Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

planner: don't decorrelate the APPLY when the inner's projection reference no column (#32370) #32571

Closed
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion expression/util.go
Original file line number Diff line number Diff line change
Expand Up @@ -162,7 +162,7 @@ func extractColumns(result []*Column, expr Expression, filter func(*Column) bool
}

// ExtractColumnSet extracts the different values of `UniqueId` for columns in expressions.
func ExtractColumnSet(exprs []Expression) *intsets.Sparse {
func ExtractColumnSet(exprs ...Expression) *intsets.Sparse {
set := &intsets.Sparse{}
for _, expr := range exprs {
extractColumnSet(expr, set)
Expand Down
16 changes: 16 additions & 0 deletions planner/core/rule_decorrelate.go
Original file line number Diff line number Diff line change
Expand Up @@ -136,6 +136,21 @@ func (s *decorrelateSolver) optimize(ctx context.Context, p LogicalPlan) (Logica
return s.optimize(ctx, p)
}
} else if proj, ok := innerPlan.(*LogicalProjection); ok {
allConst := true
for _, expr := range proj.Exprs {
if len(expression.ExtractCorColumns(expr)) > 0 || !expression.ExtractColumnSet(expr).IsEmpty() {
allConst = false
break
}
}
if allConst && apply.JoinType == LeftOuterJoin {
// If the projection just references some constant. We cannot directly pull it up when the APPLY is an outer join.
// e.g. select (select 1 from t1 where t1.a=t2.a) from t2; When the t1.a=t2.a is false the join's output is NULL.
// But if we pull the projection upon the APPLY. It will return 1 since the projection is evaluated after the join.
// We disable the decorrelation directly for now.
// TODO: Actually, it can be optimized. We need to first push the projection down to the selection. And then the APPLY can be decorrelated.
goto NoOptimize
}
for i, expr := range proj.Exprs {
proj.Exprs[i] = expr.Decorrelate(outerPlan.Schema())
}
Expand Down Expand Up @@ -288,6 +303,7 @@ func (s *decorrelateSolver) optimize(ctx context.Context, p LogicalPlan) (Logica
return s.optimize(ctx, p)
}
}
NoOptimize:
newChildren := make([]LogicalPlan, 0, len(p.Children()))
for _, child := range p.Children() {
np, err := s.optimize(ctx, child)
Expand Down
10 changes: 10 additions & 0 deletions planner/core/rule_partition_processor.go
Original file line number Diff line number Diff line change
Expand Up @@ -395,6 +395,7 @@ func makePartitionByFnCol(sctx sessionctx.Context, columns []*expression.Column,
}
}

<<<<<<< HEAD
if _, ok := monotoneIncFuncs[raw.FuncName.L]; ok {
fn = raw
args := fn.GetArgs()
Expand All @@ -405,6 +406,15 @@ func makePartitionByFnCol(sctx sessionctx.Context, columns []*expression.Column,
col = c
}
}
=======
fn = raw
monotonous = getMonotoneMode(raw.FuncName.L)
// Check the partitionExpr is in the form: fn(col, ...)
// There should be only one column argument, and it should be the first parameter.
if expression.ExtractColumnSet(args...).Len() == 1 {
if col1, ok := args[0].(*expression.Column); ok {
col = col1
>>>>>>> 991132080... planner: don't decorrelate the APPLY when the inner's projection reference no column (#32370)
}
}
case *expression.Column:
Expand Down
3 changes: 2 additions & 1 deletion planner/core/testdata/plan_suite_unexported_in.json
Original file line number Diff line number Diff line change
Expand Up @@ -131,7 +131,8 @@
"select t1.b from t t1 where t1.b in (select t2.b from t t2 where t2.a = t1.a order by t2.a)",
"select t1.b from t t1 where exists(select t2.b from t t2 where t2.a = t1.a order by t2.a)",
// `Sort` will not be eliminated, if it is not the top level operator.
"select t1.b from t t1 where t1.b = (select t2.b from t t2 where t2.a = t1.a order by t2.a limit 1)"
"select t1.b from t t1 where t1.b = (select t2.b from t t2 where t2.a = t1.a order by t2.a limit 1)",
"select (select 1 from t t1 where t1.a = t2.a) from t t2"
]
},
{
Expand Down
3 changes: 2 additions & 1 deletion planner/core/testdata/plan_suite_unexported_out.json
Original file line number Diff line number Diff line change
Expand Up @@ -117,7 +117,8 @@
"Join{DataScan(t1)->DataScan(t2)->Aggr(max(test.t.a),firstrow(test.t.b))}(test.t.b,test.t.b)->Projection->Sel([eq(test.t.b, Column#25)])->Projection",
"Join{DataScan(t1)->DataScan(t2)}(test.t.a,test.t.a)(test.t.b,test.t.b)->Projection",
"Join{DataScan(t1)->DataScan(t2)}(test.t.a,test.t.a)->Projection",
"Apply{DataScan(t1)->DataScan(t2)->Sel([eq(test.t.a, test.t.a)])->Projection->Sort->Limit}->Projection->Sel([eq(test.t.b, test.t.b)])->Projection"
"Apply{DataScan(t1)->DataScan(t2)->Sel([eq(test.t.a, test.t.a)])->Projection->Sort->Limit}->Projection->Sel([eq(test.t.b, test.t.b)])->Projection",
"Apply{DataScan(t2)->DataScan(t1)->Sel([eq(test.t.a, test.t.a)])->Projection}->Projection"
]
},
{
Expand Down
79 changes: 79 additions & 0 deletions util/ranger/detacher.go
Original file line number Diff line number Diff line change
Expand Up @@ -145,6 +145,85 @@ func getEqOrInColOffset(expr expression.Expression, cols []*expression.Column) i
return -1
}

<<<<<<< HEAD
=======
// extractIndexPointRangesForCNF extracts a CNF item from the input CNF expressions, such that the CNF item
// is totally composed of point range filters.
// e.g, for input CNF expressions ((a,b) in ((1,1),(2,2))) and a > 1 and ((a,b,c) in (1,1,1),(2,2,2))
// ((a,b,c) in (1,1,1),(2,2,2)) would be extracted.
func extractIndexPointRangesForCNF(sctx sessionctx.Context, conds []expression.Expression, cols []*expression.Column, lengths []int) (*DetachRangeResult, int, []*valueInfo, error) {
if len(conds) < 2 {
return nil, -1, nil, nil
}
var r *DetachRangeResult
columnValues := make([]*valueInfo, len(cols))
maxNumCols := int(0)
offset := int(-1)
for i, cond := range conds {
tmpConds := []expression.Expression{cond}
colSets := expression.ExtractColumnSet(cond)
if colSets.Len() == 0 {
continue
}
res, err := DetachCondAndBuildRangeForIndex(sctx, tmpConds, cols, lengths)
if err != nil {
return nil, -1, nil, err
}
if len(res.Ranges) == 0 {
return &DetachRangeResult{}, -1, nil, nil
}
// take the union of the two columnValues
columnValues = unionColumnValues(columnValues, res.ColumnValues)
if len(res.AccessConds) == 0 || len(res.RemainedConds) > 0 {
continue
}
sameLens, allPoints := true, true
numCols := int(0)
for j, ran := range res.Ranges {
if !ran.IsPoint(sctx) {
allPoints = false
break
}
if j == 0 {
numCols = len(ran.LowVal)
} else if numCols != len(ran.LowVal) {
sameLens = false
break
}
}
if !allPoints || !sameLens {
continue
}
if numCols > maxNumCols {
r = res
offset = i
maxNumCols = numCols
}
}
if r != nil {
r.IsDNFCond = false
}
return r, offset, columnValues, nil
}

func unionColumnValues(lhs, rhs []*valueInfo) []*valueInfo {
if lhs == nil {
return rhs
}
if rhs != nil {
for i, valInfo := range lhs {
if i >= len(rhs) {
break
}
if valInfo == nil && rhs[i] != nil {
lhs[i] = rhs[i]
}
}
}
return lhs
}

>>>>>>> 991132080... planner: don't decorrelate the APPLY when the inner's projection reference no column (#32370)
// detachCNFCondAndBuildRangeForIndex will detach the index filters from table filters. These conditions are connected with `and`
// It will first find the point query column and then extract the range query column.
// considerDNF is true means it will try to extract access conditions from the DNF expressions.
Expand Down