// isDecomposable checks if an aggregate function is decomposable. An aggregation function $F$ is decomposable
// if there exist aggregation functions F_1 and F_2 such that F(S_1 union all S_2) = F_2(F_1(S_1),F_1(S_2)),
// where S_1 and S_2 are two sets of values. We call S_1 and S_2 partial groups.
// It's easy to see that max, min, first row is decomposable, no matter whether it's distinct, but sum(distinct) and
// count(distinct) is not.
// Currently we don't support avg and concat.
func (a *aggPushDownSolver) isDecomposable(fun expression.AggregationFunction) bool {
switch fun.GetName() {
case ast.AggFuncAvg, ast.AggFuncGroupConcat:
// TODO: Support avg push down.
return false
case ast.AggFuncMax, ast.AggFuncMin, ast.AggFuncFirstRow:
return true
case ast.AggFuncSum, ast.AggFuncCount:
return !fun.IsDistinct()
default:
return false
}
}
// getAggFuncChildIdx gets which children it belongs to, 0 stands for left, 1 stands for right, -1 stands for both.
func (a *aggPushDownSolver) getAggFuncChildIdx(aggFunc expression.AggregationFunction, schema expression.Schema) int {
fromLeft, fromRight := false, false
var cols []*expression.Column
for _, arg := range aggFunc.GetArgs() {
cols = append(cols, expression.ExtractColumns(arg)...)
}
for _, col := range cols {
if schema.GetIndex(col) != -1 {
fromLeft = true
} else {
fromRight = true
}
}
if fromLeft && fromRight {
return -1
} else if fromLeft {
return 0
}
return 1
}
func (b *executorBuilder) newAggFuncToPBExpr(client kv.Client, aggFunc expression.AggregationFunction,
tbl *model.TableInfo) *tipb.Expr {
var tp tipb.ExprType
switch aggFunc.GetName() {
case ast.AggFuncCount:
tp = tipb.ExprType_Count
case ast.AggFuncFirstRow:
tp = tipb.ExprType_First
case ast.AggFuncGroupConcat:
tp = tipb.ExprType_GroupConcat
case ast.AggFuncMax:
tp = tipb.ExprType_Max
case ast.AggFuncMin:
tp = tipb.ExprType_Min
case ast.AggFuncSum:
tp = tipb.ExprType_Sum
case ast.AggFuncAvg:
tp = tipb.ExprType_Avg
}
if !client.SupportRequestType(kv.ReqTypeSelect, int64(tp)) {
return nil
}
children := make([]*tipb.Expr, 0, len(aggFunc.GetArgs()))
for _, arg := range aggFunc.GetArgs() {
pbArg := b.newExprToPBExpr(client, arg, tbl)
if pbArg == nil {
return nil
}
children = append(children, pbArg)
}
return &tipb.Expr{Tp: tp.Enum(), Children: children}
}
func aggFuncToPBExpr(sc *variable.StatementContext, client kv.Client, aggFunc expression.AggregationFunction) *tipb.Expr {
pc := pbConverter{client: client, sc: sc}
var tp tipb.ExprType
switch aggFunc.GetName() {
case ast.AggFuncCount:
tp = tipb.ExprType_Count
case ast.AggFuncFirstRow:
tp = tipb.ExprType_First
case ast.AggFuncGroupConcat:
tp = tipb.ExprType_GroupConcat
case ast.AggFuncMax:
tp = tipb.ExprType_Max
case ast.AggFuncMin:
tp = tipb.ExprType_Min
case ast.AggFuncSum:
tp = tipb.ExprType_Sum
case ast.AggFuncAvg:
tp = tipb.ExprType_Avg
}
if !client.SupportRequestType(kv.ReqTypeSelect, int64(tp)) {
return nil
}
children := make([]*tipb.Expr, 0, len(aggFunc.GetArgs()))
for _, arg := range aggFunc.GetArgs() {
pbArg := pc.exprToPB(arg)
if pbArg == nil {
return nil
}
children = append(children, pbArg)
}
return &tipb.Expr{Tp: tp, Children: children}
}
// decompose splits an aggregate function to two parts: a final mode function and a partial mode function. Currently
// there are no differences between partial mode and complete mode, so we can confuse them.
func (a *aggPushDownSolver) decompose(aggFunc expression.AggregationFunction, schema expression.Schema, id string) ([]expression.AggregationFunction, expression.Schema) {
// Result is a slice because avg should be decomposed to sum and count. Currently we don't process this case.
result := []expression.AggregationFunction{aggFunc.Clone()}
for _, aggFunc := range result {
schema = append(schema, &expression.Column{
ColName: model.NewCIStr(fmt.Sprintf("join_agg_%d", len(schema))), // useless but for debug
FromID: id,
Position: len(schema),
RetType: aggFunc.GetType(),
})
}
aggFunc.SetArgs(expression.Schema2Exprs(schema[len(schema)-len(result):]))
aggFunc.SetMode(expression.FinalMode)
return result, schema
}
func needValue(af expression.AggregationFunction) bool {
return af.GetName() == ast.AggFuncSum || af.GetName() == ast.AggFuncAvg || af.GetName() == ast.AggFuncFirstRow ||
af.GetName() == ast.AggFuncMax || af.GetName() == ast.AggFuncMin || af.GetName() == ast.AggFuncGroupConcat
}
func needCount(af expression.AggregationFunction) bool {
return af.GetName() == ast.AggFuncCount || af.GetName() == ast.AggFuncAvg
}