Skip to content

Improve performance of Merge for MarkDistinct #26759

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.

Sign up for GitHub

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

Draft
wants to merge 1 commit into
base: master
Choose a base branch
from
+26 −11
Draft

Improve performance of Merge for MarkDistinct #26759

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
37 changes: 26 additions & 11 deletions core/trino-main/src/main/java/io/trino/sql/planner/QueryPlanner.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,6 @@
import io.trino.sql.ir.Expression;
import io.trino.sql.ir.FieldReference;
import io.trino.sql.ir.IsNull;
import io.trino.sql.ir.Logical;
import io.trino.sql.ir.Row;
import io.trino.sql.ir.WhenClause;
import io.trino.sql.planner.RelationPlanner.PatternRecognitionComponents;
Expand Down Expand Up @@ -751,9 +750,9 @@ public MergeWriterNode plan(Merge merge)
.process(merge.getTarget());

// Assign a unique id to every target table row
Symbol uniqueIdSymbol = symbolAllocator.newSymbol("unique_id", BIGINT);
Symbol targetUniqueIdSymbol = symbolAllocator.newSymbol("target_unique_id", BIGINT);
RelationPlan planWithUniqueId = new RelationPlan(
new AssignUniqueId(idAllocator.getNextId(), targetTablePlan.getRoot(), uniqueIdSymbol),
new AssignUniqueId(idAllocator.getNextId(), targetTablePlan.getRoot(), targetUniqueIdSymbol),
mergeAnalysis.getTargetTableScope(),
targetTablePlan.getFieldMappings(),
outerContext);
Expand All @@ -774,8 +773,16 @@ public MergeWriterNode plan(Merge merge)
RelationPlan source = new RelationPlanner(analysis, symbolAllocator, idAllocator, lambdaDeclarationToSymbolMap, plannerContext, outerContext, session, recursiveSubqueries)
.process(merge.getSource());

// Assign a unique id to every source table row
Symbol sourceUniqueIdSymbol = symbolAllocator.newSymbol("source_unique_id", BIGINT);
RelationPlan sourcePlanWithUniqueId = new RelationPlan(
new AssignUniqueId(idAllocator.getNextId(), source.getRoot(), sourceUniqueIdSymbol),
source.getScope(),
source.getFieldMappings(),
outerContext);

RelationPlan joinPlan = new RelationPlanner(analysis, symbolAllocator, idAllocator, lambdaDeclarationToSymbolMap, plannerContext, outerContext, session, recursiveSubqueries)
.planJoin(merge.getPredicate(), Join.Type.RIGHT, mergeAnalysis.getJoinScope(), planWithPresentColumn, source, analysis.getSubqueries(merge)); // TODO: ir
.planJoin(merge.getPredicate(), Join.Type.RIGHT, mergeAnalysis.getJoinScope(), planWithPresentColumn, sourcePlanWithUniqueId, analysis.getSubqueries(merge)); // TODO: ir

PlanBuilder subPlan = newPlanBuilder(joinPlan, analysis, lambdaDeclarationToSymbolMap, session, plannerContext);

Expand Down Expand Up @@ -864,10 +871,11 @@ public MergeWriterNode plan(Merge merge)

List<io.trino.sql.tree.Expression> constraints = analysis.getCheckConstraints(mergeAnalysis.getTargetTable());
if (!constraints.isEmpty()) {
assignments.putIdentity(uniqueIdSymbol);
assignments.putIdentity(targetUniqueIdSymbol);
assignments.putIdentity(sourceUniqueIdSymbol);
assignments.putIdentity(presentColumn);
assignments.putIdentity(rowIdSymbol);
assignments.putIdentities(source.getFieldMappings());
assignments.putIdentities(sourcePlanWithUniqueId.getFieldMappings());
subPlan = subPlan.withNewRoot(new ProjectNode(
idAllocator.getNextId(),
subPlan.getRoot(),
Expand Down Expand Up @@ -897,7 +905,15 @@ public MergeWriterNode plan(Merge merge)
Symbol symbol = planWithPresentColumn.getFieldMappings().get(fieldIndex);
projectionAssignmentsBuilder.putIdentity(symbol);
}
projectionAssignmentsBuilder.putIdentity(uniqueIdSymbol);

// Assigns a unique id to each joined row
// The target table unique_id for matches, and the source table unique_id for non-matches
// Avoid the scenario where unique_id values become null after right join due to unmatched rows.
// It can improve performance and parallelism when handling non-matches in a mark distinct operation.
Symbol uniqueIdSymbol = symbolAllocator.newSymbol("unique_id", BIGINT);
Expression uniqueIdExpression = new Coalesce(targetUniqueIdSymbol.toSymbolReference(), sourceUniqueIdSymbol.toSymbolReference());

projectionAssignmentsBuilder.put(uniqueIdSymbol, uniqueIdExpression);
projectionAssignmentsBuilder.putIdentity(rowIdSymbol);
projectionAssignmentsBuilder.put(mergeRowSymbol, caseExpression);

Expand All @@ -919,11 +935,10 @@ public MergeWriterNode plan(Merge merge)
Symbol isDistinctSymbol = symbolAllocator.newSymbol("is_distinct", BOOLEAN);
MarkDistinctNode markDistinctNode = new MarkDistinctNode(idAllocator.getNextId(), project, isDistinctSymbol, ImmutableList.of(uniqueIdSymbol, caseNumberSymbol));

// Raise an error if unique_id symbol is non-null and the unique_id/case_number combination was not distinct
// The unique_id which originates from either the source or target table will not be null
// Raise an error if the unique_id/case_number combination was not distinct
Expression filter = ifExpression(
Logical.and(
not(metadata, isDistinctSymbol.toSymbolReference()),
not(metadata, new IsNull(uniqueIdSymbol.toSymbolReference()))),
not(metadata, isDistinctSymbol.toSymbolReference()),
new Cast(
failFunction(metadata, MERGE_TARGET_ROW_MULTIPLE_MATCHES, "One MERGE target table row matched more than one source row"),
BOOLEAN),
Expand Down
Loading