTable
      Parameters:
      optimizablePredicate - OptimizablePredicate to push down.
      Returns:
      Whether or not the predicate was pushed down.
      Throws:
      StandardException - Thrown on error
      See Also:
    • toString

      public String toString()
      Convert this object to a String. See comments in QueryTreeNode.java for how this should be done for tree printing.
      Overrides:
      toString in class FromTable
      Returns:
      This object as a String
    • printSubNodes

      void printSubNodes(int depth)
      Prints the sub-nodes of this object. See QueryTreeNode.java for how tree printing is supposed to work.
      Overrides:
      printSubNodes in class SingleChildResultSetNode
      Parameters:
      depth - The depth of this node in the tree
    • flattenableInFromSubquery

      boolean flattenableInFromSubquery(FromList fromList)
      Evaluate whether or not the subquery in a FromSubquery is flattenable. Currently, a FSqry is flattenable if all of the following are true: o Subquery is a SelectNode. o It contains no top level subqueries. (RESOLVE - we can relax this) o It does not contain a group by or having clause o It does not contain aggregates.
      Overrides:
      flattenableInFromSubquery in class SingleChildResultSetNode
      Parameters:
      fromList - The outer from list
      Returns:
      boolean Whether or not the FromSubquery is flattenable.
    • optimize

      ResultSetNode optimize(DataDictionary dataDictionary, PredicateList predicates, double outerRows) throws StandardException
      Optimize this GroupByNode.
      Overrides:
      optimize in class SingleChildResultSetNode
      Parameters:
      dataDictionary - The DataDictionary to use for optimization
      predicates - The PredicateList to optimize. This should be a join predicate.
      outerRows - The number of outer joining rows
      Returns:
      ResultSetNode The top of the optimized subtree
      Throws:
      StandardException - Thrown on error
    • makeResultDescriptors

      ResultColumnDescriptor[] makeResultDescriptors()
      Overrides:
      makeResultDescriptors in class ResultSetNode
    • isOneRowResultSet

      boolean isOneRowResultSet() throws StandardException
      Return whether or not the underlying ResultSet tree will return a single row, at most. This is important for join nodes where we can save the extra next on the right side if we know that it will return at most 1 row.
      Overrides:
      isOneRowResultSet in class SingleChildResultSetNode
      Returns:
      Whether or not the underlying ResultSet tree will return a single row.
      Throws:
      StandardException - Thrown on error
    • generate

      void generate(ActivationClassBuilder acb, MethodBuilder mb) throws StandardException
      generate the sort result set operating over the source result set. Adds distinct aggregates to the sort if necessary.
      Overrides:
      generate in class QueryTreeNode
      Parameters:
      acb - The ActivationClassBuilder for the class being built
      mb - The method for the generated code to go into
      Throws:
      StandardException - Thrown on error
    • genScalarAggregateResultSet

      private void genScalarAggregateResultSet(ActivationClassBuilder acb, MethodBuilder mb)
      Generate the code to evaluate scalar aggregates.
    • genGroupedAggregateResultSet

      private void genGroupedAggregateResultSet(ActivationClassBuilder acb, MethodBuilder mb) throws StandardException
      Generate the code to evaluate grouped aggregates.
      Throws:
      StandardException
    • getColumnReference

      private ResultColumn getColumnReference(ResultColumn targetRC, DataDictionary dd) throws StandardException
      Method for creating a new result column referencing the one passed in.
      Parameters:
      targetRC - the source
      dd -
      Returns:
      the new result column
      Throws:
      StandardException - on error
    • considerPostOptimizeOptimizations

      void considerPostOptimizeOptimizations(boolean selectHasPredicates) throws StandardException
      Consider any optimizations after the optimizer has chosen a plan. Optimizations include: o min optimization for scalar aggregates o max optimization for scalar aggregates
      Parameters:
      selectHasPredicates - true if SELECT containing this vector/scalar aggregate has a restriction
      Throws:
      StandardException - on error